Application Of Improved Convolutional Neural Network In Engine’s Remaining Useful Life Prediction

The goals of prognostic and health management(PHM)include monitoringing the equipment status,maximizing the availability of operation,reducing the maintenance costs,and improving system reliability and security.The remaining useful life is an important part of PHM and can be estimated based on historical trajectory data.In order to enhance the convergence of the neural network,limited BFGS algorithm(L-BFGS)based on trust region search is used to replace the traditional neural network optimization algorithm in the later stage of model’s gradient descent,experiments are carried out on the datasets used in the prediction of remaining useful life of aircraft engine.1.LU decomposition is introduced to improve the computational efficiency of Newton Hessian matrix’s inversion,and regular Newton method is introduced to solve the positive definiteness problem of Hessian matrix to ensure that Newton’s method can be used iteratively.2.To solve the problem of loss oscillation in the later stage of neural network’s gradient descent,L-BFGS algorithm based on trust region algorithm、Newton method and regular Newton method is used as the later stage optimization algorithm instead of the original optimization algorithm,and UCI common dataset is used to verify that the combination of L-BFGS and trust region search can have better prediction performance.3.Based on the prediction dataset of aircraft engine unit,the deep convolutional neural network is implemented by Python to predict the remaining useful life.Deep convolutional neural network based on L-BFGS algorithm combining with trust region search has good prediction ability and effectively increases the convergence performance of neural network.The prediction results are compared with the prediction obtained by the methods in the literature,and RMSE error of the test set is reduced by 3.0199%.