Skip to main content

System Automation of PEMFCs with Prognostics and Health management for Improved Reliability and Economy

Final Report Summary - SAPPHIRE (System Automation of PEMFCs with Prognostics and Health management for Improved Reliability and Economy)

Executive Summary:
The Sapphire project aimed at the extension of the lifetime of PEM fuel cells in µCHP systems.
µCHP systems are considered an important technology by the EU, according to which they should become the "reference technology for on-demand power generation" by 2050.

Sapphire aimed at extending lifetime of PEM fuel cells applying prognostics, i.e. the technique to estimate the residual life of a system by means of diagnostic methods and data analysis. This data would then be used to control the system in such a way as to maximise the lifetime.

The project produced many useful results, including controllers to efficiently counteract common conditions that may cause degradation (poisoning, humidity imbalances, fuel starvation), mostly via software and with the addition of only a single, cheap sensor to the original system.
The developed prognostics will help end users track the status of their µCHP unit and know by what time they will require a overhaul.

A fortuitous yet important discovery was caused by a series of accidental start-stops in the first batch of demonstration on µCHP systems. A frequent start-stop, due to a communication error between the system and additional testing equipment, which resulted in a software glitch, caused a the degradation of the PEM fuel cells to recede steadily over several thousand hours, resulting in world-record degradation rates. These accidental results were confirmed in a second campaign in which the systems were subject to regular start-stops in controlled conditions.

Sapphire achieved and exceeded the quantitative targets that had been set for the project, even if an exact quantification of the stack's lifetime is challenging since voltage degradation has been reduced so much: it is possible that other phenomena (currently undetected or irrelevant) may limit its exact lifetime.

Project Context and Objectives:
== Micro Combined Heat and Power (µCHP) ==
Sapphire focused on stationary µCHP systems for production of heat and power for residential and industrial usage.
A µCHP system produces both heat and power from reformed renewable or fossil fuels, e.g. natural gas, or from a source of pure hydrogen: compared to traditional power and heat sources, it is highly efficient and virtually emission-free.

== Degradation in Low-Temperature PEM Fuel Cells ==
Decay rates and durability of PEM fuel cells are strong functions of operating conditions. PEM fuel cells operate in a very narrow operating window for humidification: too little or too much water can respectively cause drying or flooding, which in extreme cases can occur at the same time in different parts of the stack. In addition, PEM fuel cells in CHP application may be susceptible to contamination with the various constituents in the reformate gas, especially carbon monoxide.

== Prognostics ==
Prognostics is the estimation of time to failure and risk for one or more existing and future failure modes. There are three main approaches: model-based, data-driven and experience-based. Model-based assumes that the degradation process can be modelled mathematically. Data-driven gauges the drift and the rate of change of the current state. Experience-based is used in statistical reliability applications to predict the probability of a failure at any time.

== Control ==
A fuel-cell system has typically several controllers, which can often be operated independently of one another, or decoupled; most prominently, they ensure the flow of reactants to the cells, maintain a given temperature of the stack, adjust the rate of reaction by means of DC/DC converters and set appropriate humidification levels with humidifying units upstream of the cell stack. Often, these controllers operate on very different time scales, which justifies the decoupling approach: for every dynamic mode to control, faster ones can be considered instantaneous, and slower ones can be assumed constant.

== Integrated control and PHM system architecture ==
Building a functional architecture integrating control and PHM management for optimal operation and increasing performance and lifetime of PEMFC. The architecture supports the use of advanced control strategies for more robust system control.

== Degradation and health assessment (diagnostic) methods ==
Developing diagnostic systems based on knowledge on degradation mechanisms and from real system data. The models ranged from first-principle to data-driven models. The developed and implemented methodology was tested and evaluated on experimental setups at FCLAB, FESB and ZSW.

== Prognostics approaches ==
Developing prognostic approaches for estimation of the remaining useful lifetime (RUL) of PEMFCs. The approaches were initially thought to be built on self-adaptive and evolving methods such as neural networks and neuro-fuzzy systems, but during project execution a shift towards model-based prognostics occurred. The developed methodology was tested and evaluated on experimental setups at FCLAB, FESB and ZSW.

== Health adaptive control system ==
Health promoting, adaptive control algorithms were designed and implemented on top of the existing regulatory-control layer already provided by project partner Dantherm Power. The higher-level control were implemented as a multivariable, adaptive and robust controller which alters parameters and set-points of the regulatory level according to the operating condition.

== Prototype testing and validation ==
Demonstrating the applicability and merits of Sapphire by implementing and testing the architecture and methods on two Dantherm Power fuel-cell systems in a real industrial case study, studying the system behaviour using different devised scenarios and measuring its performance and comparing to existing methodologies.

Project Results:
EIFER was responsible for defining the project's experimental protocols, which they updated at the end of the project based on the partners' experience. This resulted in a public deliverable that will help future projects to produce results comparable with Sapphire's.
FCLAB produced a similar public deliverable on characterisation and measurement techniques, which was also updated at the end of the project (D2.4 and D2.6).

FESB investigated and expanded diagnostic techniques, in particular the analysis of polarisation change curves. This technique allows to easily determine graphically the nature of degradation occurring in a fuel cell. More common techniques such as Electrochemical Impedance Spectroscopy (EIS) were also employed, resulting in an extended equivalent circuit model with additional resonant circuits; while the model itself is not novel, the interpretation of the resistance in these resonant circuits is, since it was found to be linearly correlated with degradation, and as such an excellent prognostic variable.

FCLAB produced robust and reliable model-based prognostic algorithms. The prognostics is based on a mixed static-dynamic model, which is regularly updated based on measured data. The update algorithm is able to provide more accurate results by making use of EIS data if these are available, but does not require them, since EIS measurements are often unavailable in the field. The model was also refined to reduce the number of parameters and identify those that have the largest impact on degradation.
EIFER has developed its own data-driven prognostic algorithm. At the time of writing this algorithm is under a patenting process, and cannot be described in further detail.

The overall structure of the control system was first specified under the leadership of EIFER in the first year of the project: the sensors would feed a diagnostic unit that would estimate the State of Health (SoH) of the stack, which would then be fed to the Prognostic unit that would calculate the Residual Useful Life (RUL), which would in turn be fed to the control system that would set an appropriate operating condition to maximise the RUL. Implementing this exact system proved however impractical, since very few variables can be modified significantly without breaking the strict guidelines of stack makers.

SINTEF developed several control loops to address several known degradation factors: carbon monoxide (CO) poisoning, dry-out, flooding and hydrogen starvation.
CO poisoning is usually compensated by air bleed, which is often set to high values to handle the worst expected level of poisoning. SINTEF implemented a feedback routine exploiting the asymmetric dynamics of CO poisoning, and can reduce air bleed by an order of magnitude, still maintaining the stack CO-free.
Dry-out is measured by observing the pressure drop across the cathode, which is the only additional sensor for the whole control system. A steady, low pressure drop indicates drying conditions, whereas an erratic stack voltage is chosen as the indicator of flooding.
Hydrogen starvation is detected by monitoring the flame temperature of the anode exhaust: a reduction of this flame temperature indicates a dangerously low hydrogen content.

Experiments were run both on single cells, stacks and entire systems. ZSW was responsible for laboratory tests on cells and stacks, whereas Dantherm Power tested two of their systems for over 6000 hours each.
ZSW also performed post-test analysis on cells and stacks to better diagnose degradation phenomena, but much of these results cannot be published due to NDAs signed with makers of Membrane-Electrode Assemblies (MEAs).
Dantherm Power's results confirmed the feasibility of Sapphire's controllers, and also (somewhat fortuitously) identified an effective rejuvenation strategy for stacks: this strategy was successfully replicated in the second batch of experiments and will be studied in later projects.

Potential Impact:
= Impact =
The project's main intended long-term impact has been to support market penetration of µCHP systems in the European and worldwide energy mix.
The project has been able to set world records in degradation rates, so low as to reach negative numbers, with one system setting a record-breaking 4 µV/h (per cell) regeneration rate over 3000 hours: this effect is obviously not indefinitely sustainable, and may recede or disappear after a time; however, during the relevant demonstration, this effect was linear with time and did not appear to be close to settle.
Other parts of the system may be impacted by the rejuvenation technique, which involves frequent restarts: in particular, the effect on the fuel processor may be significant and ought to be further investigated.

The development of accurate prognostics has not seen a strong integration with control as originally intended, mostly because the operating conditions of the µCHP system cannot be significantly changed. However, Dantherm Power considers the ability to indicate the residual useful time to customers a very useful feature, which they intend to provide in their future products.

The developed rejuvenation techniques may be applied in other fuel-cell applications other than µCHP, such as the critical automotive sector. These techniques may reduce the total cost of ownership of fuel cell systems and accelerate the market uptake of fuel cells in several sectors.

The work on fuel-cell prognostics in Sapphire has been groundbreaking, and will pave the way for further developments in the field. The Sapphire project directly generated a successor project, Giantleap, which will further develop fuel-cell prognostics to the automotive sector.

The original FCH JU call specified two targets: 20000 hours of lifetime and 100 €/kW of additional costs for the control system.
As indicated in the original proposal, Dantherm Power used to guarantee their systems for 5000 hours; since in one of the tested µCHP systems the degradation rate decreased by a factor of 10 (from 2 to 0.2 µV/h), a lifetime of 50000 hours may be extrapolated. However, the other µCHP system was even reversing its degradation, increasing voltage over time: an exact estimate of the stack lifetime is therefore difficult, but it can with sufficient certainty be set far beyond the FCH JU's original target.
The whole control system was implemented in software, with the exception of one pressure sensor for the cathodic pressure drop. This sensor was indicated to cost 68 € for a 0.9 kW system, which is therefore already within the target 100 €/kW; it should be noted that far cheaper pressure sensors are available on the market, and those may be used in a deployment of the developed control system.

= Dissemination =
Sapphire was a project with a strong research focus, and dissemination was mainly aimed at the academic community. Sapphire produced many publications at international conferences, including the HFCNC 2013, idHea 2014, JCGE 2014, UECT 2014, VPPC 2014, CARISMA 2014, IEEEPHM 2015, EFCF 2015, ADCHEM 2015, FUCE 2016, DYCOPS-CAB 2016, in addition to two articles in peer-reviewed international journals (with another in preparation).
In addition, Sapphire organised an invited session at VPPC 2014 and a booth at the 2016 Hanover Fair, and was invited to the workshops of fellow FCH JU projects Reforcell and Second Act.
In addition, Sapphire maintained a Web site for public information, on which news about the project were regularly published, including all public deliverables.

= Exploitation =
Sapphire generated two patent applications, one for the overall control system (currently managed by Électricité de France and ZSW) and another one currently being submitted by EIFER on a data-driven prognostic system.
A direct exploitation of results will be the integration of Sapphire's result in Dantherm Power's µCHP systems, pending a further refinement of the controllers; roll-out is expected by 2017.

List of Websites:
Coordinator and exploitation manager: Federico Zenith,
Specification and architecture: Philippe Moçotéguy,
Stack Development and Laboratory Experiments: Joachim Scholta,
System Development and Demonstration: Thomas Pedersen,
Diagnostics: Frano Barbir,
Prognostics: Rafael Gouriveau,
Control: Johannes Tjønnås,
Dissemination manager: Marie-Cécile Péra,