Failure Diagnosis and Prognosis of Safety Critical Systems: Applications in Aerospace Industries

Many safety-critical systems such as aircraft, space crafts, and large power plants are required to operate in a reliable and efficient working condition without any performance degradation. As a result, fault diagnosis and prognosis (FDP) is a research topic of great interest in these systems. FDP systems attempt to use historical and current data of a system, which are collected from various measurements to detect faults, diagnose the types of possible failures, predict and manage failures in advance. This thesis deals with FDP of safety-critical systems. For this purpose, two critical systems including a multifunctional spoiler (MFS) and hydro-control value system are considered, and some challenging issues from the FDP are investigated. This research work consists of three general directions, i.e., monitoring, failure diagnosis, and prognosis. The proposed FDP methods are based on data-driven and model-based approaches. The main aim of the data-driven methods is to utilize measurement data from the system and forecast the remaining useful life (RUL) of the faulty components accurately and efficiently. In this regard, two dierent methods are developed. A modular FDP method based on a divide and conquer strategy is presented for the MFS system. The modular structure contains three components:1) fault diagnosis unit, 2) failure parameter estimation unit and 3) RUL unit. The fault diagnosis unit identifies types of faults based on an integration of neural network (NN) method and discrete wavelet transform (DWT) technique. Failure parameter estimation unit observes the failure parameter via a distributed neural network. Afterward, the RUL of the system is predicted by an adaptive Bayesian method. In another work, an innovative data-driven FDP method is developed for hydro-control valve systems. The idea is to use redundancy in multi-sensor data information and enhance the performance of the FDP system. Therefore, a combination of a feature selection method and support vector machine (SVM) method is applied to select proper sensors for monitoring of the hydro-valve system and isolate types of fault. Then, adaptive neuro-fuzzy inference systems (ANFIS) method is used to estimate the failure path. Similarly, an online Bayesian algorithm is implemented for forecasting RUL. Model-based methods employ high-delity physics-based model of a system for prognosis task. In this thesis, a novel model-based approach based on an integrated extended Kalman lter (EKF) and Bayesian method is introduced for the MFS system. To monitor the MFS system, a residual estimation method using EKF is performed to capture the progress of the failure. Later, a transformation is utilized to obtain a new measure to estimate the degradation path (DP). Moreover, the recursive Bayesian algorithm is invoked to predict the RUL. Finally, relative accuracy (RA) measure is utilized to assess the performance of the proposed methods