Fault Diagnosis Of Aircraft APU Based On The Long Short-term Memory

Auxiliary Power Unit,as an important device of the aircraft,can not only ensure the safe start of the aircraft,when the aircraft is parked on the ground,but also provide air and electricity for the aircraft,ensuring cabin comfort.Therefore,it is particularly important to study the fault diagnosis of aircraft APU.The following studies are carried out for the fault diagnosis of APU:Firstly,an optimized Long Term Memory aircraft APU fault diagnosis model is proposed.LSTM has a large uncertainty and randomness of manually selected parameters,which seriously affects the fault diagnosis results.The improved particle swarm optimization algorithm is used to optimize the number of hidden layer nodes of LSTM.Improved particle swarm optimization algorithm with clustering thought divide the particle swarm into different study groups,particles can be divided into ordinary students particles and team leader particles,to update the ordinary student particles with traditional way,with the sum of the information weight of each learning group leader particle to guide the update of leader particle.This increases the information interaction between the study groups.Moreover,the modified vector is added to criticize and inherit the locally optimal particle,so as to avoid falling into the locally optimal solution,and the performance test of the adaptive particle swarm optimization algorithm shows that the APSO algorithm is better than the traditional PSO algorithm.With hidden layer nodes of LSTM as the class particles of APSO algorithm.After initialization,optimization is carried out,the optimal value obtained is substituted into the LSTM network model.The APSO-LSTM fault diagnosis model is used for fault diagnosis of aircraft APU.The experimental results show that the APSO-LSTM model has good fault recognition capability compared with the traditional LSTM and other relevant algorithms.Secondly,an adaptive quantum particle swarm optimization algorithm is proposed to replace particles with quantum state.Firstly,the population particles updated with speed and position are converted into a quantum state updated with spatial position,so that the parameters are reduced and the optimization speed is faster.Then,the AQPSO algorithm is obtained by improving the contraction and expansion coefficient of the quantum particle swarm.The performance test results show that AQPSO algorithm is closer to the ideal value than APSO algorithm.The hidden layer node of LSTM is taken as the quantum particle of AQPSO algorithm,which is optimized after initialization,and the obtained optimal hidden layer node number is substituted into the LSTM network model.In addition,aiming at the problem of gradient disappearance and gradient explosion of deep network,the paper proposes to add batch normalization idea into the network model.The batch normalization layer is set above the hidden layer of LSTM network model.Finally,the AQPSO-LSTM-BN fault diagnosis model is built,and the obtained AQPSO-LSTM-BN fault diagnosis model is used for fault diagnosis of aircraft APU.The simulation results show that the AQPSO-LSTMBN fault diagnosis model improves the fault recognition rate and model training speed,and is more stable,which can achieve better fault diagnosis effect.