System diagnosis for an auxiliary power unit

Abstract

Even though the Auxiliary Power Unit (APU) is a widely used system in modern aviation, the existing experimental, simulation and diagnostic studies for this system are very limited. The topic of this project is the System Diagnosis of an APU, and the case study that is used in this research is a Boeing 747 APU. This APU was used to develop an experimental rig in order to collect performance data under a wide range of loading and environmental conditions. The development of the experimental rig consumed considerable time and required the design and installation of structures and parts related with the control of the APU, the adjustment of the electric and pneumatic load and the data acquisition. The validation of the rig was achieved by a repeatability test, which ensures that the collected measurements are repeatable under the same boundary conditions, and by a consistency test, which ensures that the performance parameters are consistent with the imposed ambient conditions. The experimental data that are extracted from the rig were used to calibrate a physics-based (0-D) model for steady-state conditions. Data that correspond to faulty conditions were generated by injecting faults in the simulation model. Based on the most prominent APU faults, as reported by The Boeing Company, six components that belong to different sub-systems were considered in the diagnostic analysis, and for each one of them, a single fault mode was simulated. By using healthy and faulty simulation data, for each component under examination, a classification algorithm that can recognise the healthy and faulty state of the component is trained. A critical part of the diagnostic analysis is that each classifier was trained to recognise the healthy and the faulty state of the corresponding component, while other components can be either healthy or faulty. The test results showed that the proposed technique is able to diagnose both single and multiple faults, even though in many cases different component faults resulted in similar fault patterns.