In this paper we propose a hybrid modeling approach for generating reduced models of a high fidelity model of a physical system. We propose machine learning inspired representations for complex model components. These representations preserve in part the physical interpretation of the original components. Training platforms featuring automatic differentiation are used to learn the parameters of the new representations using data generated by the high-fidelity model. We showcase our approach in the context of fault diagnosis for a rail switch system. We generate three new model abstractions whose complexities are two order of magnitude smaller than the complexity of the high fidelity model, both in the number of equations and simulation time. Faster simulations ensure faster diagnosis solutions and enable the use of diagnosis algorithms relying heavily on large numbers of model simulations.
Design of Intelligent Commutation Switch System Based on HPLC Carrier Scheme
