Aero-engine Fault Diagnosis Research Based On Elman Neural Network

Aero-engine is the core component of the aircraft.Its running state affects the safety of the whole flight process.At present,the maintenance cost of aero-engine accounts for 60% of the total maintenance cost of airlines,so engine fault diagnosis and monitoring plays a key role in ensuring flight safety and economic use.In engine fault diagnosis,the engine is abstracted as a typical complex mechanical system.It is difficult to establish a high-precision fault diagnosis model because of the complex engine structure,nonlinear model,various diagnostic methods,and flight factors.Artificial neural network has the advantages of self-adaptability,self-learning and fault-tolerance,so it can be used to diagnose engine faults.In this paper,based on Elman neural network,the main research contents are as follows: For its traditional excitation function affects the convergence speed and is prone to local minimum and initial weight and threshold selection.On the basis of the original,the training algorithm and excitation function of Elman neural network are improved,and the initial weights and thresholds of Elman neural network are optimized by using the gravitational search algorithm,and then the GSA-Elman neural network is proposed.Finally,taking the fault data of engine gas path collected from airlines as samples,the simulation fault diagnosis is carried out in MATLAB environment.The results show that compared with the traditional BP and Elman network,the improved Elman neural network has faster convergence speed and smaller absolute error.The diagnostic absolute error of the improved Elman neural network is smaller than that of the other two methods under different training samples.At the same time,the improved GSA-Elman network also has good anti-noise ability.In different condition monitoring systems,although the types of input parameters have changed,the network error changes a little,which shows a strong adaptability.Finally,GM(1,n)and GSA-Elman are used to make a comparison.The results show that the diagnosis effect of GSA-Elman network is better when the number of samples is relatively large as well as when it has the characteristic of nonlinearity.