Remaining Useful Life Prediction Method Of Aero-Engine Combining Cascaded LSTM Model With Empirical Formula Optimization

As the “heart” of an aircraft,the aero-engine is the source of power for the aircraft during flight.However,aero-engines often work in extreme and complex natural environments,causing them susceptible to various types of failures.To reduce the risk of aero-engine failure and ensure the flight safety of aircraft,the remaining useful life of aero-engine needs to be accurately predicted so as to monitor and grasp the operating status of the engine in time and help maintenance personnel make the right decision before failure occurs.In the research of remaining useful life prediction techniques,Long-Short Term Memory(LSTM)networks are widely favored by researchers for their excellent performance in time series data prediction.However,because of the complex mechanical structure and operation mode of aero-engine,the accuracy of the results obtained by utilizing the traditional LSTM network to predict the remaining useful life of aero-engine is not sufficient to meet the accuracy requirements of its life prediction.Therefore,it is essential to design a new prediction model to achieve the accurate prediction of the remaining useful life of aero-engine.This topic is based on a data-driven approach to carry out the study of the remaining useful life prediction method by using the performance degradation data of aero-engines provided by NASA.The main research contents of this topic are listed below:(1)An Automatically Cascaded LSTM(ACLSTM)model is constructed on the basis of LSTM network for high-dimensional time-series data in the dataset.The ACLSTM model relies on the network structure of multiple sub-modules connecting one by one,and continuously extracts the output error of the previous level sub-module for the network training of the next level sub-module,which algorithmically constitutes a multi-level prediction mechanism of the error,ensuring that the prediction error of ACLSTM model is effectively reduced and greatly improving the prediction performance of the model.(2)To better meet the demand of aero-engine life prediction with high accuracy,a segmented linear degradation model is reconstructed based on the summary of aero-engine life decay law,and several empirical formulas are refined based on this model to further optimize the prediction results of ACLSTM model.The proposed empirical formula allows some prediction data with large deviations to be corrected in accordance with the life decay law of aero-engine,thus ensuring the accuracy and objectivity of the final prediction results.(3)In the experimental part,the prediction effectiveness of the ACLSTM model and the validity of the empirical formula are verified by applying four subsets of the Commercial Modular Aero-propulsion System Simulation dataset,and the prediction effectiveness of the ACLSTM model is compared with that of several advanced algorithms.The experimental results show that,compare with a single sub-module without cascading,the Root Mean Square Error(RMSE)of the prediction results in the four experiments is reduced by95.44% on average when the sub-modules are cascaded 10 times;and after optimizing the prediction results of the model after 10 cascades using the empirical formula,the RMSE is even reduced by an average of 96.48%;the RMSE of the ACLSTM model is the lowest compare to the prediction results of other advanced algorithms,which indicates that the use of ACLSTM model reaps more excellent prediction results.The experiments fully verify that the proposed ACLSTM model and empirical formula optimization method can effectively improve the accuracy of remaining useful life prediction of aero-engines and show the significant advantages in the field of life prediction.