Periodic Reporting for period 2 - TWINECS (Toward a Digital Twin ECS and thermal management architecture models: Improvement of MODELICA libraries and usage of Deep Learning technics)
Periodo di rendicontazione: 2022-03-01 al 2023-08-31
This project deals with the modelling of a VCS system within the framework of a system-oriented simulation of the whole ECS for control and optimization purposes. This work is aligned with the Clean Sky 2 SYS ITD WP6 – Major loads, WP6.1 – Electrical Air System and Thermal Management. Within this work package, this topic is specifically related to the subsection about the New Generation of Cooling Systems.
The aims of the TwinECS project are, on the one hand, to enhance existing Modelica libraries to simulate Environmental Control Systems (ECSs) based on the major challenges underlined in the Clean Sky 2 MALET project, and on the other hand, to develop and integrate surrogate models based on data analytics technics to reduce computational costs. The TwinECS specific objectives are summarized as follows:
• Development of thermo-fluid models: VCS heat exchangers and their successful integration in assembled VCS models.
• Development of electrical models for: motor, power inverter, ATRU, IGBT and MOSFET.
• Development of surrogate models of the aforementioned thermo-fluid and electrical components based on data atechniques technics.
• Simulation of a complete thermo-fluid-electrical VCS model using both the standard and the surrogate models.
The numerical tool developed within this project has proven to overcome the main challenges underlined in the Clean Sky 2 MALET project. The required performance in terms of accuracy, computational time, and more importantly, numerical robustness has been achieved. The tool will play a crucial role in the design of more efficient ECS within the new More Electrical Approach (MAE). The ECS optimization supported by model simulations will contribute to the global Clean-Sky objectives such as competitiveness improvement and reduction of fuel consumption.
The aims of the TwinECS project are, on the one hand, to enhance existing Modelica libraries to simulate Environmental Control Systems (ECSs) based on the major challenges underlined in the Clean Sky 2 MALET project, and on the other hand, to develop and integrate surrogate models based on data analytics technics to reduce computational costs. The TwinECS specific objectives are summarized as follows:
• Development of thermo-fluid models: VCS heat exchangers and their successful integration in assembled VCS models.
• Development of electrical models for: motor, power inverter, ATRU, IGBT and MOSFET.
• Development of surrogate models of the aforementioned thermo-fluid and electrical components based on data atechniques technics.
• Simulation of a complete thermo-fluid-electrical VCS model using both the standard and the surrogate models.
The numerical tool developed within this project has proven to overcome the main challenges underlined in the Clean Sky 2 MALET project. The required performance in terms of accuracy, computational time, and more importantly, numerical robustness has been achieved. The tool will play a crucial role in the design of more efficient ECS within the new More Electrical Approach (MAE). The ECS optimization supported by model simulations will contribute to the global Clean-Sky objectives such as competitiveness improvement and reduction of fuel consumption.
THERMO-FLUID SYSTEM MODELLING
Development of Heat Exchangers models
• A new flexible modelling approach has been implemented in order to satisfy the requirements in terms of numerical robustness and CPU time consumption.
• A calibration procedure based on reference data has been implemented to reach the expected accuracy levels.
• A complete set of numerical tests has been implemented to assess the model's numerical robustness concerning demanding configuration and operational aspects.
Development of Vapor Compression System models
• A complete VCS model has been assembled integrating the new heat exchanger models.
• The model includes appropriate technics to conduct the regulation of the thermostatic valve and to manage the refrigerant charge.
• A complete set of numerical tests has been implemented to assess the model numerical robustness for demanding configuration and operational aspects.
ELECTRICAL SYSTEMS MODELLING
Development of motor, power converters and transistor models that were rigorously tested with various operating conditions and simulation time steps.
• Dynamic phasor modelling domain adopted to enable suitability to large time-steps.
• The drive system consists of three-phase inverter which drives a permanent magnet synchronous motor (PMSM).
• Electrical drive control implemented with field-oriented control (FOC).
• Robust control design ensuring stable performance under reference speed changes and load torque changes.
• Power loss calculation for each component.
• Numerical robustness and low CPU time obtained for demanding cases including complete switch-off and switch-on cases.
SURROGATE MODELS DEVELOPMENT
Development of surrogate models for both thermal and electric components and systems.
• Electrical model surrogates have been based on the Gaussian Process (GP) method using Maximum Likelihood Estimation technics for the training. High performance has been observed based on K-fold cross-validation techniques.
• Thermal model surrogates have been based on both the Gaussian Process and the Multi-Layer Perceptron (MLP) methods (the latter is used for outputs with high nonlinearity). The training has been performed using Maximum Likelihood Estimation techniques and the Adam Optimizer respectively. Hyper Parameter Optimization has been used for the MLP to improve its accuracy. Performance assessment has been based on the Root Mean Square Error and different sizes of training datasets.
• Both methods, GP and MLP, are calculated within the Modelica environment from the trained data (the training is conducted in the Python environment).
THERMOELECTRICAL COUPLING AND SYSTEM VALIDATION
• The VCS model has been successfully validated based on reference data provided by the Topic Manager for several design steady-state operation conditions.
• The thermal system has been mechanically linked to the electrical system using two non-invasive elements (one of them integrates dynamic relaxations for both shared values, torque and speed). The linking has proven to be fast and numerically robust at demanding conditions with different time steps.
GLOBAL ASSESSMENT
Robust thermal and electric models have been developed and coupled successfully. They have proven to fulfill all the numerical requirements for their use in large ECS architecture simulations. In addition, surrogate models have also been developed.
Development of Heat Exchangers models
• A new flexible modelling approach has been implemented in order to satisfy the requirements in terms of numerical robustness and CPU time consumption.
• A calibration procedure based on reference data has been implemented to reach the expected accuracy levels.
• A complete set of numerical tests has been implemented to assess the model's numerical robustness concerning demanding configuration and operational aspects.
Development of Vapor Compression System models
• A complete VCS model has been assembled integrating the new heat exchanger models.
• The model includes appropriate technics to conduct the regulation of the thermostatic valve and to manage the refrigerant charge.
• A complete set of numerical tests has been implemented to assess the model numerical robustness for demanding configuration and operational aspects.
ELECTRICAL SYSTEMS MODELLING
Development of motor, power converters and transistor models that were rigorously tested with various operating conditions and simulation time steps.
• Dynamic phasor modelling domain adopted to enable suitability to large time-steps.
• The drive system consists of three-phase inverter which drives a permanent magnet synchronous motor (PMSM).
• Electrical drive control implemented with field-oriented control (FOC).
• Robust control design ensuring stable performance under reference speed changes and load torque changes.
• Power loss calculation for each component.
• Numerical robustness and low CPU time obtained for demanding cases including complete switch-off and switch-on cases.
SURROGATE MODELS DEVELOPMENT
Development of surrogate models for both thermal and electric components and systems.
• Electrical model surrogates have been based on the Gaussian Process (GP) method using Maximum Likelihood Estimation technics for the training. High performance has been observed based on K-fold cross-validation techniques.
• Thermal model surrogates have been based on both the Gaussian Process and the Multi-Layer Perceptron (MLP) methods (the latter is used for outputs with high nonlinearity). The training has been performed using Maximum Likelihood Estimation techniques and the Adam Optimizer respectively. Hyper Parameter Optimization has been used for the MLP to improve its accuracy. Performance assessment has been based on the Root Mean Square Error and different sizes of training datasets.
• Both methods, GP and MLP, are calculated within the Modelica environment from the trained data (the training is conducted in the Python environment).
THERMOELECTRICAL COUPLING AND SYSTEM VALIDATION
• The VCS model has been successfully validated based on reference data provided by the Topic Manager for several design steady-state operation conditions.
• The thermal system has been mechanically linked to the electrical system using two non-invasive elements (one of them integrates dynamic relaxations for both shared values, torque and speed). The linking has proven to be fast and numerically robust at demanding conditions with different time steps.
GLOBAL ASSESSMENT
Robust thermal and electric models have been developed and coupled successfully. They have proven to fulfill all the numerical requirements for their use in large ECS architecture simulations. In addition, surrogate models have also been developed.
The library to simulate ECS based on the Modelica/Dymola framework is being optimized in terms of numerical robustness, CPU time consumption and accuracy. The main focus is placed on the VCS and the electrical components.
Compared to the state of the art, this enhanced ECS simulation tool contributes to study and introduce new ECS architectures including supplemental cooling systems.
The library characteristic allows reproducing a large amount of cases (good numerical robustness) with significantly low CPU time (optimized time consumption based on appropriate modelling methods) that supports the activities needed to comply the high level objectives at demonstrator/technology level. Among them:
• Selection of new generation fluid compliant with environmental future rules and aerospace certification constrains.
• Study of variable and non-variable speed compressor centrifugal technologies
• Design and development of optimized valve for dynamic regulation of VCS.
• Control capabilities. Improvement and definition of an optimized control strategy (introduction of the surrogate modelling approach to accelerate and stabilize the non-linear and complex nature of VCS systems).
• Design and optimization of heat Exchanger focused on VCS application.
Compared to the state of the art, this enhanced ECS simulation tool contributes to study and introduce new ECS architectures including supplemental cooling systems.
The library characteristic allows reproducing a large amount of cases (good numerical robustness) with significantly low CPU time (optimized time consumption based on appropriate modelling methods) that supports the activities needed to comply the high level objectives at demonstrator/technology level. Among them:
• Selection of new generation fluid compliant with environmental future rules and aerospace certification constrains.
• Study of variable and non-variable speed compressor centrifugal technologies
• Design and development of optimized valve for dynamic regulation of VCS.
• Control capabilities. Improvement and definition of an optimized control strategy (introduction of the surrogate modelling approach to accelerate and stabilize the non-linear and complex nature of VCS systems).
• Design and optimization of heat Exchanger focused on VCS application.