Project Summary

Our vision is to achieve zero-emissions megawatt-scale fuel cell propulsion systems for regional aircraft markets utilising an innovative digital twin approach to rapidly design and control our system.In this project, building off of a previous project funded by Innovate UK Fast Start programme, we will develop a digital twin of connected surrogate models to accurately simulate the multi-physics behaviour of the fuel cell more time- and cost-effectively than traditional methods. The expected outputs will start us on our journey towards a ‘Digital Twin’ toolset to inform the design of new, innovative megawatt-scale fuel cells with improved performance and reliability. 

Project Achievements

Project activities were developed to test the proposed theory of linking multiple surrogate models together to provide a performance prediction for a larger system. The main test involved using two linked surrogate models and 3 linked surrogate models to arrive at a result for a single section of channel length. Simulations were carried out using ANSYS Fluent’s PEMFC Module and surrogate models were trained using MATLAB’s Deep Learning Toolkit. The result from the linked up trained models was compared with simulation results to assess the viability of this approach.

Conclusions

For this project, errors within 2% of the simulated result are considered ‘successful’ and within 5% are considered ‘acceptable’. In the best comparison case, the average error for two linked models was 0.39% (high of 2.57%) and for three linked models was 5.12% (high of 20.5%). This has led to the conclusion that feasibility has been proven, whilst noting that a sufficiently large and well-distributed data set is very important. Time saved to develop models using this method is thought to be more than 90%. The techniques could be useful in early design-stages to quickly reach more optimal design solutions at a fraction of the time and cost.

Next Steps

To advance this work from experimental proof of concept to higher TRLs, it is expected that simulation models should be rigorously validated against physical lab-test data to ensure both simulations and surrogate models are accurately representing real test conditions. Proving the technology in a relevant environment is expected to involve using a set of surrogate models linked in such a way that they can return an accurate performance prediction for a full fuel cell flow field. EAG is putting together project proposals to develop megawatt-scale fuel cells for aerospace applications. This development would include physical testing of large format cells for which it is hoped that this potential solution could aid in the design of by reducing the simulation burden. EAG is shaping work packages within this larger megawatt fuel cell development program to include work towards developing this solution further. It is envisaged that a sufficiently developed surrogate model set could be deployed as a design tool towards EAG’s in-house developments and design consultancy services.