| The aero-engine serves as the main power source for airplanes,and its safety and reliability are crucial for ensuring aviation safety.Establishing an effective aero-engine prediction and health management mechanism is of great significance for ensuring its safety and economy.Remaining useful life(RUL)prediction is an important part of prediction and health management technology,and with the rapid development of sensor and data mining technology.In order to effectively monitor the health status of aeroengines,sensors are usually used to monitor the performance parameters of multiple critical components.Data-driven aero-engine RUL prediction methods typically integrate monitoring data from multiple performance parameters to accurately predict RUL.However,due to the large number of sensors,noise,operating condition fluctuations,and coupling of multiple failure modes,the monitoring data for multiple performance parameters contain a lot of redundancy and irrelevance,which seriously affects the accuracy of aero-engine RUL prediction results.Therefore,conducting research on aeroengine RUL prediction related to these influencing factors is of great significance for optimizing maintenance strategies and preventing accidents.The thesis focuses on aero-engine and uses monitoring data of multiple performance parameters during the degradation process of the engine as the source information carrier.It investigates how to select aero-engine performance parameters,improves the residual life prediction method based on degradation trajectory similarity,and also proposes a residual life prediction method based on a hybrid neural network model considering multiple operating conditions,and verifies the effectiveness of the proposed methods on the publicly available NASA C-MAPSS dataset.The specific research content of the paper is as follows:(1)Propose an information theory-based screening method for aero-engine performance parameters.Due to the presence of redundant and irrelevant information in the monitoring data of engine performance parameters,direct use of raw data for RUL prediction can lead to overfitting and seriously affect the accuracy of prediction results.Therefore,this study uses information theory to compare the entropy,permutation entropy,and mutual information of each performance parameter monitoring data to measure the amount of degradation information,data trends,and data correlation,and select performance parameter monitoring data that can effectively characterize the health status of aero-engines to reduce data redundancy and irrelevance.(2)Put forward an improved remaining useful life(RUL)prediction method based on Trajectory Similarity-Based Prognostic(TSBP).TSBP method is adopted to avoid the problem of inaccurate setting of failure threshold,and accurately evaluate the remaining life of equipment by matching the historical monitoring data of similar equipment with the online monitoring data of the in-service equipment.However,the traditional method only analyzes the similarity between the degradation trajectories locally and does not consider the case where the overall length of the degradation trajectories is unequal.This study uses dynamic time warping and Euclidean distance to measure similarity both globally and locally,and improves the similarity matching method to enhance the accuracy of RUL prediction based on TSBP for aero-engines considering the impact of dynamic changes in operational conditions.(3)Provide a residual life prediction method combining a recurrent neural network model with degradation trajectory similarity method.Due to the limitations of the degradation trajectory similarity method,such as insensitivity to degradation information in the early and middle stages of the degradation process and insufficient utilization of its own historical data,the predicted results in the early and middle stages often deviate significantly from the actual residual life.To address these issues,a long short-term memory neural network model with strong long-term sequence processing capability in the recurrent neural network is adopted to vertically mine the degradation trajectory in the early and middle stages of the degradation process,predict the subsequent changes in the trajectory,and combine the predicted sequence with the original sequence to increase the degradation information contained in the trajectory.Finally,the combined sequence is used for similarity matching to improve the accuracy of the residual life prediction results in the early and middle stages of the degradation process.(4)Present a residual life prediction method based on a hybrid neural network model that considers the effects of multiple operating conditions.To strengthen the performance,such as the accuracy and robustness of the residual life prediction results,a hybrid model based on stacked denoising autoencoder and bidirectional long short-term memory neural network is proposed,which addresses the problem of complex modeling processes and violent fluctuations in monitoring data under dynamic operating conditions.The hidden features of the performance parameter monitoring data after operating condition partitioning and normalization are extracted using the stacked denoising autoencoder to achieve data dimensionality reduction and fusion.The fusion data is then used as input for the bidirectional long short-term memory neural network model to achieve residual life prediction for aero-engines under multiple operating conditions. |
