Real-Time FPGA-Based Hardware Neural Network for Fault Detection and Isolation in More Electric Aircraft

Abstract

More Electric Aircraft (MEA) has a great development prospect in the aviation industry thanks to the progressive power electronics technology and digital systems. Nevertheless, any fault occurring in the power electric system of MEA could be a fatal risk for the safety of the aircraft. To deal with the real-time fault detection and isolation (FDI) on the MEA, we put forward an FPGA-based neural network method which includes two stages: off-line construction using TensorFlow and real-time monitoring on the FPGA. Long Short-Term Memory (LSTM) network is applied because of its capability of learning from the long-term historical information in time series. A comprehensive MEA model based on the Boeing 787 power system created in PSCAD/EMTDC® and a commercial electric aircraft model based on Airbus E-Fan in Simscape are simulated to validate the effectiveness and generality of our proposed method. Adequate contrast experiments are conducted to acquire the most applicable architecture and configurations of the network. The results illustrate that the evaluation of the real-time condition can achieve accuracy over 99.5% within one sampling time on FPGA and reasonable hardware resource utilization.