Fault Detection for Aircraft Fuel System with Neural Network

Abstract

The technological advances in the aircraft industry in the last decade have increased the complexity of aircraft systems. This, in turn, makes the fault detection, diagnosis and modification/ repair processes more difficult. The presence of a fault within a system can result in changes to system function, reduce system performance and cause operational downtime. Due to this reason Condition Based Maintenance (CBM) which predicts the state of the component on based upon data gathered is widely used in aircraft MRO industries. CBM uses diagnostics and prognostics models to make decisions on appropriates maintenance actions based upon the remaining used life (RUL) of the components. In this research, we applied a Neural Network model to solve the fault detection problem, and the experimental results demonstrated the neural network model can obtain excellent performance. Fault diagnosis is a more complicated problem, and it requires diagnosing the type of fault. Therefore, fault diagnosis becomes a classification problem. More importantly, the fault state of a fuel system may relate to the previous state of the fuel system. Therefore, a Recurrent Neural Network model could be developed for fault diagnosis.