Final Report Summary - DIAMOND (Diagnosis-aided control for SOFC power systems)
Executive Summary:
The project DIAMOND tackles the problem of monitoring, diagnosing and controlling SOFC systems with a holistic approach to achieve advanced management. The consortium has developed monitoring, diagnostic and control techniques to improve performance and durability. These developments have been tested on two different SOFC systems: a conventional one, DIAMOND-C and an advanced system based on an integrated HoTbox, DIAMOND-A.
A thorough inventory of faults and failures which might occur in SOFC CHP systems has been made. The most important ones like, for example, high fuel and air utilisation, leakages, too high temperatures and out of specification steam to carbon ratios, have been of particular focus in the project.
By using fault-tree analyses for the systems, fault signature matrices have been constructed, which are used for Fault Detection and Isolation (FDI) methodology. To develop and implement model-based diagnosis and control strategies, a set of models was proposed and tested. The DIAMOND-C system model was improved using the data delivered by VTT obtained using the new stacks. All system models were properly coded for their implementation on PLC and tested. The DIAMOND-A system was modelled making use of a dynamic lumped gray-box approach and validated against experimental data.
A signal-based diagnosis was developed focusing on degradation phenomena. In order to detect degrading signatures, a spectral estimation procedure was developed to describe the distribution (over frequency) of the power contained in a signal, based on a finite set of data, and to detect signals buried in wideband noise.
The applicability of Total Harmonic Distortion as diagnostic tool has been studied on a SRU and a short stack. The latter has been tested for more than 6000 h. The experiments show that THD is a good tool for the detection of a faulty cell. Recovery strategies have been developed for the main faults; in case of a leakage in one cell the pressure in both compartments needs to be equilibrated, degradation can for a major part be mitigated by tuning the Fuel Utilization (FU).
In commercial systems the number of physical sensors will be kept at a minimum to reduce costs. Consequently, there is a need to deduce critical parameters from known parameters. So-called soft sensors have been developed and validated for the DIAMOND C system, in particular to reconstruct the minimum and maximum stack temperature as well as the oxygen to carbon ratio.
In both systems controllers have been implemented to improve their efficiency and durability. To this aim, a feed-forward-feedback controllers to ensure fast load tracking were designed. The feed-forward part of the controller reacts to the power demand as well as the stoichiometry of electro-oxidation. The feedback part performs corrections of the controlled system output by additional manipulation of system inputs. Parameters of the feedback controllers were tuned using open-loop step response experiments with both systems. A supervisory optimizer has been developed; the optimizer adjusts references for the low-level controllers to maximise the electrical efficiency of the system while satisfying a number of constraints. The supervisory control module was tested on the DIAMOND-C system and implemented on a standard PLC.A procedure to estimate the remaining useful life has been developed and validated using the DIAMOND-C system. The RUL estimation is central in predictive maintenance of the deployed devices and it can help the control system in accommodating the control strategies in order to extend the achievable lifespan.
The Final Dissemination and Exploitation Workshop was jointly organized by DIAMOND and HEALTH-CODE at European Fuel Cell Forum 2017.
Project Context and Objectives:
The project DIAMOND tackles the problem of monitoring, diagnosing and controlling (MDC) SOFC systems with a holistic approach to achieve advanced management. The close link among these three functions guarantees a comprehensive solution to the problem of achieving improved performance, maintenance scheduling, higher reliability and thus increased lifetime. The key innovation of the project consists in the coupling between diagnosis and control (i.e. the main management tasks) through condition monitoring, which supports the other two functions.
Solid Oxide Fuel Cells (SOFCs) have an outstanding position in the panorama of “green” energy conversion systems. Thanks to their features, both electrical energy and heat can be generated for cogeneration purposes using rich hydrogen fuels. Several projects in Europe, United State and in Asia have demonstrated the feasibility of SOFC technology for a wide power spectrum of power generation systems from houses micro-cogeneration use up to large scale application (some MW).
For micro-cogeneration uses several demonstration projects have been set in the EU to analyze the feasibility of SOFC and demonstrations are running accumulating hours of field experience to improve the future generations of SOFC CHP systems. The main aims of those projects are: i) to consolidate the technology (i.e. materials, configuration, and layout); ii) to improve efficiency and iii) to achieve the main challenges: durability, reliability and optimal working operations of both stack and auxiliaries.
The need for advanced diagnosis and control is of prominent significance to improve the lifetime of SOFCs. Diagnosis algorithms can detect malfunctioning of stack and system components (faults and failures) and give alarms on events that may hinder correct operations or induce either a reduction or a loss of functionality.
Monitoring and advanced diagnosis can give significant insights into the state-of-health for both stack and sub-systems and can guarantee effective control to mitigate the consequences of degradation due to chemical and thermal effects as well as load cycles or faults. Nowadays, control strategies are mainly developed for reference prototypes, without accounting systematically for drift, wear and potential faults; thus simple control strategies are in place without effective adaption or recovery plans. Knowing the status of both stack and system makes it possible to design and run the system closer to the limits, which enables savings on system costs and better operating performance. Therefore, the achievement of accurate and reliable fault detection algorithms will improve control strategies to attain system optimization and longer lifetime.
The practical framework for DIAMOND is formed by two different SOFC-based power generation units: DIAMOND-A and DIAMOND-C. The first one, DIAMOND-A, is a commercial micro-CHP unit built by HYG and based on an integrated stack module (HoTbox™) from HTc. The other one, DIAMOND-C, is a complete middle scale (~10kWe) SOFC CHP system built by VTT.
The two distinct configurations of DIAMOND-A&C cover a broad spectrum of possible SOFC system assemblies, including multi-stack, small and medium power generators, integrated fuel processing and integrated components.
Only limited work has been done on the modeling of integrated balance of plant components for control applications, while the diagnosis problem has not yet been approached at all. Control schemes and diagnosis tools are available for medium-scale systems, largely thanks to their conventional system configuration, but the integration of these results to the systems has not yet been studied. Monitoring, diagnosis and advanced control are the three pillars on which the SOFC-system management will be built. In the last years the first two research areas have been explored with increasing interest thanks to their potentialities to enhance SOFC systems reliability. On the other hand, advanced control has been scarcely investigated and, even more, its integration with diagnosis is still lacking dedicated research. For monitoring and diagnosis the most recent studies prior to the start of DIAMOND were the FCH-JU funded projects DESIGN and GENIUS2. DIAMOND is a continuation of these two projects.
The DIAMOND project aims at developing advanced diagnostic and innovative control strategies to detect stack and system faults and support stack degradation level analysis, as well as to guarantee optimal operations according to the actual status of the entire system; dedicated algorithms will monitor the actual status (condition monitoring) of both stack and components. Knowing the actual status of the entire system, diagnosis and control algorithms will perform the following main actions:
1. Identify faults and potential failures caused by material degradation, system components faults (e.g. blower, heat exchangers, and reformers) or having external causes (e.g. gas composition, erroneous control, critical load).
2. Adapt the controllers of each sub-system to guarantee that all components work at their highest performance levels; this will allow achieving the optimal performance (e.g. maximum efficiency, delivery targeted output power).
The status of the system derived from the monitoring algorithm will support the management functions with advanced features that will overcome the limitations of actual diagnosis and control methods. Therefore a general improvement in accuracy, component faults discrimination and actions anticipation will be also achieved. Thus monitoring will provide the management system with the information needed for the implementation of actions that are not feasible yet for on-board use.
The combination of the above actions will guarantee, above all, the improvement of system reliability, which is the main objective to be pursued to guarantee the successful deployment of SOFC systems.
Within DIAMOND new monitoring, diagnosis and control concepts for two SOFC systems with different layouts. The first objective (1) is to develop a monitoring model structure, which has the capability to simulate different SOFC configurations. Due to the constraints imposed by diagnosis and control, the trade-off between computational time (on-line use) and accuracy (on-field implementation) will be considered as the main issue to be addressed when designing and building the algorithms. Moreover, with the objective of limiting development costs (2) and guaranteeing the feasibility at industrial level (3), the modeling approaches will be selected with the objectives of minimizing the number of experiments required for both system characterization and validation (4). The advanced diagnosis algorithms (5) will be based on both model- and signal-based algorithms. For model-based approach the results achieved before will be improved with respect to the actual state of the art; among others, inverse models will be considered to strengthen the fault isolation process (6).
The achievement of other objectives is expected in the area of SOFC control. Primarily, a hierarchical control structure (7) will be built to guarantee fast load tracking and efficient operation. Adaptive controllers (8), based on either linear or non-linear approaches, will be considered to overcome the variability of the system during lifetime and in case of drift from nominal conditions.
In an industrial perspective, the main goals of the project are the development of advanced diagnosis concepts and new control functions that will be designed, manufactured and installed on a fuel cell system. These concepts will be implemented and tested in the framework of the DIAMOND project.
Summarizing the main S&T objectives of DIAMOND are:
a) Improve management of system components to enhance SOFC performance and lifetime. Warranty on lifetime can be increased. The aim is to achieve a system lifetime of more than 10 years
b) Monitor SOFC stack to predict any deviation from its safe operation envelope, thus increasing time to maintenance. The customer on-site assistance service could be significantly reduced, as the remote monitoring will give much more information on the failure state.
c) Reduce stack degradation rate through a tight management of control variables by adapting their values according to the actual status of the stack.
d) Increase mean time between failures (MTBF).
e) Use existing FC system HW and SW architectures and minimize the implementation burden.
f) Leverage methodologies from mature sectors to reduce development costs of diagnosis and control strategies; among others, Software-In-the-Loop concept will be considered for their testing before on-board implementation.
Project Results:
The project started by specifying the functional requirements for advanced diagnosis and control of SOFC CHP systems
The partners identified the critical faults/failures, their causes and effects, the way how to detect faults and failures along with the countermeasures to be taken. An example of a critical fault or failure is fuel starvation. Fuel starvation can be caused by: a defective MFC / MFM, a software failure, erroneous (too low) electrical current process value (PV), or stack sealing failure; The effect will be a damaged stack; Detection can be done using the following signals and methods: signature in the voltage, deviation in stack performance, T monitoring, e.g. T afterburner and flow monitoring, e.g. towards the afterburner; Possible countermeasure are decrease SP for FU (increase fuel flow), or close NG supply in case of leakage somewhere in the system.
The faults and failures were ranked; fuel starvation and air starvation are considered to be the most critical fault/failure. However, some faults and failures are estimated to be of low importance, they should not be neglected, because each of them can lead to a catastrophic degradation or destruction of the stack. The purpose of ranking is to create the order in which problems should be solved.
Testing of diagnostic tools and controls will be done on two type of systems, an advanced system DIAMOND A, and a conventional system DIAMOND C.
The Diamond A system is an integrated micro-CHP system by HyGear comprising a HTc-HoTbox 2500™. The balance of plant is designed and developed by HyGear. The maximal output of the HoTbox 2500™ is in the range of 2.8 – 2.9 kWe. It is operated as a stand-alone system and as load-demand following. In idle mode the system will be in hot stand-by.
The HB is started up and shut down using the burner. The burner provides the necessary heat to preheat the cathode air inlet and reach the operating target temperature. The necessary fuel to heat up the HB-2500 is referred to as make-up gas. Forming gas is used to avoid nickel re-oxidation during start-up and also to protect the stack in case of system failure. The maximum speed for heating up and cooling down is 100 oC/hr. A specific start-up procedure has been developed by HTc in which important aspects are heat-up rate and the temperature gradients over the stacks. At certain temperature levels the system is switched to a next step in the procedure. The operating window of the system has been defined in terms of minimum and maximum values for the various parameters. During operation the system is controlled by only 8 variables.
In total four modified HoTboxes have been manufactured and delivered during the reporting period to HyGear for integration into the DIAMOND A system and testing. The first HB contained SOFC stack dummies allowing the consortium perform testing and tuning of all operating condition without the risk of stack internal damage that leads to the mixing of fuel and air inside the stack. The dummy stacks contained as many parts from the real stack as possible to mimic the thermal mass and dynamic response of the real stack. This ensured a swift transition from dummy stack operation to real stack.
Each HoTbox was integrated with the “cold” Balance of Plant, followed by functionality tests of the components.
For the preliminary experiments and characterisation tests a system control using a PLC with a computer as User Interface was developed. The start-up sequence as described by HTc was programmed to run automatically. A user intervention is possible by changing the air flow to the burner to control the heat produced in the first phase. The control keeps the variables, like heat-up speed, stack temperature gradient and maximum temperatures within the allowed limits.
Pressures, temperatures, flows, current and voltage are measured, recorded and used to calculate the various pressure drops, temperature gradients, fuel and air utilization, and power output.
Essentially, the system layout is such that under normal operating conditions pressures and pressure drops will be within limits at the maximum flows. If a pressure drop or a pressure exceeds the upper limit, the stack power will be reduced and hence the flow. Similarly, temperatures and temperature gradients will be within the allowed domain at standard operation. Any deviation will be indicative of a problem, and appropriate corrective action will be taken.
The testing of SOFC CHP systems at HyGear is mainly aimed at testing a limited number of operating conditions. Therefore the control is not fully automated. When a change in condition is due, the operator has to adjust the values of the mass flow controllers, blowers, and inverter.
The Diamond C system is a mid-scale (10 kW) SOFC system having a conventional lay-out with separate system components built by VTT. The system is started-up by first heating the stack from ambient to operating temperature and then ramping up the load current to a desired value. Heat is provided to the system by inline electric heaters. During the heat-up, air is supplied to the system and the anode is protected from re-oxidation by providing a reducing atmosphere with either
1) standard pre-mixed safety gases or
2) safety-gas free measures.
Current load-up is carried out after a sufficient operating temperature is reached. Constraints regarding stack temperature and/or temperature distribution have to be obeyed during current load-up.
During nominal operation, the system operating domain is confined by the constraints giving the minimum and maximum values for the various parameters.
No upper-level control strategy is (automatically) implemented in the DIAMOND C control system. Instead, all the controller set points are explicitly or implicitly set by the human operator (cf. the look-up-tables during start-up and shutdown).
To enable the communication of an external device (“DIAMOND device”) with the DIAMOND-C power plant and its existing automation and control system, two data transfer interfaces were prepared:
1) One based on the OPC protocol and the system internal tag name mapping
2) One based on the Modbus/TCP protocol and an external data registry mapping
Both OPC and Modbus are industry standard communication protocols for automation and process control purposes and currently, both protocols are available for system development in the DIAMOND project.
The OPC-based communication protocol utilizes the readily existing tag name mapping, which is provided by the OPC server existing as a part of the DIAMOND-C automation system. Therefore no additional data I/O variable-name mapping was required for this purpose. The downside of OPC is that it is not preferred for fast control actions due to the data always traversing through the OPC server.
To enable direct communication of an external DIAMOND device and the DIAMOND-C control unit (“PLC”), a Modbus/TCP protocol interface was defined. As a part of this task 72 new interface variables and 132 Modbus registry definitions were configured in the system and mapped against the key input and output variables in the DIAMOND-C system. The full interface mapping was documented and distributed to the integration partners, especially INEA for usage in the algorithm embedding efforts.
At the start of the project a short heat-up, load-up and shutdown experiment was carried out with the DIAMOND-C system. The system was operated at maximum current of 50 A, whereas its nominal current is 160 A. The reason for this is that the stack was so worn out during the experiments that voltages of individual cells within the stack decreased too much to allow for safe increase of the stack current.
Between ca. 40 to 90 hours the system was let cool down as maintenance had to be carried out on the exhaust heat recovery heat exchanger. At its end-of-life state, the stack produced so much heat that heat recovery at the exhaust had to be improved in order for the exhaust temperature to remain within safe limits. All in all, the system operated reliably and the updated HMI/SCADA as well as the OPC and Modbus interfaces operated as expected.
For further testing a new stack module and new stacks were manufactured for and installed in the DIAMOND-C system. The DIAMOND-C system BoP part had to be modified slightly to fit the new stack module. A new fuel gas preheater was installed between the BoP module and the stack module to guarantee sufficient fuel temperature at the stack inlet.
Due to the mentioned modifications, it was considered necessary to carry out additional nominal operation testing of the DIAMOND-C system.
The control of fuel cell systems at HyGear and VTT is kept simple with much involvement from an operator. The intention of the project is to operate the systems unattended, and in the second phase of the project to provide software and advance control to adapt the control parameters, like operating window, temperatures, fuel and air utilisation and more, to the changed system health.
Diagnostics
The application of suitable diagnostic methodologies coupled with control actions and reasoned operation manoeuvres can improve FC systems lifetime and reduce their maintenance costs. Indeed, diagnosis allows to quickly detect malfunction states in the entire system and to process adaptive control strategies, so as to bring the system towards safer operating conditions. Generally speaking, fault diagnosis is characterized by three main processes: i) detection, ii) isolation and iii) identification. The first process focuses on the detection of a faulty state occurring in the considered system. Then, the location of the fault is singled out through the isolation process, identifying the component(s) under undesired state. Finally, through the identification process, the fault size and its time-varying behaviour are appraised. A fault diagnosis methodology can be classified according to its performance: FD (fault detection), FDI (fault detection and isolation) and FDII (fault detection, isolation and identification).
In the current literature, three kinds of diagnostic methodologies are usually considered, namely model-, knowledge- and signal-based diagnosis. The model-based approach exploits a mathematical model, physical, grey- or black-box, to simulate the system variables in several operating conditions. Knowledge-based diagnosis is, instead, based on human knowledge or qualitative reasoning, whereas the signal-based approach relates to the direct processing of signals measured on the system. It is worth observing that during the detection process, the acquired information is treated to obtain system state indicators, whose nature depends on methodology used. The distinctive features of each methodology might drive the choice from one approach to another. As an example, with respect to both knowledge- and signal-based approaches, a model-based approach does not require the same amount of experimental data and shows a greater generalizability, due to more evident physical content. However, a main drawback consists in the necessity of specific details on the system features, such as geometry, material, etc. Nevertheless, this methodology is recommended when experimental data are not easy to be performed, due to the complexity of the condition to be reached or the constraints on the system operation.
Despite the considered approach, the extraction of significant features, which feed the detection process, is the core activity of the system monitoring process. Once a feature is extracted (e.g. a residual) it is treated to single out the occurrence of a fault. As an example, model-based approach exploits models to simulate monitored system variables. The residuals are generated as a difference between measured and simulated variables. Then, for the purpose of detection, the residual is compared to specific thresholds, so as to generate an analytical symptom. A symptom is a binary variable representing the occurrence of an abnormal state in the system. If a residual exceeds the threshold, a symptom arises (becoming 1), meaning that an unexpected event happened. If the residual lies below the threshold, the related symptom is 0, meaning that the system is behaving normally. Threshold values are generally computed taking into account several aspects, such as system complexity and measurement uncertainty, and considering the trade-off between accuracy, robustness and false alarm and missed faults probabilities. The robustness of the approach can be improved by using adaptive threshold instead of fixed ones. The symptoms generation is only the first step towards isolation of the faulty component(s). They should be related to the specific fault occurring in the system (fault identification). This process can be performed through Fault Tree Analysis (FTA) methodology, whose main outcome is an inferential isolation tool, namely Fault Signature Matrix (FSM). This matrix univocally links the faults to the symptoms detected during the detection process. The comparison of the collected symptoms with those gathered within the FSM, allows the fault isolation. If a perfect match occurs, the faulty component is located. As a final step, the estimation of residuals magnitude and time behaviour provides further significant information (identification process). Indeed, it is essential to characterize the fault time behaviour to take into account the detection and isolation delay, especially for incipient faults.
Suitable tools and methodologies for enhancing real-time monitoring, diagnosis and control of generic SOFC systems, considering both typical fuel cell system assembly (i.e. DIAMOND-C experimental set-up), as well as Integrated stack modules (ISM, i.e. the DIAMOND-A Hot-box developed and instrumented by HTc) have been under development.
This work mainly consisted of four tasks, whose respective evolution and achievements are synthesized hereinafter. The first task focused on an extended literature analysis on SOFC models and diagnostic schemes. Such a state of the art investigation was particularly intended to serve as a basis for the subsequent development, within the dedicated tasks, of models and methods for condition monitoring and on-field diagnosis. Therefore, this study indicated the most suited modelling approaches and methodologies for SOFC systems monitoring and diagnosis. A deep investigation has been also performed on the state of art related to control-oriented models, in particular 1-D, 0-D and black box models of both SOFC and SOFC systems, as one of the main objectives of the work package is to address the necessary counter-measures to be undertaken whenever a manageable fault arises.
On the basis of the indications provided by the above described state of the art analysis, two tasks, which lasted over the entire project length, were carried-out in parallel in such a way as to achieve the target goals in terms of real-world deployability of models and diagnostic algorithms. Particularly, the second task focused on the selection and development of the most suitable diagnostic methodologies for the SOFC systems under-study. The detection and isolation of faulty states within an SOFC system is a challenging issue, particularly due to the complex and non-linear interactions among the system components. According to several Fault Detection and Isolation (FDI) techniques, an abnormal or undesired state can be detected by comparing residuals to suitable threshold levels. The residuals can be generated by relating the measurements of monitored variables, acquired on the system by means of available devices, to numerical values generated (and this is the case for the DIAMOND project, with particular regard to the Diamond-C system) by a mathematical model. Once at least one residual crosses the related threshold level, an abnormal state is detected and a corresponding analytical symptom is activated. To accomplish the correct isolation of the detected state, all the generated symptoms are compared to specific information relating the aforementioned symptoms to a fault. This latter task can be achieved by means of a Fault Signature Matrix (FSM), which univocally correlates a set of analytical symptoms (i.e. monitored variables) to all the most probable faults the system can encounter. The design of the FSM is usually accomplished following a Fault Tree Analysis (FTA). In the DIAMOND project, a generic FSM was developed, holding directly valid for the DIAMOND-C, whereas a revised version shall be adapted to the DIAMOND-A ISM. More in detail, a Fault Signature Matrix (FSM) is a two-dimensional matrix, in whose rows are listed the faults considered in the FTA, whereas its columns list all the symptoms gathered by accounting for each developed FT. Clearly, repeating symptoms are considered only once. In this way, a complete list of all the variables to be monitored is obtained. In its most generic version, the FSM includes five faults: i) air blower, ii) air leakage between air compressor and pre-heater, iii) temperature controller failure, iv) pre-reformer fault and v) stack fault. Each row has been designed following for each fault the previous FTA and several FTs have been depicted. From all the symptoms, fifteen variables have been chosen as the most significant to be monitored. The definition of an FMS only through an FTA approach might lead to a non-optimized isolation tool, since the correlations among symptoms and faults are defined only through qualitative links, not accounting for the quantitative effect of each fault on the deviation of the affected variables from the nominal conditions. According to the FTA approach, a symptom is triggered as soon as a fault occurs. However, it should be considered that the affected variables might experience small deviations from their nominal values, depending on the fault magnitude. Indeed, for incipient faults, some variables may deviate, but still remaining near their operating condition. The use of a model to simulate the effect of several fault on an SOFC system can thus improve an FSM, evaluating the quantitative deviation of each residual and defining the arisen symptoms with respect to a specific fault magnitude and threshold value (assumed for the fault detection).
The need of suitable models, matching the conflicting needs of accuracy and computational burden, provided clear guidelines to the researchers active within the third task, during which comprehensive models were developed for both DIAMOND-C and DIAMOND-A systems. The activity mainly focused on the development of a model structure destined to on-line monitoring use. The monitoring models were thus conceived in such a way as to support diagnosis and control algorithms, which will be implemented on-board. Consistently with those applications, both black-box and grey-box approaches have been considered to maximize accuracy and minimize both computational time and required experimental data for model characterization. As a first stage, models for the conventional DIAMOND-C system were developed. The modelling work for both stacks and system components has been performed considering a modular structure to use common algorithms for similar components and/or phenomena, thus creating a comprehensive models library. This approach was proven to be beneficial, not only to speed up the second part of the modelling work, which concentrated on the more complex layout of the DIAMOND-A, but also to enable the development of resilient model-based diagnostic tools, capable of performing both fault detection and isolation even with a reduction of number of available sensors. Starting from the results obtained by the DIAMOND-C model, subcomponents were arranged in a new model structure, taking into account thermal exchanges between components. Radiative exchange was modelled through the evaluation of view factors depending on geometrical data due to the layout. The resulting model was proven to be highly versatile and allows enabling or not the thermal exchange of each component. It is thus suitable to support offline configuration sizing. The monitoring model is conceived in such a way as to support diagnostic and control algorithms, entirely implemented on-board. Consistently with those applications, a grey-box approach has been taken into account in order to maximize accuracy and minimize both computational time and the amount of experimental data needed for model characterization. It is finally worth remarking how the overall model development phase greatly benefited from the close interactions with experimenters and hardware assemblers working in the dedicated work-packages.
The final task dealt with the development of both signal- and model-based diagnostic algorithm. The former activity focused on the phenomena that occur progressively and continuously, and do not cause instantaneous irreversible damage, but may be fatal in the long term. The goal is identifying and characterizing possible degradation signatures for diagnostic purposes in line with the typical approaches signal-based procedures present in "blind tests". In order to reach this goal, it is necessary to compare measured data with respect to the typical (i.e. expected) behavior of SOFCs. The frequency analysis, also referred to as spectral analysis, is based on estimation procedures applied to spectrum of the signal computed by the Fast Fourier Transform (FFT) algorithm. The goal of spectral estimation is to describe the distribution (over frequency) of the power contained in a signal, based on a finite set of data. Estimation of power spectra is useful in a variety of applications, including the detection of signals buried in wideband noise. Since voltage is measured in different states, the described approach is applied for every data segment where the state is kept constant (by visual inspection), henceforth referred to as a “voltage segment”: the aim is searching characteristic frequencies of the voltage time series (PSD maxima) and how they vary from a segment to another. The signal-based procedure acting in the frequency domain is applied to data set provided by partners. In particular, it is applied to the output voltage signal measured on the Diamond-C system. Voltage PSD shows some peaky behavior in the presence of considerable variation of Fuel Utilization (FU) w.r.t the typical value used for the SOFC system under test. In Diamond-C only small variation of FU are present. This implies that no faulty conditions due to high FU are present in the analysed data set (without excluding other faulty conditions), and, on the other hand, confirms that PSD-based techniques can be useful to isolate high FU utilization faults as remarked during the Design project.
After the design of appropriate residual generators, a fault is inferred based on the time evolution of a relevant subset of residuals. A statistical method for change detection in residuals was developed within the WT4.4. The non-parametric algorithm builds on the quantification of the change in statistical pattern of the corresponding residual. The simulated results show great potential for detecting change in the residuals, and thus facilitating fault detection.
Regarding the model-based FDI algorithm, the algorithm development started from the characterization of a proper mathematical model through available experimental data, describing the system as a whole and the related components. The adopted model of course consisted in the modelling architecture developed within task 3, as described above. An FTA is performed, once the faults of interest are defined after a fault analysis, to design a heuristic FSM. Then, this FSM is enhanced by fault simulations through the system model, enriched with proper fault sub-models. A sensitivity analysis, imposing fault magnitudes and specific threshold levels, led to the definition of such improved FSM. Moreover, the analysis via isolated component sub-models guaranteed the definition of an Adjoint FSM, which can increase residuals redundancy and solve fault clustering problems. This latter step closed the whole diagnostic algorithm development, with a final outcome consisting in the Adjoint FSM tuned on fault magnitude, threshold level and available measurements.
Nevertheless, the system model (and related component sub-models) can play a significant role in control algorithm design, as well as diagnostic algorithm on-board implementation. Indeed, a mathematical reduction procedure is applied to reduce model complexity and to allow its implementation on specific hardware for on-board uses. For example, PLC devices could not handle complex nonlinear differential equations, and a proper finite difference approach should be chosen to retain model accuracy and allow on-board residual computation (for both complete system and related components). Moreover, the implementation of proper look-up tables can be of advantage for highly reducing computational efforts, while keeping an acceptable physical adherence with respect to the modelled phenomena. The above observations were suitably referred to when interacting with the work-package focusing on PLC hardware implementation of the above described lack of sensor resilient fault detection and isolation methodology.
Diagnostic tools
In state-of-the-art fuel cell stack monitoring techniques, the voltage of either single cells or cell-blocks are monitored separately. This complicates system integration and has a negative impact on the cost of the fuel cell system as a large number of voltage channels may have to be monitored. Total Harmonic Distortion (THD) is a promising low cost technique for detecting a critical cell and stack operation status, such as fuel starvation, from the stack sum voltage only. The technique has been developed at different levels: single repeat unit, short-stack and SOFC system.
First, laboratory characterizations were performed on small size cells in order to qualify and quantify the suitable excitation signal characteristics for the THD analysis. In particular a special attention has been paid to determine suitable frequencies and amplitudes of excitation to be applied for high fuel utilization detection. Then, THD with suitable characteristics was implemented at short stack level with the aim to determine if this technique is promising for detecting faults from the stack sum voltage only. Finally, a THD analysis algorithm was implemented and evaluated online on DIAMOND-A SOFC system at VTT.
The following main results were obtained:
- Based on THD measurements at single cell level, it is deduced that the THD index can be used to observe changes in the FU experienced by the SOFC. A special attention has been paid on the identification of the best parameters for the excitation:
Excitation signal frequency. Changes related to cell FU are most visible in the THD index when the excitation signal frequency is between 1 and 0.01 Hz. This is reasonable, since in this test arrangement the non-linear behaviour of the i-V relationship is caused by changes in the SOFC that are related to fuel mass transport, which is a relatively sluggish process.
Excitation signal amplitude. The THD index obtained with a 1% excitation amplitude appears to be corrupted by process and measurement noise and therefore an AC excitation of only 1% of the DC current is not sufficient to obtain adequate data for reliable THD index calculation. With an AC excitation of 5% of the DC load or more, the THD index increases consistently when FU>0.8 (over the relevant frequency range).
- The implementation of THD has been done on a 6-cells short stack from SOLIDpower. It appears that detection of a faulty cell among 6 cells can be done from the stack voltage only. In this case high FU is detected with THD measurements performed at FU=70 (iDC=28.9 A) with an excitation amplitude iAC=0.8 A (i.e. 10 mA/cm², only 2.8% of iDC) during 10 periods and frequency of 0.01 Hz. It will be very sensitive to use the amplitude of the 1st harmonic as a criteria for the detection. Hence, decreasing the number of cell voltage probes in a stack seems to be realistic regarding this kind of fault.
- Finally, online THD implementation has been done on DIAMOND-C SOFC system. The experimental results have demonstrated that THDA can be a reliable means for quantitatively monitoring the FU rate experienced by the stack and that the algorithm is simple to implement on an embedded controller.
In conclusion, testing efforts have been focused on THD analysis applied at different scale levels: single repeat unit, short-stack and SOFC system. First, the best parameters for the THD excitation have been identified on single repeat units. Then, it has been demonstrated that this THA analysis can be relevant for detecting online high fuel utilization occurring in a stack by measuring the sum stack voltage only.
Also work has been done on the development of stack state-of-health and Remaining Useful Life (RUL) prediction. The novel approach proposed in the DIAMOND project overcomes the limitations of the known approaches and facilitates a reliable RUL prediction in non-stationary operating conditions. The internal resistance of the SOFC stack is estimated from available instrument readings. More specifically, we reconstruct the internal aggregated Ohmic stack area specific resistance (ASR) by using an Unscented Kalman filter (UKF). Since this parameter describes an internal resistance of the stack, it is directly associated with efficiency of power conversion. Therefore, the maximal loss of efficiency that defines RUL of the stack can be easily correlated with the terminal value of ASR.
The entire algorithm consists of the three main parts, executed continuously online: (i) estimation of area specific resistance (ASR) of the stack, (ii) prediction of its future progress based on collected data, and (iii) prediction of RUL. In the first step, a UKF employing a nonlinear SOFC stack model estimates the stack ASR, which is independent of the momentary operating conditions. The UKF employs an SOFC model and measured operating conditions (including temperatures voltage, current, and fuel gas flow rate) to estimate ASR. In the second step, by assuming a linear structure of the temporal drift in the estimated ASR, a momentary change rate of the said ASR estimate is identified by utilizing a standard linear Kalman filter. In the third step, given the momentary estimate of the ASR and its momentary change rate at any time instant, a statistical prognosis of the future development of the stack ASR is performed by using Monte Carlo simulation. Hence the algorithm results in an estimate of the current ASR value as well as the RUL at each time instant of stack operation. The methodology was evaluated on two different stacks, showing in both cases accordance between the predicted and real RUL. The algorithm requires some time to be able to provide reliable RUL predictions, but accurate RUL predictions are provided long before the end-of-the-life of the stacks.
To reduce the number of sensors in a system algorithms for soft sensors have been designed; one for determining the minimum and maximum stack temperatures, and a second to estimate the gas composition at the inlet of a stack.
Thermal stress is one of the main factors affecting the degradation rate of SOFC stacks. In order to mitigate the possibility of fatal thermal stress, stack temperatures and the corresponding thermal gradients need to be continuously controlled during operation. An efficient and consistent approach to data-driven design of the estimator for maximum and minimum stack temperatures was proposed intended (i) to be of high precision, (ii) to be simple to implement on conventional platforms like programmable logic controllers, and (iii) to maintain reliability in spite of degradation processes. By careful application of subspace identification, supported by physical arguments, we derive a simple static estimator capable of producing estimates with 3% error irrespective of the evolving stack degradation. The degradation drift is handled without any explicit modelling. The approach is experimentally validated on a 10 kW SOFC system from VTT.
Degradation and poisoning of SOFC stacks are continuously shortening the lifespan of SOFC systems. Poisoning mechanisms, such as carbon deposition, form a coating layer, hence rapidly decreasing the efficiency of the fuel cells. Gas composition of inlet gases is known to have great impact on the rate of coke formation. Therefore, monitoring of these variables can be of great benefit for overall management of SOFCs. We proposed three distinct approaches for the design of gas composition estimators of an SOFC system in anode off-gas recycle configuration which are (i) accurate, and (ii) easy to implement on a programmable logic controller. Firstly, a classical approach was briefly revisited and problems related to implementation complexity was discussed. Secondly, the model is simplified and adapted for easy implementation. Further, an alternative data-driven approach for gas composition estimation is developed. Finally, a hybrid estimator employing experimental data and 1st principles is proposed. Despite the structural simplicity of the estimators, the experimental validation shows a high precision for all of the approaches. Experimental validation is performed on a 10 kW SOFC system from VTT.
Control
As a first step in the development of advanced control for the DIAMOND systems the control concepts and architectures for μCHP SOFC systems recently adopted in practice have been reviewed. Control system manages and regulates the behaviour of the system. It monitors whether or not system parameters, like temperature or pressure stay within given boundaries. In case of exceeding either lower or upper limit the system should respond in a way that ensures further safe operation. Usually several options exist to do this depending on the control philosophy taken by a particular approach. A primary function of the control is to regulate supply and demand of system output power and heat. For a heat demand driven CHP system an increased demand in heat should lead to an increase in heat produced. This can be achieved by either increasing the stack power and thus fuel supply when operating at a constant utilisation or by solely increasing the fuel supply while keeping the stack power constant,and thus reducing the fuel utilisation. This increased excess of fuel is then burned out in the after-burner producing extra heat. In the former option the extra produced power will be supplied to the grid.
In a power demand driven system an increase in power will lead to an increase in stack power and thus current. When operating at a constant flow the fuel utilisation will increase, and the heat output will decrease; when operating at a constant utilisation, the fuel flow will increase and also the heat output.
One of the main challenges in commercializing the SOFC power systems is how to achieve high electrical efficiency without increasing the degradation rate of fuel cells. The rationale of the work is to design a hierarchical control structure that will ensure fast load tracking and efficient operation of the SOFC systems while taking care of a range of imposed operational constraints and state of health. Such control includes soft sensors algorithms for relevant immeasurable states and disturbances (e.g. maximum stack temperature, fuel concentration, degradation rate of the SOFCs, etc).
For good operation of a SOFC power system well designed low-level controllers are indispensable. The purpose of low-level controllers is to i) supply enough fuel into the system in order to achieve the required stack current and prevent fuel starvation in cells, ii) maintain stack temperature and the corresponding thermal gradients at all times in order to reduce stack degradation due to thermal stress and iii) supply enough water in the system which is needed.
Low-level controllers for SOFC power system were designed by using the feedforward-feedback approach. The feedforward controllers are based on stoichiometry of electro-oxidation, reforming and combustion reactions, which allow immediate response to variable current demand. The feedback controllers perform additional fine adjustment of fuel and air supply in order to minimize the undesired system temperatures variations.
Controllers were implemented and verified on two models of the SOFC power system:
• The detailed physical model of the SOFC Hotbox power (DIAMOND-A) system in gPROMS environment. The model was received from the SOLIDpower and was previously developed by EPFL-LENI.
• The model of the DIAMOND-C power system in Matlab/Simulink environment developed by UNISA within the DIAMOND project.
The proposed low-level control scheme for the DIAMOND-A system consists of four feedforward controllers and three PI control loops and is rather simple for practical implementation.
The selection of pairings of manipulated and controlled variables for control is based on physical knowledge of the system. Input/output pairing for single-loop feedback control is assessed by the relative gain analysis.
Parameters of the PI controllers were tuned from the open-loop step response experiments performed with the DIAMOND-A model at the given operating point. The so-called Magnitude Optimum Multiple Integration (MOMI) method was applied which provides a relatively fast and non-oscillatory closed-loop response for a large class of processes. The method calculates parameters by integrating the process open-loop response after applying the step-change to the process input.
Performance of the proposed control scheme was verified by applying the standard daily load profile of residential houses and large step load changes. The simulations show that the proposed controllers maintain controlled temperatures within defined ranges and provide system electrical efficiency of about 43 %. The lowest efficiency is obtained at load on the lower bound due to the fuel that has to be feed into the burner to maintain the burner temperature. The maximum stack temperature was almost constant, whereas the minimum stack temperature exhibits greater variations. The latter might be a problem during the high load periods when the stack temperature difference is the highest. This can lead to degradation of the electrochemical performance, or in the worst case, malfunction of the system.
Low-level control scheme for DIAMOND C power system consists of four feedforward controllers and seven PI controllers. The control scheme has been tested on the daily load profile and on step-wise changes of the stack current. Simulation results indicate that the strategy successfully controls the main stack variables while taking into account some of the system limitations and delivers high system electrical efficiency (above 65 %). Due to its simplicity, the control appears appropriate for commercial applications.
Running SOFC power systems at the edge of efficiency and maximal longevity is important for their successful market deployment. To properly handle variable load conditions and inevitable degradation processes, on-line optimization is needed to adjust the process variables to run the process at the operational optimum. We proposed a two-level control system in which the low-level control for the DIAMOND-A system is upgraded with a supervisory optimizer. The optimizer adjusts references for the low-level controllers to maximize the electrical efficiency of the system while satisfying a number of constraints. The optimization problem is solved by using the extremum-seeking approach where optimum is sought directly on the process.
A steady-state efficiency analysis was performed at a nominal operating point of a model with different combinations of available optimization variables. Based on these results the references of burner lambda and outlet air temperature were selected as the optimization variables (arguments) of the supervisory optimizer.
The selected optimization problem targets maximization of efficiency and minimization of stack temperature violations. Efficient operation of the optimizer is governed by properly tuned parameters of the optimization algorithm.
A two-level control system with a supervisory controller is verified on a DIAMOND-A model on a range of operational scenarios. The presented results were obtained by simulating the standard load profile of the residential houses. The references of the control scheme without optimizer were fixed at constant values.
Simulations show, that the optimizer improves electrical efficiency of the system by about 8% compared to the efficiency of the system controlled by the low-level controllers. It also keeps the system temperatures within the safe ranges. Due to its simplicity, the optimizer appears appropriate for practical applications.
Similarly, supervisory controller for the DIAMOND-C system is developed and verified on a model of a system in Matlab/Simulink environment. The control system in this case consists of interconnected, independent modules such as: low-level control, soft sensors, condition monitoring module and supervisory control. The purpose of the low-level controllers is to keep the system running at corresponding operating conditions, defined by the controller set-points. The developed soft sensors provide information of non-measurable process variables to the remaining three modules. Namely, the reconstructed temperatures can be used as controlled variables, while the condition monitoring module employs the estimations of gas flow components at the anode inlet. The condition monitoring module provides information on the instantaneous degradation rate, i.e. a measure of stack performance deterioration. Finally, this information is fed to the supervisory controller, which calculates the references for the low-level controller by employing the information about performance degradation rate and the overall efficiency of the system.
The references of the low-level controllers were in this case tightly connected with each other. The reference of the stack outlet air temperature completely defines operating point of the system.
The optimisation problem was formulated to seek an optimal operating point of the SOFC system so that either (i) degradation rate is minimal, (ii) system efficiency is maximal, or (iii) the compromise between the two is achieved.
The optimization problem is solved on the basis of well-known extremum seeking control.
An analysis of the performance criteria on the system level with implemented low-level controller was performed to understand the behaviour of the supervisory controller. To this aim, the individual contributions to the criterion function are evaluated for the only remaining manipulative variable that defines the operating point of the system, cathode outlet temperature. Two facts are apparent from the analysis: (i) system efficiency first increases and then decreases with the increasing temperature of the stack outlet temperature and (ii) the degradation rate drops with the increasing temperature.
The proposed two-level control system for the DIAMOND-C system was verified under different simulation scenarios. Simulation scenarios show that the supervisory controller reaches the optimal operating point which corresponds to the end user optimization objective. Even though the oscillating load current significantly hinders the performance of the supervisory controller, the optimum is still reached. The simulation validation of the proposed control system shows promising preliminary results. Although drawing unambiguous conclusions is unreasonable, one major conclusion is indisputable: “The developed control system has major impact on the degradation rate, and, in turn, to the total life-span system".
For successful integration of the developed low-level and supervisory control software and diagnostics software into the existing CHP system the existing level of instrumentation and control system was reviewed. The detailed system analysis was taken to evaluate a list of signals, measurement devices and actuators used on the system. Within analysis the set point parameters, exchange process values and triggers to influence the system operation were defined.
Where needed the adaptation measures of the system software to be operated in parallel to the prediction and monitoring system were considered. After system examination the standard industrial PLC system was chosen and delivered to HyGear by INEA.
On that platform the development of advanced control algorithms and diagnostics methods were prepared. The advanced control was set up as higher level control (outer control loop) to existing control system. The control and diagnostics was realized as a PLC program library.
The original control software was adopted, to enable it poll for new set points, receive trigger signals and send process variables to the control and diagnostics algorithms.
Advanced control system was applied both to fuel cell and accompanying BoP systems.
The implementation of the developed low-level and supervisory control software and diagnostics software into DIAMOND C was for a large part identical to the procedures followed for DIAMOND A.
As the DIAMOND-C system was already highly instrumented, including a full PLC-based control and automation system, the focus was on the implementation of the advanced control and diagnosis algorithms into a suitable software and computational environment. This was provided with an additional PLC system which communicates with the PLC system through a standard Modbus protocol. A significant work force was spared by not completely changing the current programmable logic controller (PLC) equipment of the VTT system.
Tight teamwork of all participants were needed to complete integration part of control and diagnostics on Diamond-A and Diamond-C. On both systems teams from JSI and INEA work remotely connected to the Diamond systems at HyGear and VTT. On site personnel helped with all necessary adaptations and operator actions on the actual SOFC systems. During both tasks continuous data logging were provided to run off-line analysis for tuning controllers and executing diagnostics algorithms.
The algorithms that were developed in DIAMOND and implemented in industry standard process control devices and subsequently integrated into SOFC systems have been evaluated on both systems.
Extensive testing efforts took place with both systems. In particular, cold testing, hot dummy-testing (system w/o SOFC stack), nominal characterization tests of both systems and finally complete system level algorithm evaluation tests. In addition, existing measurement data from the systems’ operation was utilized for algorithm development before fresh data became available.
Several algorithms for general monitoring and estimation as well as control had been developed before the DIAMOND project. Also many of these were adopted for the SOFC in either a simulated environment or in experimental laboratory arrangements. Also many of the DIAMOND partners (including HYG, UNISA, VTT, CEA, HTc) have participated in this development effort in preceding European projects such as DeSign and GENIUS. The final implementation and validation phase of algorithms in an industrially relevant setting has, however, not been reached in the existing works to the extent as was done in the DIAMOND project.
Several separate system level test runs were prepared and executed during the project. These are described below per test system.
DIAMOND-A:
1) Prototype operational testing; short test run with one stack box dedicated for the testing purposes
2) System characterization; system refurbished with new stack box, step changes in current, I-V curves, and long term operation at various conditions >6000 hours
3) Algorithm evaluation test run; system refurbished with new stack box, tuning control and parameterization of algorithms, ca. 1000 hours
DIAMOND-C:
1) System operability test; short test run with old, de-functional stacks to check balance-of-plant and instrumentation operability after downtime period
2) Hot dummy testing; system operability testing with dummy-stack after stack module modifications related to stack refurbishment, ca. 300 hours
3) System characterization; system refurbished with new stacks and stack module, operational testing and characterization of system in various operating conditions, ca. 400 hours
4) Cold testing with modified system for embedded THD algorithm development, ca. 150 hours
5) Algorithm evaluation; tuning of controllers, parameterization of algorithms and evaluation of algorithm output under various conditions, ca. 700 hours
In addition, several component tests were carried out to facilitate system refurbishment and modification.
The testing effort was a genuine co-operation of all project partners and realized a thorough pre-commercial evaluation of the developed and implemented algorithms in a 100% industrially relevant setup. The testing work carried out in the project revealed several aspects of SOFC system monitoring, diagnostic and control algorithm development work, which have to be taken into account in similar future projects. At the same time, the status of commercially relevant monitoring, diagnostic and control algorithms for SOFC systems was pushed forward significantly.
In summary the following main results were obtained in the testing work, all in an industrially/commercially relevant system level setting:
a) The validation of the functionality and effectiveness of an advanced SOFC system control algorithm, and the applied development procedures
b) The validation of the functionality of novel SOFC stack state-of-health estimation and stack lifetime prognosis algorithm
Subsequently, the validation of the functionality a supervisory controller algorithm based on estimated SOFC stack remaining useful life and efficiency
c) The validation of the functionality and effectiveness of such diagnostically relevant algorithms as (i) the SOFC stack THD index calculation algorithm for realized fuel utilization monitoring and (ii) the online fuel gas composition estimation
d) The validation of correct and consistent operation of a model-based SOFC system diagnostic algorithm.
In conclusion a significant collaborative system level testing effort was carried out as the co-operation of all the DIAMOND project partners. Two separate SOFC power systems were used in the testing and multiple algorithms for the monitoring, control and diagnostics of SOFC systems were validated in a commercially-relevant context as result of the tests.
Activities have been carried out to promote and coordinate dissemination and exploitation activities according to the DoW and to the decision taken by the Project Bodies. The main instruments for the implementation of these actions were the website (http://www.diamond-sofc-project.eu) and the newsletters. A first joint workshop with the project ENDURANCE was organized in the first part of the project (with DIAMOND partners giving seven presentations), whereas a second workshop was jointly organized with the HEALTH-CODE project.
In the first part of the project the main actions of focused on the building the website and setting-up the newsletters editorial tool. Attention to the format was given to have a coherent graphical and communication strategy. The Coordinator selected the IT platforms for the implementation of the website and the newsletters; the former was hosted by HyGear and administered with WordPress open source software, whereas the newsletters was managed on the MailChimp platform. The editing work of website and newsletter (content, graphics and structure) was performed by UNISA under the responsibility of the Dissemination Manager. In the second part of the project the main actions of were the update of publishable information on the website and the sending of the newsletters. In the first period, the DIAMOND website registered 3653 visits; in the second period the visits were 3250, with the following percentage in terms of visiting areas: United States 24.93%, Italy 9.37%, United Kingdom 4.27%, Russia 3.82%, China 3.80%, etc.
As basic communication strategy, each newsletter summarized the most recent activities, which are detailed on the relevant website sections (http://www.diamond-sofc-project.eu/about/newsletter/). Several documents and deliverables concerning modelling, diagnostic and control activities have been made available on the website. Seven newsletters have been circulated, aiming at updating the industry and research communities active in the field of modelling, monitoring, diagnostics and control of SOFC. The newsletters mailing list has been populated carefully by including relevant recipients belonging to FC and Hydrogen industry, research, JU and other stakeholders. The initial database had 156 email addresses, which growth up to 329 contacts by the end of the project.
The results of DIAMOND have been presented at the following events:
• 8th Bruges Workshop on Progress in Fuel Cell Systems, June 2-3, 2015 (Bruges, Belgium);
• ENDURANCE Joint Workshop “SOFC Diagnostic” September 14, 2015 (Genoa, Italy);
• FCH-JU Programme Review 2015, November 17-19, 2015 (Bruxelles, Belgium);
• FCH-JU Programme Review 2016, November 21-22, 2016 (Bruxelles, Belgium);
• ENDURANCE & SOPHIA Workshop “Degradation on SOFC”, February 17, 2017 (Barcelona, Spain).
Twelve papers were written and presented at six International Conferences:
• Feedforward-Feedback Control of a SOFC Power System: A Simulation Study, ECS Conference on Electrochemical Energy Conversion & Storage with SOFC-XIV, July 26-31, 2015;
• Soft sensor design for estimation of a SOFC stack temperatures and oxygen-to-carbon ratio. ECS Conference on Electrochemical Energy Conversion & Storage with SOFC-XIV, July 26-31, 2015;
• A Lumped Dynamic Modelling Approach for Model-Based Control and Diagnosis of Solid Oxide Fuel Cell System with Anode Off-Gas Recycling, ECS Conference on Electrochemical Energy Conversion & Storage with SOFC-XIV, July 26-31, 2015;
• Improved fault isolability for solid oxide fuel cell diagnosis through sub-system analysis, 8th International Conference on Applied Energy (ICAE), October 8-11, 2016, Beijing, China;
• Development of a dynamic model for diagnosis and control of an integrated stack module based on solid oxide fuel cells. 8th International Conference on Applied Energy (ICAE), October 8-11, 2016, Beijing, China;
• Estimating the level of degradation in fuel cell stacks, Proceedings of the 25th International Electrotechnical and Computer Science Conference ERK 2016, Portoroz, Slovenia, September 19-21, 2016;
• Control design for a 2.5 kW solid oxide fuel cell power system, Proceedings of the 25th International Electrotechnical and Computer Science Conference ERK 2016, Portoroz, Slovenia, September 19-21, 2016;
• Accounting for modelling errors in model-based diagnosis by using Gaussian process models, Proceedings of the 3rd International Conference on Control and Fault-Tolerant Systems, SysTOL 2016, Barcelona, September 7-9, 2016, pp. 510-515.
• Supervisory control of a 2.5 kW SOFC power system. Proceedings of the 10th Professional Conference on Automation in Industry and Economy, AIG’ 2017, Maribor, Slovenia, April 6-7, 2017, pp. 159-166.
• The new concept of the solid-oxide fuel cells stack control based on device status. Proceedings of the 10th Professional Conference on Automation in Industry and Economy, AIG’ 2017, Maribor, Slovenia, April 6-7, 2017, pp. 151-158.
• Hybrid approach to remaining useful life prediction of solid oxide fuel cell stack. 15th International Symposium on Solid Oxide Fuel Cells SOFC-XV, July 23-28, 2017, Hollywood, Florida, USA. ECS transactions, vol. 78, no. 1, 2251-2264.
• Improving operation of a 2.5kW SOFC power system with supervisory control. 15th International Symposium on Solid Oxide Fuel Cells SOFC-XV, July 23-28, 2017, Hollywood, Florida, USA. ECS Transactions, vol. 78, no., pp 265-274.
Five papers have been published in four international peer-reviewed journals:
• A comprehensive 3-D modeling of a single planar solid oxide fuel cell, International Journal of Hydrogen Energy, vol. 41, n. 5: 3613-3627;
• Online estimation of internal stack temperatures in solid oxide fuel cell power generating units, Journal of Power Sources vol. 336: 251-260;
• Online gas composition estimation in solid oxide fuel cell systems with anode off-gas recycle configuration, Journal of Power Sources, vol. 336:251-260;
• A model-based diagnostic technique to enhance fault isolability in Solid Oxide Fuel Cell systems, Applied Energy, vol 204:1198-1214;
• State of health estimation and remaining useful life prediction of solid oxide fuel cell stack. Energy Conversion and Management, 2017, vol. 148, pp. 993-1002.
One journal article manuscript is close to finalization:
• Online total harmonic distortion analysis for solid oxide fuel cell stack monitoring in a system application, 2017
A poster was presented at the 9th IFAC Symposium SAFEPROCESS that took place in Paris on September 3, 2015. The poster was at display in the special session “Industrial day”, dedicated to dissemination of the latest cases of industry-academia cooperation. The contribution of the consortium team was Diagnosis-aided control for SOFC power systems (DIAMOND), 9th IFAC Symposium SAFEPROCESS, Paris 2-4 September 2015. Posters accepted for presentation at the 7th European Fuel Cell Technology & Applications Conference – Piero Lunghi Conference EFC2017, 12th -15th December 2017, Naples (ITALY) are: The model-based diagnostic algorithm implementation and testing; The fault simulation study performed with the DIAMOND-A monitoring model.
Moreover, IJS partner organized three seminars at the Department of Systems and Control, Jozef Stefan Institute, Ljubljana, Slovenia and invited UNISA representative at two-days lecture for the Slovenian Simulation Society.
Two Workshops have been planned to exchange the latest results and developments of the project: Mid-term exploitation workshop (December 16th, 2015) and Final and exploitation workshop (July 4th, 2017). A joint workshop with the project ENDURANCE has been held on 14th September 2015 and a joint workshop with ENDURANCE & SOPHIA projects has been held on 17th February 2017. Moreover, a final joint workshop was held on 4th July 2017 in Lucerne with HEALTH-CODE project
Mid-term exploitation Workshop – Workshop on Monitoring, Diagnostics and Control for SOFC systems. Improving SOFC-based CHP performance through innovative diagnosis and control focused on the most recent advances and the current research on monitoring, diagnostics and control for SOFC systems. The objective of the workshop was to present the DIAMOND approach and the recent project results for CHP systems equipped with both conventional stacks and integrated modules. The workshop was hosted by 2015 European Fuel Cell Conference & Exhibition at Hotel Royal Continental, Naples (I).
Joint workshop on SOFC diagnostic Diamond & Endurance FCH-JU Projects focused on the activities of both DIAMOND and ENDURANCE projects whose objectives converge towards the improvement of SOFC performance and durability. The workshop was hosted by University of Genova (I). Joint workshop Degradation mechanisms in Solid Oxide cells and systems Endurance & SOPHIA FCH-JU Projects aimed to provide a platform for professional discussions and exchange of expertise in one of the most challenging issues in the commercialization of SOFC: degradation and durability.
Final Dissemination and Exploitation Workshop – One Day Workshop on Monitoring, Diagnostics and Control for FC focused on the implementation and use of the technology beyond the project duration and summarised the progress towards the exploitation by industrial partners and potential customers.
The results achieved for DIAMOND-C and DIAMOND-A SOFC-based systems together with PEM-based µ-CHP and Backup systems were presented in a unique and comprehensive framework. Presentations dealt with methodologies and applications for both SOFC and PEMFC. Control and diagnostic techniques were reviewed along with experimental and modelling techniques suited for their design and implementation on real FC systems. A wide range of Model-based and EIS-based approaches were described and together with their exploitation for fault detection in stack and BOP. A special contribution was given by Y. Tsur (TECHNION): “Impedance spectroscopy analysis using genetic programming (ISGP): a short tutorial”. The workshop was jointly organized by FP7 & H2020 FCH-JU funded projects DIAMOND and HEALTH-CODE at European Fuel Cell Forum – KKL Lucerne (CH). More than 45 attendants from 16 countries were hosted free of charge. Lunch, coffee and beverages were offered to all participants.
Concerning the exploitation strategy, the industrial roadmap on behalf of all the relevant technologies of diagnostics and advanced SOFC CHP system control was summarised in both deliverables D8.5 and D8.6. The possible exploitable results will be the developed diagnostic tools, the advanced control and the system component models and the methodology to develop them. At the various levels the possible markets are Stack manufacturers, System integrators, Automation industry and Boiler companies. The general approach of exploitation of results is directly linked to the interests of the partners involved in DIAMOND. Both, technical and strategic exploitation can be applied to internal and external purposes of the partners. The internal usage will definitely lead to improvements of the demonstrated technology but also to future research on this topic. The external exploitation requires more effort on this task. Here, not only the involved partners are affected. The external exploitation has to address public and commercial stakeholders and the end-users of the demonstrated technology. The specific details for the different stakeholders need to include data on reliable operation of the μ-CHP system using the diagnosis and advanced control.
Potential Impact:
Potential impact
The impact from the R&D done within the DIAMOND project can be divided in various categories:
• Technical impact
• Commercial
• International
• Educational
• Integration in Europe
• Environmental
Technical impacts
Within the project monitoring, diagnostic and control techniques have been developed to improve performance and durability of SOFC systems. These developments have been tested on two different SOFC systems: a conventional one, DIAMOND-C and an advanced system based on an integrated HoTbox, DIAMOND-A.
In more detail the following technical improvements were reached:
• System modes were developed and validated;
• Fault Detection and Isolation (FDI) methodology using fault signature matrices based on fault-tree analyses has been developed and validated;
• A signal-based diagnosis methodology was developed;
• The applicability of Total Harmonic Distortion as diagnostic tool has been verified on stack level;
• Advanced low-level control has been developed, implemented and tested;
• Supervisory control has been developed and tested;
• Soft sensors have been developed;
• A Remaining Useful Life (RUL) estimator has bas been developed and validated.
Commercial impacts
The project was set up to develop and validate advanced control, monitoring and diagnostic tools for SOFC CHP-systems. On system level commercial impacts will arise only after several years. Additional development and cost reductions are needed. The developed methodologies will be used in future systems
HyGear produces hydrogen generation systems but is expanding its portfolio also to fuel cell system which consume hydrogen. The DIAMOND project has provided HyGear with the knowledge to broaden its technology base for in future years.
HTc is a major developer and supplier of SOFC systems. It is constantly improving its cells, stacks, and systems. The results obtained in the DIAMOND project will assist HTc in this effort.
INEA is a supplier of embedded systems for various applications. The outcome of the DIAMOND project will strengthen their visibility in the fuel cell market.
International Impacts
The outcome of the project will help the partners within the project to compete on the international stage with SOFC systems and its components. Performance, durability and reliability, all addressed within the project, are important factors for success in the market.
Educational Impacts
UNISA is an important institute for education, training and PhD students in the field of system modelling, diagnostics and fuel cell and electrolyser research. The research institutes VTT, CEA, and IJS all involve master and PhD students in their research. During this project 2 PhD students have worked on the project (IJS, INEA)
Impact on Europe
The project has strengthened the European position in the field of SOFC-systems development of academia, research institutes and SMEs/industry. Much knowledge has been gained on diagnostic tools, fault detection and identification methodology, system and system component modelling, and control.
The RTD performers gained knowledge and experience for future research and support of the industry. The SMEs will introduce the gained knowledge into their products. Additionally the knowledge gained can also be applied to other fuel cell technologies, but also to electrolysis systems.
Environmental and Economic Impacts
The technology developed can be used to improve future SOFC systems. Widespread adoption of SOFC technologies will give the environmental benefits of reduced fossil fuel usage (via increased efficiency) and reduced emissions.
Dissemination activities
The results of DIAMOND have been presented at the following events:
• 8th Bruges Workshop on Progress in Fuel Cell Systems, June 2-3, 2015 (Bruges, Belgium);
• ENDURANCE Joint Workshop “SOFC Diagnostic” September 14, 2015 (Genoa, Italy);
• FCH-JU Programme Review 2015, November 17-19, 2015 (Bruxelles, Belgium);
• FCH-JU Programme Review 2016, November 21-22, 2016 (Bruxelles, Belgium);
• ENDURANCE & SOPHIA Workshop “Degradation on SOFC”, February 17, 2017 (Barcelona, Spain).
Twelve papers were written and presented at six International Conferences:
• Feedforward-Feedback Control of a SOFC Power System: A Simulation Study, ECS Conference on Electrochemical Energy Conversion & Storage with SOFC-XIV, July 26-31, 2015;
• Soft sensor design for estimation of a SOFC stack temperatures and oxygen-to-carbon ratio. ECS Conference on Electrochemical Energy Conversion & Storage with SOFC-XIV, July 26-31, 2015;
• A Lumped Dynamic Modelling Approach for Model-Based Control and Diagnosis of Solid Oxide Fuel Cell System with Anode Off-Gas Recycling, ECS Conference on Electrochemical Energy Conversion & Storage with SOFC-XIV, July 26-31, 2015;
• Improved fault isolability for solid oxide fuel cell diagnosis through sub-system analysis, 8th International Conference on Applied Energy (ICAE), October 8-11, 2016, Beijing, China;
• Development of a dynamic model for diagnosis and control of an integrated stack module based on solid oxide fuel cells. 8th International Conference on Applied Energy (ICAE), October 8-11, 2016, Beijing, China;
• Estimating the level of degradation in fuel cell stacks, Proceedings of the 25th International Electrotechnical and Computer Science Conference ERK 2016, Portoroz, Slovenia, September 19-21, 2016;
• Control design for a 2.5 kW solid oxide fuel cell power system, Proceedings of the 25th International Electrotechnical and Computer Science Conference ERK 2016, Portoroz, Slovenia, September 19-21, 2016;
• Accounting for modelling errors in model-based diagnosis by using Gaussian process models, Proceedings of the 3rd International Conference on Control and Fault-Tolerant Systems, SysTOL 2016, Barcelona, September 7-9, 2016, pp. 510-515.
• Supervisory control of a 2.5 kW SOFC power system. Proceedings of the 10th Professional Conference on Automation in Industry and Economy, AIG’ 2017, Maribor, Slovenia, April 6-7, 2017, pp. 159-166.
• The new concept of the solid-oxide fuel cells stack control based on device status. Proceedings of the 10th Professional Conference on Automation in Industry and Economy, AIG’ 2017, Maribor, Slovenia, April 6-7, 2017, pp. 151-158.
• Hybrid approach to remaining useful life prediction of solid oxide fuel cell stack. 15th International Symposium on Solid Oxide Fuel Cells SOFC-XV, July 23-28, 2017, Hollywood, Florida, USA. ECS transactions, vol. 78, no. 1, 2251-2264.
• Improving operation of a 2.5kW SOFC power system with supervisory control. 15th International Symposium on Solid Oxide Fuel Cells SOFC-XV, July 23-28, 2017, Hollywood, Florida, USA. ECS Transactions, vol. 78, no., pp 265-274.
Five papers have been published in four international peer-reviewed journals:
• A comprehensive 3-D modeling of a single planar solid oxide fuel cell, International Journal of Hydrogen Energy, vol. 41, n. 5: 3613-3627;
• Online estimation of internal stack temperatures in solid oxide fuel cell power generating units, Journal of Power Sources vol. 336: 251-260;
• Online gas composition estimation in solid oxide fuel cell systems with anode off-gas recycle configuration, Journal of Power Sources, vol. 336:251-260;
• A model-based diagnostic technique to enhance fault isolability in Solid Oxide Fuel Cell systems, Applied Energy, vol 204:1198-1214;
• State of health estimation and remaining useful life prediction of solid oxide fuel cell stack. Energy Conversion and Management, 2017, vol. 148, pp. 993-1002.
One journal article manuscript is close to finalization:
• Online total harmonic distortion analysis for solid oxide fuel cell stack monitoring in a system application, 2017
A poster was presented at the 9th IFAC Symposium SAFEPROCESS that took place in Paris on September 3, 2015. The poster was at display in the special session “Industrial day”, dedicated to dissemination of the latest cases of industry-academia cooperation. The contribution of the consortium team was Diagnosis-aided control for SOFC power systems (DIAMOND), 9th IFAC Symposium SAFEPROCESS, Paris 2-4 September 2015. Posters accepted for presentation at the 7th European Fuel Cell Technology & Applications Conference – Piero Lunghi Conference EFC2017, 12th -15th December 2017, Naples (ITALY) are: The model-based diagnostic algorithm implementation and testing; The fault simulation study performed with the DIAMOND-A monitoring model.
Moreover, IJS partner organized three seminars at the Department of Systems and Control, Jozef Stefan Institute, Ljubljana, Slovenia and invited UNISA representative at two-days lecture for the Slovenian Simulation Society.
In the framework of the task 8.3 two Workshops have been planned to exchange the latest results and developments of the project: Mid-term exploitation workshop (December 16th, 2015) and Final and exploitation workshop (July 4th, 2017). A joint workshop with the project ENDURANCE has been held on 14th September 2015 and a joint workshop with ENDURANCE & SOPHIA projects has been held on 17th February 2017. Moreover, a final joint workshop was held on 4th July 2017 in Lucerne with HEALTH-CODE project
Mid-term exploitation Workshop – Workshop on Monitoring, Diagnostics and Control for SOFC systems. Improving SOFC-based CHP performance through innovative diagnosis and control focused on the most recent advances and the current research on monitoring, diagnostics and control for SOFC systems. The objective of the workshop was to present the DIAMOND approach and the recent project results for CHP systems equipped with both conventional stacks and integrated modules. The workshop was hosted by 2015 European Fuel Cell Conference & Exhibition at Hotel Royal Continental, Naples (I).
Joint workshop on SOFC diagnostic Diamond & Endurance FCH-JU Projects focused on the activities of both DIAMOND and ENDURANCE projects whose objectives converge towards the improvement of SOFC performance and durability. The workshop was hosted by University of Genova (I). Joint workshop Degradation mechanisms in Solid Oxide cells and systems Endurance & SOPHIA FCH-JU Projects aimed to provide a platform for professional discussions and exchange of expertise in one of the most challenging issues in the commercialization of SOFC: degradation and durability.
Final Dissemination and Exploitation Workshop – One Day Workshop on Monitoring, Diagnostics and Control for FC focused on the implementation and use of the technology beyond the project duration and summarised the progress towards the exploitation by industrial partners and potential customers.
The results achieved for DIAMOND-C and DIAMOND-A SOFC-based systems together with PEM-based µ-CHP and Backup systems were presented in a unique and comprehensive framework. Presentations dealt with methodologies and applications for both SOFC and PEMFC. Control and diagnostic techniques were reviewed along with experimental and modelling techniques suited for their design and implementation on real FC systems. A wide range of Model-based and EIS-based approaches were described and together with their exploitation for fault detection in stack and BOP. A special contribution was given by Y. Tsur (TECHNION): “Impedance spectroscopy analysis using genetic programming (ISGP): a short tutorial”. The workshop was jointly organized by FP7 & H2020 FCH-JU funded projects DIAMOND and HEALTH-CODE at European Fuel Cell Forum – KKL Lucerne (CH). More than 45 attendants from 16 countries were hosted free of charge. Lunch, coffee and beverages were offered to all participants.
Exploitation of results
The DIAMOND consortium can be divided into two partner groups, Industry and RTD-performers. Each group will identify the best type of exploitation action and the specific audience of the exploitation activities. Some of the exploitation activities will have direct impact on the technical improvements within a short time frame, while others will lead to future impact for the end users and technology manufacturers.
The industrial partners within DIAMOND will focus their exploitation activities on improving their current technology and business position in existing markets and on the creation of new markets beyond the markets addressed in DIAMOND. The industrial partners will use the technical improvements in a direct manner to shorten turn-around times to product readiness.
The exploitation goals of the RTD performers are different, yet complementary to those of the industrial partners. The technical improvements and developments will be integrated into the teaching courses but will also flow into the research activities of the RTD performers. For instance CEA’s business model consists in developing technologies for the industry and bringing them to market thanks to technology transfer to industry, either large industries, SME’s or start-ups. VTT has a similar approach.
This approach will lead to being more attractive for Ph.D. master and graduate level students to join the RTD performers. More strategically, graduates from academic partners will be trained for working in the related industry but also in related research work. The long-term pay-off of the RTD performers will lift DAMOND’s technology being general part of teaching lessons.
As an example the methodology developed by UNISA in the frame of DIAMOND for modelling systems is very versatile as has been shown. It can be easily applied to other system architectures.
The exploitable results are listed below in groups, each group followed by short plans for exploitation.
Exploitable result: Diagnostics
1 Total Harmonic Distortion
2 Failure detection and Isolation methodology
3 Soft Sensors
4 SOFC system modelling
Exploitation plans
1 The developed method will provide CEA a good international visibility and will allow CEA to participate in future projects for further development, either national or international, collaborative or bilateral with industries.
2 The methodology is versatile and can be applied in other Fuel Cell systems. It will provide UNISA a good international visibility and will allow UNISA to participate in future projects for further development, either national or international, collaborative or bilateral with industries.
3 The developed methodology will be exploited in future projects for further development, either national or international, collaborative or bilateral with industries
4 All models developed in the frame of DIAMOND by UNISA will be used for several forthcoming studies and projects dedicated to fuel cells.
Exploitable result: Controls
1 Embedded control
2 Supervisory control
Exploitation plans
INEA and IJS will exploit the developed control methodology in future projects in the field of fuel cell systems.
List of Websites:
http://www.diamond-sofc-project.eu/
The project DIAMOND tackles the problem of monitoring, diagnosing and controlling SOFC systems with a holistic approach to achieve advanced management. The consortium has developed monitoring, diagnostic and control techniques to improve performance and durability. These developments have been tested on two different SOFC systems: a conventional one, DIAMOND-C and an advanced system based on an integrated HoTbox, DIAMOND-A.
A thorough inventory of faults and failures which might occur in SOFC CHP systems has been made. The most important ones like, for example, high fuel and air utilisation, leakages, too high temperatures and out of specification steam to carbon ratios, have been of particular focus in the project.
By using fault-tree analyses for the systems, fault signature matrices have been constructed, which are used for Fault Detection and Isolation (FDI) methodology. To develop and implement model-based diagnosis and control strategies, a set of models was proposed and tested. The DIAMOND-C system model was improved using the data delivered by VTT obtained using the new stacks. All system models were properly coded for their implementation on PLC and tested. The DIAMOND-A system was modelled making use of a dynamic lumped gray-box approach and validated against experimental data.
A signal-based diagnosis was developed focusing on degradation phenomena. In order to detect degrading signatures, a spectral estimation procedure was developed to describe the distribution (over frequency) of the power contained in a signal, based on a finite set of data, and to detect signals buried in wideband noise.
The applicability of Total Harmonic Distortion as diagnostic tool has been studied on a SRU and a short stack. The latter has been tested for more than 6000 h. The experiments show that THD is a good tool for the detection of a faulty cell. Recovery strategies have been developed for the main faults; in case of a leakage in one cell the pressure in both compartments needs to be equilibrated, degradation can for a major part be mitigated by tuning the Fuel Utilization (FU).
In commercial systems the number of physical sensors will be kept at a minimum to reduce costs. Consequently, there is a need to deduce critical parameters from known parameters. So-called soft sensors have been developed and validated for the DIAMOND C system, in particular to reconstruct the minimum and maximum stack temperature as well as the oxygen to carbon ratio.
In both systems controllers have been implemented to improve their efficiency and durability. To this aim, a feed-forward-feedback controllers to ensure fast load tracking were designed. The feed-forward part of the controller reacts to the power demand as well as the stoichiometry of electro-oxidation. The feedback part performs corrections of the controlled system output by additional manipulation of system inputs. Parameters of the feedback controllers were tuned using open-loop step response experiments with both systems. A supervisory optimizer has been developed; the optimizer adjusts references for the low-level controllers to maximise the electrical efficiency of the system while satisfying a number of constraints. The supervisory control module was tested on the DIAMOND-C system and implemented on a standard PLC.A procedure to estimate the remaining useful life has been developed and validated using the DIAMOND-C system. The RUL estimation is central in predictive maintenance of the deployed devices and it can help the control system in accommodating the control strategies in order to extend the achievable lifespan.
The Final Dissemination and Exploitation Workshop was jointly organized by DIAMOND and HEALTH-CODE at European Fuel Cell Forum 2017.
Project Context and Objectives:
The project DIAMOND tackles the problem of monitoring, diagnosing and controlling (MDC) SOFC systems with a holistic approach to achieve advanced management. The close link among these three functions guarantees a comprehensive solution to the problem of achieving improved performance, maintenance scheduling, higher reliability and thus increased lifetime. The key innovation of the project consists in the coupling between diagnosis and control (i.e. the main management tasks) through condition monitoring, which supports the other two functions.
Solid Oxide Fuel Cells (SOFCs) have an outstanding position in the panorama of “green” energy conversion systems. Thanks to their features, both electrical energy and heat can be generated for cogeneration purposes using rich hydrogen fuels. Several projects in Europe, United State and in Asia have demonstrated the feasibility of SOFC technology for a wide power spectrum of power generation systems from houses micro-cogeneration use up to large scale application (some MW).
For micro-cogeneration uses several demonstration projects have been set in the EU to analyze the feasibility of SOFC and demonstrations are running accumulating hours of field experience to improve the future generations of SOFC CHP systems. The main aims of those projects are: i) to consolidate the technology (i.e. materials, configuration, and layout); ii) to improve efficiency and iii) to achieve the main challenges: durability, reliability and optimal working operations of both stack and auxiliaries.
The need for advanced diagnosis and control is of prominent significance to improve the lifetime of SOFCs. Diagnosis algorithms can detect malfunctioning of stack and system components (faults and failures) and give alarms on events that may hinder correct operations or induce either a reduction or a loss of functionality.
Monitoring and advanced diagnosis can give significant insights into the state-of-health for both stack and sub-systems and can guarantee effective control to mitigate the consequences of degradation due to chemical and thermal effects as well as load cycles or faults. Nowadays, control strategies are mainly developed for reference prototypes, without accounting systematically for drift, wear and potential faults; thus simple control strategies are in place without effective adaption or recovery plans. Knowing the status of both stack and system makes it possible to design and run the system closer to the limits, which enables savings on system costs and better operating performance. Therefore, the achievement of accurate and reliable fault detection algorithms will improve control strategies to attain system optimization and longer lifetime.
The practical framework for DIAMOND is formed by two different SOFC-based power generation units: DIAMOND-A and DIAMOND-C. The first one, DIAMOND-A, is a commercial micro-CHP unit built by HYG and based on an integrated stack module (HoTbox™) from HTc. The other one, DIAMOND-C, is a complete middle scale (~10kWe) SOFC CHP system built by VTT.
The two distinct configurations of DIAMOND-A&C cover a broad spectrum of possible SOFC system assemblies, including multi-stack, small and medium power generators, integrated fuel processing and integrated components.
Only limited work has been done on the modeling of integrated balance of plant components for control applications, while the diagnosis problem has not yet been approached at all. Control schemes and diagnosis tools are available for medium-scale systems, largely thanks to their conventional system configuration, but the integration of these results to the systems has not yet been studied. Monitoring, diagnosis and advanced control are the three pillars on which the SOFC-system management will be built. In the last years the first two research areas have been explored with increasing interest thanks to their potentialities to enhance SOFC systems reliability. On the other hand, advanced control has been scarcely investigated and, even more, its integration with diagnosis is still lacking dedicated research. For monitoring and diagnosis the most recent studies prior to the start of DIAMOND were the FCH-JU funded projects DESIGN and GENIUS2. DIAMOND is a continuation of these two projects.
The DIAMOND project aims at developing advanced diagnostic and innovative control strategies to detect stack and system faults and support stack degradation level analysis, as well as to guarantee optimal operations according to the actual status of the entire system; dedicated algorithms will monitor the actual status (condition monitoring) of both stack and components. Knowing the actual status of the entire system, diagnosis and control algorithms will perform the following main actions:
1. Identify faults and potential failures caused by material degradation, system components faults (e.g. blower, heat exchangers, and reformers) or having external causes (e.g. gas composition, erroneous control, critical load).
2. Adapt the controllers of each sub-system to guarantee that all components work at their highest performance levels; this will allow achieving the optimal performance (e.g. maximum efficiency, delivery targeted output power).
The status of the system derived from the monitoring algorithm will support the management functions with advanced features that will overcome the limitations of actual diagnosis and control methods. Therefore a general improvement in accuracy, component faults discrimination and actions anticipation will be also achieved. Thus monitoring will provide the management system with the information needed for the implementation of actions that are not feasible yet for on-board use.
The combination of the above actions will guarantee, above all, the improvement of system reliability, which is the main objective to be pursued to guarantee the successful deployment of SOFC systems.
Within DIAMOND new monitoring, diagnosis and control concepts for two SOFC systems with different layouts. The first objective (1) is to develop a monitoring model structure, which has the capability to simulate different SOFC configurations. Due to the constraints imposed by diagnosis and control, the trade-off between computational time (on-line use) and accuracy (on-field implementation) will be considered as the main issue to be addressed when designing and building the algorithms. Moreover, with the objective of limiting development costs (2) and guaranteeing the feasibility at industrial level (3), the modeling approaches will be selected with the objectives of minimizing the number of experiments required for both system characterization and validation (4). The advanced diagnosis algorithms (5) will be based on both model- and signal-based algorithms. For model-based approach the results achieved before will be improved with respect to the actual state of the art; among others, inverse models will be considered to strengthen the fault isolation process (6).
The achievement of other objectives is expected in the area of SOFC control. Primarily, a hierarchical control structure (7) will be built to guarantee fast load tracking and efficient operation. Adaptive controllers (8), based on either linear or non-linear approaches, will be considered to overcome the variability of the system during lifetime and in case of drift from nominal conditions.
In an industrial perspective, the main goals of the project are the development of advanced diagnosis concepts and new control functions that will be designed, manufactured and installed on a fuel cell system. These concepts will be implemented and tested in the framework of the DIAMOND project.
Summarizing the main S&T objectives of DIAMOND are:
a) Improve management of system components to enhance SOFC performance and lifetime. Warranty on lifetime can be increased. The aim is to achieve a system lifetime of more than 10 years
b) Monitor SOFC stack to predict any deviation from its safe operation envelope, thus increasing time to maintenance. The customer on-site assistance service could be significantly reduced, as the remote monitoring will give much more information on the failure state.
c) Reduce stack degradation rate through a tight management of control variables by adapting their values according to the actual status of the stack.
d) Increase mean time between failures (MTBF).
e) Use existing FC system HW and SW architectures and minimize the implementation burden.
f) Leverage methodologies from mature sectors to reduce development costs of diagnosis and control strategies; among others, Software-In-the-Loop concept will be considered for their testing before on-board implementation.
Project Results:
The project started by specifying the functional requirements for advanced diagnosis and control of SOFC CHP systems
The partners identified the critical faults/failures, their causes and effects, the way how to detect faults and failures along with the countermeasures to be taken. An example of a critical fault or failure is fuel starvation. Fuel starvation can be caused by: a defective MFC / MFM, a software failure, erroneous (too low) electrical current process value (PV), or stack sealing failure; The effect will be a damaged stack; Detection can be done using the following signals and methods: signature in the voltage, deviation in stack performance, T monitoring, e.g. T afterburner and flow monitoring, e.g. towards the afterburner; Possible countermeasure are decrease SP for FU (increase fuel flow), or close NG supply in case of leakage somewhere in the system.
The faults and failures were ranked; fuel starvation and air starvation are considered to be the most critical fault/failure. However, some faults and failures are estimated to be of low importance, they should not be neglected, because each of them can lead to a catastrophic degradation or destruction of the stack. The purpose of ranking is to create the order in which problems should be solved.
Testing of diagnostic tools and controls will be done on two type of systems, an advanced system DIAMOND A, and a conventional system DIAMOND C.
The Diamond A system is an integrated micro-CHP system by HyGear comprising a HTc-HoTbox 2500™. The balance of plant is designed and developed by HyGear. The maximal output of the HoTbox 2500™ is in the range of 2.8 – 2.9 kWe. It is operated as a stand-alone system and as load-demand following. In idle mode the system will be in hot stand-by.
The HB is started up and shut down using the burner. The burner provides the necessary heat to preheat the cathode air inlet and reach the operating target temperature. The necessary fuel to heat up the HB-2500 is referred to as make-up gas. Forming gas is used to avoid nickel re-oxidation during start-up and also to protect the stack in case of system failure. The maximum speed for heating up and cooling down is 100 oC/hr. A specific start-up procedure has been developed by HTc in which important aspects are heat-up rate and the temperature gradients over the stacks. At certain temperature levels the system is switched to a next step in the procedure. The operating window of the system has been defined in terms of minimum and maximum values for the various parameters. During operation the system is controlled by only 8 variables.
In total four modified HoTboxes have been manufactured and delivered during the reporting period to HyGear for integration into the DIAMOND A system and testing. The first HB contained SOFC stack dummies allowing the consortium perform testing and tuning of all operating condition without the risk of stack internal damage that leads to the mixing of fuel and air inside the stack. The dummy stacks contained as many parts from the real stack as possible to mimic the thermal mass and dynamic response of the real stack. This ensured a swift transition from dummy stack operation to real stack.
Each HoTbox was integrated with the “cold” Balance of Plant, followed by functionality tests of the components.
For the preliminary experiments and characterisation tests a system control using a PLC with a computer as User Interface was developed. The start-up sequence as described by HTc was programmed to run automatically. A user intervention is possible by changing the air flow to the burner to control the heat produced in the first phase. The control keeps the variables, like heat-up speed, stack temperature gradient and maximum temperatures within the allowed limits.
Pressures, temperatures, flows, current and voltage are measured, recorded and used to calculate the various pressure drops, temperature gradients, fuel and air utilization, and power output.
Essentially, the system layout is such that under normal operating conditions pressures and pressure drops will be within limits at the maximum flows. If a pressure drop or a pressure exceeds the upper limit, the stack power will be reduced and hence the flow. Similarly, temperatures and temperature gradients will be within the allowed domain at standard operation. Any deviation will be indicative of a problem, and appropriate corrective action will be taken.
The testing of SOFC CHP systems at HyGear is mainly aimed at testing a limited number of operating conditions. Therefore the control is not fully automated. When a change in condition is due, the operator has to adjust the values of the mass flow controllers, blowers, and inverter.
The Diamond C system is a mid-scale (10 kW) SOFC system having a conventional lay-out with separate system components built by VTT. The system is started-up by first heating the stack from ambient to operating temperature and then ramping up the load current to a desired value. Heat is provided to the system by inline electric heaters. During the heat-up, air is supplied to the system and the anode is protected from re-oxidation by providing a reducing atmosphere with either
1) standard pre-mixed safety gases or
2) safety-gas free measures.
Current load-up is carried out after a sufficient operating temperature is reached. Constraints regarding stack temperature and/or temperature distribution have to be obeyed during current load-up.
During nominal operation, the system operating domain is confined by the constraints giving the minimum and maximum values for the various parameters.
No upper-level control strategy is (automatically) implemented in the DIAMOND C control system. Instead, all the controller set points are explicitly or implicitly set by the human operator (cf. the look-up-tables during start-up and shutdown).
To enable the communication of an external device (“DIAMOND device”) with the DIAMOND-C power plant and its existing automation and control system, two data transfer interfaces were prepared:
1) One based on the OPC protocol and the system internal tag name mapping
2) One based on the Modbus/TCP protocol and an external data registry mapping
Both OPC and Modbus are industry standard communication protocols for automation and process control purposes and currently, both protocols are available for system development in the DIAMOND project.
The OPC-based communication protocol utilizes the readily existing tag name mapping, which is provided by the OPC server existing as a part of the DIAMOND-C automation system. Therefore no additional data I/O variable-name mapping was required for this purpose. The downside of OPC is that it is not preferred for fast control actions due to the data always traversing through the OPC server.
To enable direct communication of an external DIAMOND device and the DIAMOND-C control unit (“PLC”), a Modbus/TCP protocol interface was defined. As a part of this task 72 new interface variables and 132 Modbus registry definitions were configured in the system and mapped against the key input and output variables in the DIAMOND-C system. The full interface mapping was documented and distributed to the integration partners, especially INEA for usage in the algorithm embedding efforts.
At the start of the project a short heat-up, load-up and shutdown experiment was carried out with the DIAMOND-C system. The system was operated at maximum current of 50 A, whereas its nominal current is 160 A. The reason for this is that the stack was so worn out during the experiments that voltages of individual cells within the stack decreased too much to allow for safe increase of the stack current.
Between ca. 40 to 90 hours the system was let cool down as maintenance had to be carried out on the exhaust heat recovery heat exchanger. At its end-of-life state, the stack produced so much heat that heat recovery at the exhaust had to be improved in order for the exhaust temperature to remain within safe limits. All in all, the system operated reliably and the updated HMI/SCADA as well as the OPC and Modbus interfaces operated as expected.
For further testing a new stack module and new stacks were manufactured for and installed in the DIAMOND-C system. The DIAMOND-C system BoP part had to be modified slightly to fit the new stack module. A new fuel gas preheater was installed between the BoP module and the stack module to guarantee sufficient fuel temperature at the stack inlet.
Due to the mentioned modifications, it was considered necessary to carry out additional nominal operation testing of the DIAMOND-C system.
The control of fuel cell systems at HyGear and VTT is kept simple with much involvement from an operator. The intention of the project is to operate the systems unattended, and in the second phase of the project to provide software and advance control to adapt the control parameters, like operating window, temperatures, fuel and air utilisation and more, to the changed system health.
Diagnostics
The application of suitable diagnostic methodologies coupled with control actions and reasoned operation manoeuvres can improve FC systems lifetime and reduce their maintenance costs. Indeed, diagnosis allows to quickly detect malfunction states in the entire system and to process adaptive control strategies, so as to bring the system towards safer operating conditions. Generally speaking, fault diagnosis is characterized by three main processes: i) detection, ii) isolation and iii) identification. The first process focuses on the detection of a faulty state occurring in the considered system. Then, the location of the fault is singled out through the isolation process, identifying the component(s) under undesired state. Finally, through the identification process, the fault size and its time-varying behaviour are appraised. A fault diagnosis methodology can be classified according to its performance: FD (fault detection), FDI (fault detection and isolation) and FDII (fault detection, isolation and identification).
In the current literature, three kinds of diagnostic methodologies are usually considered, namely model-, knowledge- and signal-based diagnosis. The model-based approach exploits a mathematical model, physical, grey- or black-box, to simulate the system variables in several operating conditions. Knowledge-based diagnosis is, instead, based on human knowledge or qualitative reasoning, whereas the signal-based approach relates to the direct processing of signals measured on the system. It is worth observing that during the detection process, the acquired information is treated to obtain system state indicators, whose nature depends on methodology used. The distinctive features of each methodology might drive the choice from one approach to another. As an example, with respect to both knowledge- and signal-based approaches, a model-based approach does not require the same amount of experimental data and shows a greater generalizability, due to more evident physical content. However, a main drawback consists in the necessity of specific details on the system features, such as geometry, material, etc. Nevertheless, this methodology is recommended when experimental data are not easy to be performed, due to the complexity of the condition to be reached or the constraints on the system operation.
Despite the considered approach, the extraction of significant features, which feed the detection process, is the core activity of the system monitoring process. Once a feature is extracted (e.g. a residual) it is treated to single out the occurrence of a fault. As an example, model-based approach exploits models to simulate monitored system variables. The residuals are generated as a difference between measured and simulated variables. Then, for the purpose of detection, the residual is compared to specific thresholds, so as to generate an analytical symptom. A symptom is a binary variable representing the occurrence of an abnormal state in the system. If a residual exceeds the threshold, a symptom arises (becoming 1), meaning that an unexpected event happened. If the residual lies below the threshold, the related symptom is 0, meaning that the system is behaving normally. Threshold values are generally computed taking into account several aspects, such as system complexity and measurement uncertainty, and considering the trade-off between accuracy, robustness and false alarm and missed faults probabilities. The robustness of the approach can be improved by using adaptive threshold instead of fixed ones. The symptoms generation is only the first step towards isolation of the faulty component(s). They should be related to the specific fault occurring in the system (fault identification). This process can be performed through Fault Tree Analysis (FTA) methodology, whose main outcome is an inferential isolation tool, namely Fault Signature Matrix (FSM). This matrix univocally links the faults to the symptoms detected during the detection process. The comparison of the collected symptoms with those gathered within the FSM, allows the fault isolation. If a perfect match occurs, the faulty component is located. As a final step, the estimation of residuals magnitude and time behaviour provides further significant information (identification process). Indeed, it is essential to characterize the fault time behaviour to take into account the detection and isolation delay, especially for incipient faults.
Suitable tools and methodologies for enhancing real-time monitoring, diagnosis and control of generic SOFC systems, considering both typical fuel cell system assembly (i.e. DIAMOND-C experimental set-up), as well as Integrated stack modules (ISM, i.e. the DIAMOND-A Hot-box developed and instrumented by HTc) have been under development.
This work mainly consisted of four tasks, whose respective evolution and achievements are synthesized hereinafter. The first task focused on an extended literature analysis on SOFC models and diagnostic schemes. Such a state of the art investigation was particularly intended to serve as a basis for the subsequent development, within the dedicated tasks, of models and methods for condition monitoring and on-field diagnosis. Therefore, this study indicated the most suited modelling approaches and methodologies for SOFC systems monitoring and diagnosis. A deep investigation has been also performed on the state of art related to control-oriented models, in particular 1-D, 0-D and black box models of both SOFC and SOFC systems, as one of the main objectives of the work package is to address the necessary counter-measures to be undertaken whenever a manageable fault arises.
On the basis of the indications provided by the above described state of the art analysis, two tasks, which lasted over the entire project length, were carried-out in parallel in such a way as to achieve the target goals in terms of real-world deployability of models and diagnostic algorithms. Particularly, the second task focused on the selection and development of the most suitable diagnostic methodologies for the SOFC systems under-study. The detection and isolation of faulty states within an SOFC system is a challenging issue, particularly due to the complex and non-linear interactions among the system components. According to several Fault Detection and Isolation (FDI) techniques, an abnormal or undesired state can be detected by comparing residuals to suitable threshold levels. The residuals can be generated by relating the measurements of monitored variables, acquired on the system by means of available devices, to numerical values generated (and this is the case for the DIAMOND project, with particular regard to the Diamond-C system) by a mathematical model. Once at least one residual crosses the related threshold level, an abnormal state is detected and a corresponding analytical symptom is activated. To accomplish the correct isolation of the detected state, all the generated symptoms are compared to specific information relating the aforementioned symptoms to a fault. This latter task can be achieved by means of a Fault Signature Matrix (FSM), which univocally correlates a set of analytical symptoms (i.e. monitored variables) to all the most probable faults the system can encounter. The design of the FSM is usually accomplished following a Fault Tree Analysis (FTA). In the DIAMOND project, a generic FSM was developed, holding directly valid for the DIAMOND-C, whereas a revised version shall be adapted to the DIAMOND-A ISM. More in detail, a Fault Signature Matrix (FSM) is a two-dimensional matrix, in whose rows are listed the faults considered in the FTA, whereas its columns list all the symptoms gathered by accounting for each developed FT. Clearly, repeating symptoms are considered only once. In this way, a complete list of all the variables to be monitored is obtained. In its most generic version, the FSM includes five faults: i) air blower, ii) air leakage between air compressor and pre-heater, iii) temperature controller failure, iv) pre-reformer fault and v) stack fault. Each row has been designed following for each fault the previous FTA and several FTs have been depicted. From all the symptoms, fifteen variables have been chosen as the most significant to be monitored. The definition of an FMS only through an FTA approach might lead to a non-optimized isolation tool, since the correlations among symptoms and faults are defined only through qualitative links, not accounting for the quantitative effect of each fault on the deviation of the affected variables from the nominal conditions. According to the FTA approach, a symptom is triggered as soon as a fault occurs. However, it should be considered that the affected variables might experience small deviations from their nominal values, depending on the fault magnitude. Indeed, for incipient faults, some variables may deviate, but still remaining near their operating condition. The use of a model to simulate the effect of several fault on an SOFC system can thus improve an FSM, evaluating the quantitative deviation of each residual and defining the arisen symptoms with respect to a specific fault magnitude and threshold value (assumed for the fault detection).
The need of suitable models, matching the conflicting needs of accuracy and computational burden, provided clear guidelines to the researchers active within the third task, during which comprehensive models were developed for both DIAMOND-C and DIAMOND-A systems. The activity mainly focused on the development of a model structure destined to on-line monitoring use. The monitoring models were thus conceived in such a way as to support diagnosis and control algorithms, which will be implemented on-board. Consistently with those applications, both black-box and grey-box approaches have been considered to maximize accuracy and minimize both computational time and required experimental data for model characterization. As a first stage, models for the conventional DIAMOND-C system were developed. The modelling work for both stacks and system components has been performed considering a modular structure to use common algorithms for similar components and/or phenomena, thus creating a comprehensive models library. This approach was proven to be beneficial, not only to speed up the second part of the modelling work, which concentrated on the more complex layout of the DIAMOND-A, but also to enable the development of resilient model-based diagnostic tools, capable of performing both fault detection and isolation even with a reduction of number of available sensors. Starting from the results obtained by the DIAMOND-C model, subcomponents were arranged in a new model structure, taking into account thermal exchanges between components. Radiative exchange was modelled through the evaluation of view factors depending on geometrical data due to the layout. The resulting model was proven to be highly versatile and allows enabling or not the thermal exchange of each component. It is thus suitable to support offline configuration sizing. The monitoring model is conceived in such a way as to support diagnostic and control algorithms, entirely implemented on-board. Consistently with those applications, a grey-box approach has been taken into account in order to maximize accuracy and minimize both computational time and the amount of experimental data needed for model characterization. It is finally worth remarking how the overall model development phase greatly benefited from the close interactions with experimenters and hardware assemblers working in the dedicated work-packages.
The final task dealt with the development of both signal- and model-based diagnostic algorithm. The former activity focused on the phenomena that occur progressively and continuously, and do not cause instantaneous irreversible damage, but may be fatal in the long term. The goal is identifying and characterizing possible degradation signatures for diagnostic purposes in line with the typical approaches signal-based procedures present in "blind tests". In order to reach this goal, it is necessary to compare measured data with respect to the typical (i.e. expected) behavior of SOFCs. The frequency analysis, also referred to as spectral analysis, is based on estimation procedures applied to spectrum of the signal computed by the Fast Fourier Transform (FFT) algorithm. The goal of spectral estimation is to describe the distribution (over frequency) of the power contained in a signal, based on a finite set of data. Estimation of power spectra is useful in a variety of applications, including the detection of signals buried in wideband noise. Since voltage is measured in different states, the described approach is applied for every data segment where the state is kept constant (by visual inspection), henceforth referred to as a “voltage segment”: the aim is searching characteristic frequencies of the voltage time series (PSD maxima) and how they vary from a segment to another. The signal-based procedure acting in the frequency domain is applied to data set provided by partners. In particular, it is applied to the output voltage signal measured on the Diamond-C system. Voltage PSD shows some peaky behavior in the presence of considerable variation of Fuel Utilization (FU) w.r.t the typical value used for the SOFC system under test. In Diamond-C only small variation of FU are present. This implies that no faulty conditions due to high FU are present in the analysed data set (without excluding other faulty conditions), and, on the other hand, confirms that PSD-based techniques can be useful to isolate high FU utilization faults as remarked during the Design project.
After the design of appropriate residual generators, a fault is inferred based on the time evolution of a relevant subset of residuals. A statistical method for change detection in residuals was developed within the WT4.4. The non-parametric algorithm builds on the quantification of the change in statistical pattern of the corresponding residual. The simulated results show great potential for detecting change in the residuals, and thus facilitating fault detection.
Regarding the model-based FDI algorithm, the algorithm development started from the characterization of a proper mathematical model through available experimental data, describing the system as a whole and the related components. The adopted model of course consisted in the modelling architecture developed within task 3, as described above. An FTA is performed, once the faults of interest are defined after a fault analysis, to design a heuristic FSM. Then, this FSM is enhanced by fault simulations through the system model, enriched with proper fault sub-models. A sensitivity analysis, imposing fault magnitudes and specific threshold levels, led to the definition of such improved FSM. Moreover, the analysis via isolated component sub-models guaranteed the definition of an Adjoint FSM, which can increase residuals redundancy and solve fault clustering problems. This latter step closed the whole diagnostic algorithm development, with a final outcome consisting in the Adjoint FSM tuned on fault magnitude, threshold level and available measurements.
Nevertheless, the system model (and related component sub-models) can play a significant role in control algorithm design, as well as diagnostic algorithm on-board implementation. Indeed, a mathematical reduction procedure is applied to reduce model complexity and to allow its implementation on specific hardware for on-board uses. For example, PLC devices could not handle complex nonlinear differential equations, and a proper finite difference approach should be chosen to retain model accuracy and allow on-board residual computation (for both complete system and related components). Moreover, the implementation of proper look-up tables can be of advantage for highly reducing computational efforts, while keeping an acceptable physical adherence with respect to the modelled phenomena. The above observations were suitably referred to when interacting with the work-package focusing on PLC hardware implementation of the above described lack of sensor resilient fault detection and isolation methodology.
Diagnostic tools
In state-of-the-art fuel cell stack monitoring techniques, the voltage of either single cells or cell-blocks are monitored separately. This complicates system integration and has a negative impact on the cost of the fuel cell system as a large number of voltage channels may have to be monitored. Total Harmonic Distortion (THD) is a promising low cost technique for detecting a critical cell and stack operation status, such as fuel starvation, from the stack sum voltage only. The technique has been developed at different levels: single repeat unit, short-stack and SOFC system.
First, laboratory characterizations were performed on small size cells in order to qualify and quantify the suitable excitation signal characteristics for the THD analysis. In particular a special attention has been paid to determine suitable frequencies and amplitudes of excitation to be applied for high fuel utilization detection. Then, THD with suitable characteristics was implemented at short stack level with the aim to determine if this technique is promising for detecting faults from the stack sum voltage only. Finally, a THD analysis algorithm was implemented and evaluated online on DIAMOND-A SOFC system at VTT.
The following main results were obtained:
- Based on THD measurements at single cell level, it is deduced that the THD index can be used to observe changes in the FU experienced by the SOFC. A special attention has been paid on the identification of the best parameters for the excitation:
Excitation signal frequency. Changes related to cell FU are most visible in the THD index when the excitation signal frequency is between 1 and 0.01 Hz. This is reasonable, since in this test arrangement the non-linear behaviour of the i-V relationship is caused by changes in the SOFC that are related to fuel mass transport, which is a relatively sluggish process.
Excitation signal amplitude. The THD index obtained with a 1% excitation amplitude appears to be corrupted by process and measurement noise and therefore an AC excitation of only 1% of the DC current is not sufficient to obtain adequate data for reliable THD index calculation. With an AC excitation of 5% of the DC load or more, the THD index increases consistently when FU>0.8 (over the relevant frequency range).
- The implementation of THD has been done on a 6-cells short stack from SOLIDpower. It appears that detection of a faulty cell among 6 cells can be done from the stack voltage only. In this case high FU is detected with THD measurements performed at FU=70 (iDC=28.9 A) with an excitation amplitude iAC=0.8 A (i.e. 10 mA/cm², only 2.8% of iDC) during 10 periods and frequency of 0.01 Hz. It will be very sensitive to use the amplitude of the 1st harmonic as a criteria for the detection. Hence, decreasing the number of cell voltage probes in a stack seems to be realistic regarding this kind of fault.
- Finally, online THD implementation has been done on DIAMOND-C SOFC system. The experimental results have demonstrated that THDA can be a reliable means for quantitatively monitoring the FU rate experienced by the stack and that the algorithm is simple to implement on an embedded controller.
In conclusion, testing efforts have been focused on THD analysis applied at different scale levels: single repeat unit, short-stack and SOFC system. First, the best parameters for the THD excitation have been identified on single repeat units. Then, it has been demonstrated that this THA analysis can be relevant for detecting online high fuel utilization occurring in a stack by measuring the sum stack voltage only.
Also work has been done on the development of stack state-of-health and Remaining Useful Life (RUL) prediction. The novel approach proposed in the DIAMOND project overcomes the limitations of the known approaches and facilitates a reliable RUL prediction in non-stationary operating conditions. The internal resistance of the SOFC stack is estimated from available instrument readings. More specifically, we reconstruct the internal aggregated Ohmic stack area specific resistance (ASR) by using an Unscented Kalman filter (UKF). Since this parameter describes an internal resistance of the stack, it is directly associated with efficiency of power conversion. Therefore, the maximal loss of efficiency that defines RUL of the stack can be easily correlated with the terminal value of ASR.
The entire algorithm consists of the three main parts, executed continuously online: (i) estimation of area specific resistance (ASR) of the stack, (ii) prediction of its future progress based on collected data, and (iii) prediction of RUL. In the first step, a UKF employing a nonlinear SOFC stack model estimates the stack ASR, which is independent of the momentary operating conditions. The UKF employs an SOFC model and measured operating conditions (including temperatures voltage, current, and fuel gas flow rate) to estimate ASR. In the second step, by assuming a linear structure of the temporal drift in the estimated ASR, a momentary change rate of the said ASR estimate is identified by utilizing a standard linear Kalman filter. In the third step, given the momentary estimate of the ASR and its momentary change rate at any time instant, a statistical prognosis of the future development of the stack ASR is performed by using Monte Carlo simulation. Hence the algorithm results in an estimate of the current ASR value as well as the RUL at each time instant of stack operation. The methodology was evaluated on two different stacks, showing in both cases accordance between the predicted and real RUL. The algorithm requires some time to be able to provide reliable RUL predictions, but accurate RUL predictions are provided long before the end-of-the-life of the stacks.
To reduce the number of sensors in a system algorithms for soft sensors have been designed; one for determining the minimum and maximum stack temperatures, and a second to estimate the gas composition at the inlet of a stack.
Thermal stress is one of the main factors affecting the degradation rate of SOFC stacks. In order to mitigate the possibility of fatal thermal stress, stack temperatures and the corresponding thermal gradients need to be continuously controlled during operation. An efficient and consistent approach to data-driven design of the estimator for maximum and minimum stack temperatures was proposed intended (i) to be of high precision, (ii) to be simple to implement on conventional platforms like programmable logic controllers, and (iii) to maintain reliability in spite of degradation processes. By careful application of subspace identification, supported by physical arguments, we derive a simple static estimator capable of producing estimates with 3% error irrespective of the evolving stack degradation. The degradation drift is handled without any explicit modelling. The approach is experimentally validated on a 10 kW SOFC system from VTT.
Degradation and poisoning of SOFC stacks are continuously shortening the lifespan of SOFC systems. Poisoning mechanisms, such as carbon deposition, form a coating layer, hence rapidly decreasing the efficiency of the fuel cells. Gas composition of inlet gases is known to have great impact on the rate of coke formation. Therefore, monitoring of these variables can be of great benefit for overall management of SOFCs. We proposed three distinct approaches for the design of gas composition estimators of an SOFC system in anode off-gas recycle configuration which are (i) accurate, and (ii) easy to implement on a programmable logic controller. Firstly, a classical approach was briefly revisited and problems related to implementation complexity was discussed. Secondly, the model is simplified and adapted for easy implementation. Further, an alternative data-driven approach for gas composition estimation is developed. Finally, a hybrid estimator employing experimental data and 1st principles is proposed. Despite the structural simplicity of the estimators, the experimental validation shows a high precision for all of the approaches. Experimental validation is performed on a 10 kW SOFC system from VTT.
Control
As a first step in the development of advanced control for the DIAMOND systems the control concepts and architectures for μCHP SOFC systems recently adopted in practice have been reviewed. Control system manages and regulates the behaviour of the system. It monitors whether or not system parameters, like temperature or pressure stay within given boundaries. In case of exceeding either lower or upper limit the system should respond in a way that ensures further safe operation. Usually several options exist to do this depending on the control philosophy taken by a particular approach. A primary function of the control is to regulate supply and demand of system output power and heat. For a heat demand driven CHP system an increased demand in heat should lead to an increase in heat produced. This can be achieved by either increasing the stack power and thus fuel supply when operating at a constant utilisation or by solely increasing the fuel supply while keeping the stack power constant,and thus reducing the fuel utilisation. This increased excess of fuel is then burned out in the after-burner producing extra heat. In the former option the extra produced power will be supplied to the grid.
In a power demand driven system an increase in power will lead to an increase in stack power and thus current. When operating at a constant flow the fuel utilisation will increase, and the heat output will decrease; when operating at a constant utilisation, the fuel flow will increase and also the heat output.
One of the main challenges in commercializing the SOFC power systems is how to achieve high electrical efficiency without increasing the degradation rate of fuel cells. The rationale of the work is to design a hierarchical control structure that will ensure fast load tracking and efficient operation of the SOFC systems while taking care of a range of imposed operational constraints and state of health. Such control includes soft sensors algorithms for relevant immeasurable states and disturbances (e.g. maximum stack temperature, fuel concentration, degradation rate of the SOFCs, etc).
For good operation of a SOFC power system well designed low-level controllers are indispensable. The purpose of low-level controllers is to i) supply enough fuel into the system in order to achieve the required stack current and prevent fuel starvation in cells, ii) maintain stack temperature and the corresponding thermal gradients at all times in order to reduce stack degradation due to thermal stress and iii) supply enough water in the system which is needed.
Low-level controllers for SOFC power system were designed by using the feedforward-feedback approach. The feedforward controllers are based on stoichiometry of electro-oxidation, reforming and combustion reactions, which allow immediate response to variable current demand. The feedback controllers perform additional fine adjustment of fuel and air supply in order to minimize the undesired system temperatures variations.
Controllers were implemented and verified on two models of the SOFC power system:
• The detailed physical model of the SOFC Hotbox power (DIAMOND-A) system in gPROMS environment. The model was received from the SOLIDpower and was previously developed by EPFL-LENI.
• The model of the DIAMOND-C power system in Matlab/Simulink environment developed by UNISA within the DIAMOND project.
The proposed low-level control scheme for the DIAMOND-A system consists of four feedforward controllers and three PI control loops and is rather simple for practical implementation.
The selection of pairings of manipulated and controlled variables for control is based on physical knowledge of the system. Input/output pairing for single-loop feedback control is assessed by the relative gain analysis.
Parameters of the PI controllers were tuned from the open-loop step response experiments performed with the DIAMOND-A model at the given operating point. The so-called Magnitude Optimum Multiple Integration (MOMI) method was applied which provides a relatively fast and non-oscillatory closed-loop response for a large class of processes. The method calculates parameters by integrating the process open-loop response after applying the step-change to the process input.
Performance of the proposed control scheme was verified by applying the standard daily load profile of residential houses and large step load changes. The simulations show that the proposed controllers maintain controlled temperatures within defined ranges and provide system electrical efficiency of about 43 %. The lowest efficiency is obtained at load on the lower bound due to the fuel that has to be feed into the burner to maintain the burner temperature. The maximum stack temperature was almost constant, whereas the minimum stack temperature exhibits greater variations. The latter might be a problem during the high load periods when the stack temperature difference is the highest. This can lead to degradation of the electrochemical performance, or in the worst case, malfunction of the system.
Low-level control scheme for DIAMOND C power system consists of four feedforward controllers and seven PI controllers. The control scheme has been tested on the daily load profile and on step-wise changes of the stack current. Simulation results indicate that the strategy successfully controls the main stack variables while taking into account some of the system limitations and delivers high system electrical efficiency (above 65 %). Due to its simplicity, the control appears appropriate for commercial applications.
Running SOFC power systems at the edge of efficiency and maximal longevity is important for their successful market deployment. To properly handle variable load conditions and inevitable degradation processes, on-line optimization is needed to adjust the process variables to run the process at the operational optimum. We proposed a two-level control system in which the low-level control for the DIAMOND-A system is upgraded with a supervisory optimizer. The optimizer adjusts references for the low-level controllers to maximize the electrical efficiency of the system while satisfying a number of constraints. The optimization problem is solved by using the extremum-seeking approach where optimum is sought directly on the process.
A steady-state efficiency analysis was performed at a nominal operating point of a model with different combinations of available optimization variables. Based on these results the references of burner lambda and outlet air temperature were selected as the optimization variables (arguments) of the supervisory optimizer.
The selected optimization problem targets maximization of efficiency and minimization of stack temperature violations. Efficient operation of the optimizer is governed by properly tuned parameters of the optimization algorithm.
A two-level control system with a supervisory controller is verified on a DIAMOND-A model on a range of operational scenarios. The presented results were obtained by simulating the standard load profile of the residential houses. The references of the control scheme without optimizer were fixed at constant values.
Simulations show, that the optimizer improves electrical efficiency of the system by about 8% compared to the efficiency of the system controlled by the low-level controllers. It also keeps the system temperatures within the safe ranges. Due to its simplicity, the optimizer appears appropriate for practical applications.
Similarly, supervisory controller for the DIAMOND-C system is developed and verified on a model of a system in Matlab/Simulink environment. The control system in this case consists of interconnected, independent modules such as: low-level control, soft sensors, condition monitoring module and supervisory control. The purpose of the low-level controllers is to keep the system running at corresponding operating conditions, defined by the controller set-points. The developed soft sensors provide information of non-measurable process variables to the remaining three modules. Namely, the reconstructed temperatures can be used as controlled variables, while the condition monitoring module employs the estimations of gas flow components at the anode inlet. The condition monitoring module provides information on the instantaneous degradation rate, i.e. a measure of stack performance deterioration. Finally, this information is fed to the supervisory controller, which calculates the references for the low-level controller by employing the information about performance degradation rate and the overall efficiency of the system.
The references of the low-level controllers were in this case tightly connected with each other. The reference of the stack outlet air temperature completely defines operating point of the system.
The optimisation problem was formulated to seek an optimal operating point of the SOFC system so that either (i) degradation rate is minimal, (ii) system efficiency is maximal, or (iii) the compromise between the two is achieved.
The optimization problem is solved on the basis of well-known extremum seeking control.
An analysis of the performance criteria on the system level with implemented low-level controller was performed to understand the behaviour of the supervisory controller. To this aim, the individual contributions to the criterion function are evaluated for the only remaining manipulative variable that defines the operating point of the system, cathode outlet temperature. Two facts are apparent from the analysis: (i) system efficiency first increases and then decreases with the increasing temperature of the stack outlet temperature and (ii) the degradation rate drops with the increasing temperature.
The proposed two-level control system for the DIAMOND-C system was verified under different simulation scenarios. Simulation scenarios show that the supervisory controller reaches the optimal operating point which corresponds to the end user optimization objective. Even though the oscillating load current significantly hinders the performance of the supervisory controller, the optimum is still reached. The simulation validation of the proposed control system shows promising preliminary results. Although drawing unambiguous conclusions is unreasonable, one major conclusion is indisputable: “The developed control system has major impact on the degradation rate, and, in turn, to the total life-span system".
For successful integration of the developed low-level and supervisory control software and diagnostics software into the existing CHP system the existing level of instrumentation and control system was reviewed. The detailed system analysis was taken to evaluate a list of signals, measurement devices and actuators used on the system. Within analysis the set point parameters, exchange process values and triggers to influence the system operation were defined.
Where needed the adaptation measures of the system software to be operated in parallel to the prediction and monitoring system were considered. After system examination the standard industrial PLC system was chosen and delivered to HyGear by INEA.
On that platform the development of advanced control algorithms and diagnostics methods were prepared. The advanced control was set up as higher level control (outer control loop) to existing control system. The control and diagnostics was realized as a PLC program library.
The original control software was adopted, to enable it poll for new set points, receive trigger signals and send process variables to the control and diagnostics algorithms.
Advanced control system was applied both to fuel cell and accompanying BoP systems.
The implementation of the developed low-level and supervisory control software and diagnostics software into DIAMOND C was for a large part identical to the procedures followed for DIAMOND A.
As the DIAMOND-C system was already highly instrumented, including a full PLC-based control and automation system, the focus was on the implementation of the advanced control and diagnosis algorithms into a suitable software and computational environment. This was provided with an additional PLC system which communicates with the PLC system through a standard Modbus protocol. A significant work force was spared by not completely changing the current programmable logic controller (PLC) equipment of the VTT system.
Tight teamwork of all participants were needed to complete integration part of control and diagnostics on Diamond-A and Diamond-C. On both systems teams from JSI and INEA work remotely connected to the Diamond systems at HyGear and VTT. On site personnel helped with all necessary adaptations and operator actions on the actual SOFC systems. During both tasks continuous data logging were provided to run off-line analysis for tuning controllers and executing diagnostics algorithms.
The algorithms that were developed in DIAMOND and implemented in industry standard process control devices and subsequently integrated into SOFC systems have been evaluated on both systems.
Extensive testing efforts took place with both systems. In particular, cold testing, hot dummy-testing (system w/o SOFC stack), nominal characterization tests of both systems and finally complete system level algorithm evaluation tests. In addition, existing measurement data from the systems’ operation was utilized for algorithm development before fresh data became available.
Several algorithms for general monitoring and estimation as well as control had been developed before the DIAMOND project. Also many of these were adopted for the SOFC in either a simulated environment or in experimental laboratory arrangements. Also many of the DIAMOND partners (including HYG, UNISA, VTT, CEA, HTc) have participated in this development effort in preceding European projects such as DeSign and GENIUS. The final implementation and validation phase of algorithms in an industrially relevant setting has, however, not been reached in the existing works to the extent as was done in the DIAMOND project.
Several separate system level test runs were prepared and executed during the project. These are described below per test system.
DIAMOND-A:
1) Prototype operational testing; short test run with one stack box dedicated for the testing purposes
2) System characterization; system refurbished with new stack box, step changes in current, I-V curves, and long term operation at various conditions >6000 hours
3) Algorithm evaluation test run; system refurbished with new stack box, tuning control and parameterization of algorithms, ca. 1000 hours
DIAMOND-C:
1) System operability test; short test run with old, de-functional stacks to check balance-of-plant and instrumentation operability after downtime period
2) Hot dummy testing; system operability testing with dummy-stack after stack module modifications related to stack refurbishment, ca. 300 hours
3) System characterization; system refurbished with new stacks and stack module, operational testing and characterization of system in various operating conditions, ca. 400 hours
4) Cold testing with modified system for embedded THD algorithm development, ca. 150 hours
5) Algorithm evaluation; tuning of controllers, parameterization of algorithms and evaluation of algorithm output under various conditions, ca. 700 hours
In addition, several component tests were carried out to facilitate system refurbishment and modification.
The testing effort was a genuine co-operation of all project partners and realized a thorough pre-commercial evaluation of the developed and implemented algorithms in a 100% industrially relevant setup. The testing work carried out in the project revealed several aspects of SOFC system monitoring, diagnostic and control algorithm development work, which have to be taken into account in similar future projects. At the same time, the status of commercially relevant monitoring, diagnostic and control algorithms for SOFC systems was pushed forward significantly.
In summary the following main results were obtained in the testing work, all in an industrially/commercially relevant system level setting:
a) The validation of the functionality and effectiveness of an advanced SOFC system control algorithm, and the applied development procedures
b) The validation of the functionality of novel SOFC stack state-of-health estimation and stack lifetime prognosis algorithm
Subsequently, the validation of the functionality a supervisory controller algorithm based on estimated SOFC stack remaining useful life and efficiency
c) The validation of the functionality and effectiveness of such diagnostically relevant algorithms as (i) the SOFC stack THD index calculation algorithm for realized fuel utilization monitoring and (ii) the online fuel gas composition estimation
d) The validation of correct and consistent operation of a model-based SOFC system diagnostic algorithm.
In conclusion a significant collaborative system level testing effort was carried out as the co-operation of all the DIAMOND project partners. Two separate SOFC power systems were used in the testing and multiple algorithms for the monitoring, control and diagnostics of SOFC systems were validated in a commercially-relevant context as result of the tests.
Activities have been carried out to promote and coordinate dissemination and exploitation activities according to the DoW and to the decision taken by the Project Bodies. The main instruments for the implementation of these actions were the website (http://www.diamond-sofc-project.eu) and the newsletters. A first joint workshop with the project ENDURANCE was organized in the first part of the project (with DIAMOND partners giving seven presentations), whereas a second workshop was jointly organized with the HEALTH-CODE project.
In the first part of the project the main actions of focused on the building the website and setting-up the newsletters editorial tool. Attention to the format was given to have a coherent graphical and communication strategy. The Coordinator selected the IT platforms for the implementation of the website and the newsletters; the former was hosted by HyGear and administered with WordPress open source software, whereas the newsletters was managed on the MailChimp platform. The editing work of website and newsletter (content, graphics and structure) was performed by UNISA under the responsibility of the Dissemination Manager. In the second part of the project the main actions of were the update of publishable information on the website and the sending of the newsletters. In the first period, the DIAMOND website registered 3653 visits; in the second period the visits were 3250, with the following percentage in terms of visiting areas: United States 24.93%, Italy 9.37%, United Kingdom 4.27%, Russia 3.82%, China 3.80%, etc.
As basic communication strategy, each newsletter summarized the most recent activities, which are detailed on the relevant website sections (http://www.diamond-sofc-project.eu/about/newsletter/). Several documents and deliverables concerning modelling, diagnostic and control activities have been made available on the website. Seven newsletters have been circulated, aiming at updating the industry and research communities active in the field of modelling, monitoring, diagnostics and control of SOFC. The newsletters mailing list has been populated carefully by including relevant recipients belonging to FC and Hydrogen industry, research, JU and other stakeholders. The initial database had 156 email addresses, which growth up to 329 contacts by the end of the project.
The results of DIAMOND have been presented at the following events:
• 8th Bruges Workshop on Progress in Fuel Cell Systems, June 2-3, 2015 (Bruges, Belgium);
• ENDURANCE Joint Workshop “SOFC Diagnostic” September 14, 2015 (Genoa, Italy);
• FCH-JU Programme Review 2015, November 17-19, 2015 (Bruxelles, Belgium);
• FCH-JU Programme Review 2016, November 21-22, 2016 (Bruxelles, Belgium);
• ENDURANCE & SOPHIA Workshop “Degradation on SOFC”, February 17, 2017 (Barcelona, Spain).
Twelve papers were written and presented at six International Conferences:
• Feedforward-Feedback Control of a SOFC Power System: A Simulation Study, ECS Conference on Electrochemical Energy Conversion & Storage with SOFC-XIV, July 26-31, 2015;
• Soft sensor design for estimation of a SOFC stack temperatures and oxygen-to-carbon ratio. ECS Conference on Electrochemical Energy Conversion & Storage with SOFC-XIV, July 26-31, 2015;
• A Lumped Dynamic Modelling Approach for Model-Based Control and Diagnosis of Solid Oxide Fuel Cell System with Anode Off-Gas Recycling, ECS Conference on Electrochemical Energy Conversion & Storage with SOFC-XIV, July 26-31, 2015;
• Improved fault isolability for solid oxide fuel cell diagnosis through sub-system analysis, 8th International Conference on Applied Energy (ICAE), October 8-11, 2016, Beijing, China;
• Development of a dynamic model for diagnosis and control of an integrated stack module based on solid oxide fuel cells. 8th International Conference on Applied Energy (ICAE), October 8-11, 2016, Beijing, China;
• Estimating the level of degradation in fuel cell stacks, Proceedings of the 25th International Electrotechnical and Computer Science Conference ERK 2016, Portoroz, Slovenia, September 19-21, 2016;
• Control design for a 2.5 kW solid oxide fuel cell power system, Proceedings of the 25th International Electrotechnical and Computer Science Conference ERK 2016, Portoroz, Slovenia, September 19-21, 2016;
• Accounting for modelling errors in model-based diagnosis by using Gaussian process models, Proceedings of the 3rd International Conference on Control and Fault-Tolerant Systems, SysTOL 2016, Barcelona, September 7-9, 2016, pp. 510-515.
• Supervisory control of a 2.5 kW SOFC power system. Proceedings of the 10th Professional Conference on Automation in Industry and Economy, AIG’ 2017, Maribor, Slovenia, April 6-7, 2017, pp. 159-166.
• The new concept of the solid-oxide fuel cells stack control based on device status. Proceedings of the 10th Professional Conference on Automation in Industry and Economy, AIG’ 2017, Maribor, Slovenia, April 6-7, 2017, pp. 151-158.
• Hybrid approach to remaining useful life prediction of solid oxide fuel cell stack. 15th International Symposium on Solid Oxide Fuel Cells SOFC-XV, July 23-28, 2017, Hollywood, Florida, USA. ECS transactions, vol. 78, no. 1, 2251-2264.
• Improving operation of a 2.5kW SOFC power system with supervisory control. 15th International Symposium on Solid Oxide Fuel Cells SOFC-XV, July 23-28, 2017, Hollywood, Florida, USA. ECS Transactions, vol. 78, no., pp 265-274.
Five papers have been published in four international peer-reviewed journals:
• A comprehensive 3-D modeling of a single planar solid oxide fuel cell, International Journal of Hydrogen Energy, vol. 41, n. 5: 3613-3627;
• Online estimation of internal stack temperatures in solid oxide fuel cell power generating units, Journal of Power Sources vol. 336: 251-260;
• Online gas composition estimation in solid oxide fuel cell systems with anode off-gas recycle configuration, Journal of Power Sources, vol. 336:251-260;
• A model-based diagnostic technique to enhance fault isolability in Solid Oxide Fuel Cell systems, Applied Energy, vol 204:1198-1214;
• State of health estimation and remaining useful life prediction of solid oxide fuel cell stack. Energy Conversion and Management, 2017, vol. 148, pp. 993-1002.
One journal article manuscript is close to finalization:
• Online total harmonic distortion analysis for solid oxide fuel cell stack monitoring in a system application, 2017
A poster was presented at the 9th IFAC Symposium SAFEPROCESS that took place in Paris on September 3, 2015. The poster was at display in the special session “Industrial day”, dedicated to dissemination of the latest cases of industry-academia cooperation. The contribution of the consortium team was Diagnosis-aided control for SOFC power systems (DIAMOND), 9th IFAC Symposium SAFEPROCESS, Paris 2-4 September 2015. Posters accepted for presentation at the 7th European Fuel Cell Technology & Applications Conference – Piero Lunghi Conference EFC2017, 12th -15th December 2017, Naples (ITALY) are: The model-based diagnostic algorithm implementation and testing; The fault simulation study performed with the DIAMOND-A monitoring model.
Moreover, IJS partner organized three seminars at the Department of Systems and Control, Jozef Stefan Institute, Ljubljana, Slovenia and invited UNISA representative at two-days lecture for the Slovenian Simulation Society.
Two Workshops have been planned to exchange the latest results and developments of the project: Mid-term exploitation workshop (December 16th, 2015) and Final and exploitation workshop (July 4th, 2017). A joint workshop with the project ENDURANCE has been held on 14th September 2015 and a joint workshop with ENDURANCE & SOPHIA projects has been held on 17th February 2017. Moreover, a final joint workshop was held on 4th July 2017 in Lucerne with HEALTH-CODE project
Mid-term exploitation Workshop – Workshop on Monitoring, Diagnostics and Control for SOFC systems. Improving SOFC-based CHP performance through innovative diagnosis and control focused on the most recent advances and the current research on monitoring, diagnostics and control for SOFC systems. The objective of the workshop was to present the DIAMOND approach and the recent project results for CHP systems equipped with both conventional stacks and integrated modules. The workshop was hosted by 2015 European Fuel Cell Conference & Exhibition at Hotel Royal Continental, Naples (I).
Joint workshop on SOFC diagnostic Diamond & Endurance FCH-JU Projects focused on the activities of both DIAMOND and ENDURANCE projects whose objectives converge towards the improvement of SOFC performance and durability. The workshop was hosted by University of Genova (I). Joint workshop Degradation mechanisms in Solid Oxide cells and systems Endurance & SOPHIA FCH-JU Projects aimed to provide a platform for professional discussions and exchange of expertise in one of the most challenging issues in the commercialization of SOFC: degradation and durability.
Final Dissemination and Exploitation Workshop – One Day Workshop on Monitoring, Diagnostics and Control for FC focused on the implementation and use of the technology beyond the project duration and summarised the progress towards the exploitation by industrial partners and potential customers.
The results achieved for DIAMOND-C and DIAMOND-A SOFC-based systems together with PEM-based µ-CHP and Backup systems were presented in a unique and comprehensive framework. Presentations dealt with methodologies and applications for both SOFC and PEMFC. Control and diagnostic techniques were reviewed along with experimental and modelling techniques suited for their design and implementation on real FC systems. A wide range of Model-based and EIS-based approaches were described and together with their exploitation for fault detection in stack and BOP. A special contribution was given by Y. Tsur (TECHNION): “Impedance spectroscopy analysis using genetic programming (ISGP): a short tutorial”. The workshop was jointly organized by FP7 & H2020 FCH-JU funded projects DIAMOND and HEALTH-CODE at European Fuel Cell Forum – KKL Lucerne (CH). More than 45 attendants from 16 countries were hosted free of charge. Lunch, coffee and beverages were offered to all participants.
Concerning the exploitation strategy, the industrial roadmap on behalf of all the relevant technologies of diagnostics and advanced SOFC CHP system control was summarised in both deliverables D8.5 and D8.6. The possible exploitable results will be the developed diagnostic tools, the advanced control and the system component models and the methodology to develop them. At the various levels the possible markets are Stack manufacturers, System integrators, Automation industry and Boiler companies. The general approach of exploitation of results is directly linked to the interests of the partners involved in DIAMOND. Both, technical and strategic exploitation can be applied to internal and external purposes of the partners. The internal usage will definitely lead to improvements of the demonstrated technology but also to future research on this topic. The external exploitation requires more effort on this task. Here, not only the involved partners are affected. The external exploitation has to address public and commercial stakeholders and the end-users of the demonstrated technology. The specific details for the different stakeholders need to include data on reliable operation of the μ-CHP system using the diagnosis and advanced control.
Potential Impact:
Potential impact
The impact from the R&D done within the DIAMOND project can be divided in various categories:
• Technical impact
• Commercial
• International
• Educational
• Integration in Europe
• Environmental
Technical impacts
Within the project monitoring, diagnostic and control techniques have been developed to improve performance and durability of SOFC systems. These developments have been tested on two different SOFC systems: a conventional one, DIAMOND-C and an advanced system based on an integrated HoTbox, DIAMOND-A.
In more detail the following technical improvements were reached:
• System modes were developed and validated;
• Fault Detection and Isolation (FDI) methodology using fault signature matrices based on fault-tree analyses has been developed and validated;
• A signal-based diagnosis methodology was developed;
• The applicability of Total Harmonic Distortion as diagnostic tool has been verified on stack level;
• Advanced low-level control has been developed, implemented and tested;
• Supervisory control has been developed and tested;
• Soft sensors have been developed;
• A Remaining Useful Life (RUL) estimator has bas been developed and validated.
Commercial impacts
The project was set up to develop and validate advanced control, monitoring and diagnostic tools for SOFC CHP-systems. On system level commercial impacts will arise only after several years. Additional development and cost reductions are needed. The developed methodologies will be used in future systems
HyGear produces hydrogen generation systems but is expanding its portfolio also to fuel cell system which consume hydrogen. The DIAMOND project has provided HyGear with the knowledge to broaden its technology base for in future years.
HTc is a major developer and supplier of SOFC systems. It is constantly improving its cells, stacks, and systems. The results obtained in the DIAMOND project will assist HTc in this effort.
INEA is a supplier of embedded systems for various applications. The outcome of the DIAMOND project will strengthen their visibility in the fuel cell market.
International Impacts
The outcome of the project will help the partners within the project to compete on the international stage with SOFC systems and its components. Performance, durability and reliability, all addressed within the project, are important factors for success in the market.
Educational Impacts
UNISA is an important institute for education, training and PhD students in the field of system modelling, diagnostics and fuel cell and electrolyser research. The research institutes VTT, CEA, and IJS all involve master and PhD students in their research. During this project 2 PhD students have worked on the project (IJS, INEA)
Impact on Europe
The project has strengthened the European position in the field of SOFC-systems development of academia, research institutes and SMEs/industry. Much knowledge has been gained on diagnostic tools, fault detection and identification methodology, system and system component modelling, and control.
The RTD performers gained knowledge and experience for future research and support of the industry. The SMEs will introduce the gained knowledge into their products. Additionally the knowledge gained can also be applied to other fuel cell technologies, but also to electrolysis systems.
Environmental and Economic Impacts
The technology developed can be used to improve future SOFC systems. Widespread adoption of SOFC technologies will give the environmental benefits of reduced fossil fuel usage (via increased efficiency) and reduced emissions.
Dissemination activities
The results of DIAMOND have been presented at the following events:
• 8th Bruges Workshop on Progress in Fuel Cell Systems, June 2-3, 2015 (Bruges, Belgium);
• ENDURANCE Joint Workshop “SOFC Diagnostic” September 14, 2015 (Genoa, Italy);
• FCH-JU Programme Review 2015, November 17-19, 2015 (Bruxelles, Belgium);
• FCH-JU Programme Review 2016, November 21-22, 2016 (Bruxelles, Belgium);
• ENDURANCE & SOPHIA Workshop “Degradation on SOFC”, February 17, 2017 (Barcelona, Spain).
Twelve papers were written and presented at six International Conferences:
• Feedforward-Feedback Control of a SOFC Power System: A Simulation Study, ECS Conference on Electrochemical Energy Conversion & Storage with SOFC-XIV, July 26-31, 2015;
• Soft sensor design for estimation of a SOFC stack temperatures and oxygen-to-carbon ratio. ECS Conference on Electrochemical Energy Conversion & Storage with SOFC-XIV, July 26-31, 2015;
• A Lumped Dynamic Modelling Approach for Model-Based Control and Diagnosis of Solid Oxide Fuel Cell System with Anode Off-Gas Recycling, ECS Conference on Electrochemical Energy Conversion & Storage with SOFC-XIV, July 26-31, 2015;
• Improved fault isolability for solid oxide fuel cell diagnosis through sub-system analysis, 8th International Conference on Applied Energy (ICAE), October 8-11, 2016, Beijing, China;
• Development of a dynamic model for diagnosis and control of an integrated stack module based on solid oxide fuel cells. 8th International Conference on Applied Energy (ICAE), October 8-11, 2016, Beijing, China;
• Estimating the level of degradation in fuel cell stacks, Proceedings of the 25th International Electrotechnical and Computer Science Conference ERK 2016, Portoroz, Slovenia, September 19-21, 2016;
• Control design for a 2.5 kW solid oxide fuel cell power system, Proceedings of the 25th International Electrotechnical and Computer Science Conference ERK 2016, Portoroz, Slovenia, September 19-21, 2016;
• Accounting for modelling errors in model-based diagnosis by using Gaussian process models, Proceedings of the 3rd International Conference on Control and Fault-Tolerant Systems, SysTOL 2016, Barcelona, September 7-9, 2016, pp. 510-515.
• Supervisory control of a 2.5 kW SOFC power system. Proceedings of the 10th Professional Conference on Automation in Industry and Economy, AIG’ 2017, Maribor, Slovenia, April 6-7, 2017, pp. 159-166.
• The new concept of the solid-oxide fuel cells stack control based on device status. Proceedings of the 10th Professional Conference on Automation in Industry and Economy, AIG’ 2017, Maribor, Slovenia, April 6-7, 2017, pp. 151-158.
• Hybrid approach to remaining useful life prediction of solid oxide fuel cell stack. 15th International Symposium on Solid Oxide Fuel Cells SOFC-XV, July 23-28, 2017, Hollywood, Florida, USA. ECS transactions, vol. 78, no. 1, 2251-2264.
• Improving operation of a 2.5kW SOFC power system with supervisory control. 15th International Symposium on Solid Oxide Fuel Cells SOFC-XV, July 23-28, 2017, Hollywood, Florida, USA. ECS Transactions, vol. 78, no., pp 265-274.
Five papers have been published in four international peer-reviewed journals:
• A comprehensive 3-D modeling of a single planar solid oxide fuel cell, International Journal of Hydrogen Energy, vol. 41, n. 5: 3613-3627;
• Online estimation of internal stack temperatures in solid oxide fuel cell power generating units, Journal of Power Sources vol. 336: 251-260;
• Online gas composition estimation in solid oxide fuel cell systems with anode off-gas recycle configuration, Journal of Power Sources, vol. 336:251-260;
• A model-based diagnostic technique to enhance fault isolability in Solid Oxide Fuel Cell systems, Applied Energy, vol 204:1198-1214;
• State of health estimation and remaining useful life prediction of solid oxide fuel cell stack. Energy Conversion and Management, 2017, vol. 148, pp. 993-1002.
One journal article manuscript is close to finalization:
• Online total harmonic distortion analysis for solid oxide fuel cell stack monitoring in a system application, 2017
A poster was presented at the 9th IFAC Symposium SAFEPROCESS that took place in Paris on September 3, 2015. The poster was at display in the special session “Industrial day”, dedicated to dissemination of the latest cases of industry-academia cooperation. The contribution of the consortium team was Diagnosis-aided control for SOFC power systems (DIAMOND), 9th IFAC Symposium SAFEPROCESS, Paris 2-4 September 2015. Posters accepted for presentation at the 7th European Fuel Cell Technology & Applications Conference – Piero Lunghi Conference EFC2017, 12th -15th December 2017, Naples (ITALY) are: The model-based diagnostic algorithm implementation and testing; The fault simulation study performed with the DIAMOND-A monitoring model.
Moreover, IJS partner organized three seminars at the Department of Systems and Control, Jozef Stefan Institute, Ljubljana, Slovenia and invited UNISA representative at two-days lecture for the Slovenian Simulation Society.
In the framework of the task 8.3 two Workshops have been planned to exchange the latest results and developments of the project: Mid-term exploitation workshop (December 16th, 2015) and Final and exploitation workshop (July 4th, 2017). A joint workshop with the project ENDURANCE has been held on 14th September 2015 and a joint workshop with ENDURANCE & SOPHIA projects has been held on 17th February 2017. Moreover, a final joint workshop was held on 4th July 2017 in Lucerne with HEALTH-CODE project
Mid-term exploitation Workshop – Workshop on Monitoring, Diagnostics and Control for SOFC systems. Improving SOFC-based CHP performance through innovative diagnosis and control focused on the most recent advances and the current research on monitoring, diagnostics and control for SOFC systems. The objective of the workshop was to present the DIAMOND approach and the recent project results for CHP systems equipped with both conventional stacks and integrated modules. The workshop was hosted by 2015 European Fuel Cell Conference & Exhibition at Hotel Royal Continental, Naples (I).
Joint workshop on SOFC diagnostic Diamond & Endurance FCH-JU Projects focused on the activities of both DIAMOND and ENDURANCE projects whose objectives converge towards the improvement of SOFC performance and durability. The workshop was hosted by University of Genova (I). Joint workshop Degradation mechanisms in Solid Oxide cells and systems Endurance & SOPHIA FCH-JU Projects aimed to provide a platform for professional discussions and exchange of expertise in one of the most challenging issues in the commercialization of SOFC: degradation and durability.
Final Dissemination and Exploitation Workshop – One Day Workshop on Monitoring, Diagnostics and Control for FC focused on the implementation and use of the technology beyond the project duration and summarised the progress towards the exploitation by industrial partners and potential customers.
The results achieved for DIAMOND-C and DIAMOND-A SOFC-based systems together with PEM-based µ-CHP and Backup systems were presented in a unique and comprehensive framework. Presentations dealt with methodologies and applications for both SOFC and PEMFC. Control and diagnostic techniques were reviewed along with experimental and modelling techniques suited for their design and implementation on real FC systems. A wide range of Model-based and EIS-based approaches were described and together with their exploitation for fault detection in stack and BOP. A special contribution was given by Y. Tsur (TECHNION): “Impedance spectroscopy analysis using genetic programming (ISGP): a short tutorial”. The workshop was jointly organized by FP7 & H2020 FCH-JU funded projects DIAMOND and HEALTH-CODE at European Fuel Cell Forum – KKL Lucerne (CH). More than 45 attendants from 16 countries were hosted free of charge. Lunch, coffee and beverages were offered to all participants.
Exploitation of results
The DIAMOND consortium can be divided into two partner groups, Industry and RTD-performers. Each group will identify the best type of exploitation action and the specific audience of the exploitation activities. Some of the exploitation activities will have direct impact on the technical improvements within a short time frame, while others will lead to future impact for the end users and technology manufacturers.
The industrial partners within DIAMOND will focus their exploitation activities on improving their current technology and business position in existing markets and on the creation of new markets beyond the markets addressed in DIAMOND. The industrial partners will use the technical improvements in a direct manner to shorten turn-around times to product readiness.
The exploitation goals of the RTD performers are different, yet complementary to those of the industrial partners. The technical improvements and developments will be integrated into the teaching courses but will also flow into the research activities of the RTD performers. For instance CEA’s business model consists in developing technologies for the industry and bringing them to market thanks to technology transfer to industry, either large industries, SME’s or start-ups. VTT has a similar approach.
This approach will lead to being more attractive for Ph.D. master and graduate level students to join the RTD performers. More strategically, graduates from academic partners will be trained for working in the related industry but also in related research work. The long-term pay-off of the RTD performers will lift DAMOND’s technology being general part of teaching lessons.
As an example the methodology developed by UNISA in the frame of DIAMOND for modelling systems is very versatile as has been shown. It can be easily applied to other system architectures.
The exploitable results are listed below in groups, each group followed by short plans for exploitation.
Exploitable result: Diagnostics
1 Total Harmonic Distortion
2 Failure detection and Isolation methodology
3 Soft Sensors
4 SOFC system modelling
Exploitation plans
1 The developed method will provide CEA a good international visibility and will allow CEA to participate in future projects for further development, either national or international, collaborative or bilateral with industries.
2 The methodology is versatile and can be applied in other Fuel Cell systems. It will provide UNISA a good international visibility and will allow UNISA to participate in future projects for further development, either national or international, collaborative or bilateral with industries.
3 The developed methodology will be exploited in future projects for further development, either national or international, collaborative or bilateral with industries
4 All models developed in the frame of DIAMOND by UNISA will be used for several forthcoming studies and projects dedicated to fuel cells.
Exploitable result: Controls
1 Embedded control
2 Supervisory control
Exploitation plans
INEA and IJS will exploit the developed control methodology in future projects in the field of fuel cell systems.
List of Websites:
http://www.diamond-sofc-project.eu/