Aircraft APU Starter Health Monitoring and Failure Prognostics

Abstract

Auxiliary power unit (APU) is a gas turbine engine on aircraft that provides energy for functions other than propulsion. Its starter is a crucial component that outputs assistant power to support the APU starting process. Starter performance degradation significantly impairs the whole APU life and raises risks for the aircraft flight. However, the current maintenance policy for the starter is still "run it till it breaks". An effective technique for the starter diagnostics and prognostics has not been reported yet. The aim of this thesis is to propose a framework for enabling the online detection and prediction of starter degradation. For this purpose, the thesis makes use of a dataset containing information about 52 APU "inability to start" failure events that were collected from actual aircraft operations over a period of ten years. Through the establishment of the relationship between the starter degradation and gas turbine engine starting performance, 13 of these 52 failures were identified as being caused by the starter degradation. Once this determination has been made, an online classifier based on moving autocorrelation is designed to detect the initial phase of degradation for each failure. Finally, a particle filtering based approach with an associated system state model is proposed to achieve the fault diagnostics and failure prognostics. The results demonstrate that a condition based maintenance program for the APU starter can be implemented to avoid unnecessary economic losses and to enhance aircraft operating safety. i First and foremost, I would like to express my deepest gratitude to my supervisor Dr. LIU Jie for his enthusiastic supervision and patient guidance. He is the best supervisor I had ever met, with encourage, enthusiasm, and immense knowledge. He supported and guided me throughout the entire master program. Without him, it would not have been possible to have this thesis. Also, I would like to sincerely thank our research collaborator Dr. YANG Chunsheng from National Research Council Canada for his insights and nice helps in this two-year project.

Performance Sensing Data Prediction for an Aircraft Auxiliary Power Unit Using the Optimized Extreme Learning Machine

“…PHM has strong theoretical and engineering research demand for condition-based maintenance. PHM can provide effective information support for failure warning [8], anomaly detection [9,10], remaining useful life (RUL) prediction [11,12], and fault detection [13]. The provided information can help APU maintenance in implementing repair and maintenance in a reasonable manner to save cost and enhance reliability [14].…”

Performance Estimation and Fault Diagnostics for the Starter of Auxiliary Power Unit

“…Deterioration of the performance parameters were observed to occur sequentially and the failure was detected based on measurements within multiple time steps [57]. Lou employed measurements from the APU sensors as indicators for the starter health monitoring and failure detection [58]. However, since the monitored performance parameters were selected at the system level, the results could potentially be affected by Most diagnostic approaches in the literature are based on statistical analysis and datadriven techniques [12], [57]- [60].…”

“…Some previous works in the literature attempt to estimate the performance of the starter based on statistical analysis and data-driven methods [63], [64]. These approaches lack transparency and do not possess tangible physical meanings [58], [60], [65]. They also significantly depend on historical measurement data.…”

“…Alternative to the physics-based model in Chapter 2, a back-propagation, feedforward neural network is used for APU performance monitoring to compare with the physicsbased results. Previous works [58], [59] have attempted to estimate the starter degrading trend, however a clear symptom for the starter failure is not provided, which could be instrumental for the CBM. Both physics-based model and neural network model are implemented in the healthy and faulty conditions, and their capabilities as monitoring tools for the starter failure are verified.…”

“…Traditionally, the threshold is determined in one dimension for individual indicators [57], [58]. That can cause confusion in the decision making when working with multiple indicators, as any indicator may exceed its respective threshold at a different time.…”