Fault simulations and diagnostics for a Boeing 747 Auxiliary Power Unit

Health monitoring of aircraft systems is of great interest to aircraft manufacturers and operators because it minimises the aircraft downtime (due to avoiding unscheduled maintenance), which in turn reduces the operating costs. The work that is presented in this paper explores, for a Boeing 747 APU, fault simulation and diagnostics for single and multiple component faults. Data that corresponds to healthy and faulty conditions is generated by a calibrated simulation model, and a set of performance parameters (symptom vector) are selected to characterise the components health state. For each component under examination, a classification algorithm is used to identify its health state (healthy or faulty) and the training strategy that is used considers the existence of multiple faults in the system. The proposed diagnostic technique is tested against single and multiple fault cases and shows good results for the compressor, turbine, Load Control Valve (LCV) and Fuel Metering Valve (FMV), even though these faults present similar fault patterns. On the contrary, the classifiers for the Speed Sensor (SS) and the generator do not provide reliable predictions. As regards the SS, the sensitivity assessment for this component showed that the existence of faults in the other components can sometimes mask the SS fault. The reason that the generator diagnosis fails under the proposed diagnostic technique is attributed to the fact that it has only a very slight influence on the other symptom vector parameters. In both cases, additional diagnostic strategies are suggested.