Fault Sensor Isolation Method for Liquid Rocket Engines Based on Multi-Model

This paper presents a multiple-mode-based approach for sensor fault isolation in liquid rocket engines. The approach comprises a “two-step” precise model identification technique and a particle filter bank-based sensor fault isolation strategy. By leveraging the “two-step” precise model identification process, this approach effectively addresses the challenges associated with inaccurate modeling in traditional model-based sensor fault diagnosis methods. Moreover, the particle filter bank-based strategy facilitates the accurate identification and isolation of faulty sensors. The “two-step” method integrates subspace model identification method with forecast error detection. Initially, the subspace identification technique estimates the engine's coefficient matrix to produce a preliminary model. Subsequently, the prediction error method refines and corrects the coefficient matrix, yielding an accurate system model. Once the system model is established, a series of particle filters are constructed based on this model, with each filter specifically tailored to an individual sensor. In the event of a sensor failure, only the particle filter designed for that particular sensor remains unaffected, enabling effective fault isolation through residual analysis. The algorithm's performance is thoroughly evaluated using both simulation data from the Space Shuttle Main Engine (SSME) and experimental data from the SSME fuel flow path simulation test bench. These evaluations demonstrate the robustness of the approach under various operating conditions and confirm its applicability in real-world scenarios.