A Methodological Framework for Prognosis Using Control-Oriented Models: Application to an Aeronautical Power Converter

Abstract

Developing Prognostics and Health Management (PHM) for safety-critical systems faces a major challenge. Obtaining degradation and failure data is both expensive and time-consuming, especially during the design and development phases. Models built at this stage for control specification and verification, such as those in MATLAB/Simulink using the Specialized Power Systems toolbox, were not designed to capture component faults or track degradation over multi-year aging horizons. In addition, their single-domain electrical focus neglects important multi-physics interactions like electro-thermal feedback. To address this gap, this work applies a parametric four-stage workflow that leverages the computational efficiency of electrical models, making long-duration degradation simulations practical for data generation. We apply this approach to an aeronautical power converter, with the Silicon Carbide (SiC) Metal-Oxide-Semiconductor Field-Effect Transistor (MOSFET) selected as the most reliability-critical component. The increase in on-state resistance is used as the principal degradation indicator. Without altering the model's structure, degradation is introduced through controlled parametric fault injection to generate structured degradation datasets, followed by systematic feature engineering of electrical signatures to identify degradation-sensitive patterns. Top-ranked features are then used to train regression models that provide a diagnostic estimate of fault severity. This end-to-end workflow validates the PHM pipeline using available design-stage tools, establishing a performance baseline and reducing technical risk before transitioning to higher-fidelity multi-physics co-simulation.