Research On Remaining Useful Life Prediction And Fault Diagnosis Of Aircraft Engines Based On Data-driven Model

An aircraft-engine is a very delicate and complex mechanical system with a wide range of failures.In order to detect abnormalities in an aircraft engine,it is necessary to predict the remaining useful life(RUL)and determine the type of failure based on the engine’s operating condition data,then make maintenance and repair plans based on the predicted information to ensure the safe and reliable operation of the aircraft.With the rise of big data and computer technology,RUL prediction and fault diagnosis modelling methods for aircraft-engines,as the core component of Prognostics Health Management(PHM),are also being expanded.In order to further improve the fault prediction accuracy of airfraft-engine,this thesis will adopt a data-driven approach to investigate the following issues:(1)To address the problems that the current aircraft-engine fault diagnosis model relies on a priori knowledge and cannot handle nonlinear data,and the uncertain parameters of Support Vector Machine(SVM)lead to fluctuations in model classification accuracy,this thesis proposes aircraft-engine fault diagnosis model based on a data-driven approach.The model introduces an improved particle swarm algorithm in SVM to increase the probability of SVM finding the best parameters,and combines with Sparse Auto Encoder(SAE)for dimensionality reduction of nonlinear data to achieve fault diagnosis of aircraft-engine.The validation experiments were conducted on the aircraftengine fault dataset,and the results showed that the fault diagnosis accuracy of this algorithm reached 97.01%.(2)In the field of RUL prediction,most current prediction methods have the problems of complex structure,poor generalization and insufficient extraction of multiscale information,a predication model in this thesis combining Multi-Head SelfAttention(MSA)with Bi-directional Long and Short Term Memory Network(BLSTM)for aircraft-engine RUL prediction model is proposed.This model improves on the basis of Self-Attention,extends the Self-Attention subspace,and performs an average weighting operation on the scores of key features of a single self-attention space in the form of multiple heads,which can effectively reduce the errors generated by a single SelfAttention in the process of weight assignment,and then combines BLSTM to mine the nonlinear relationship of time series data to form a linear mapping relationship,therefore achieved the RUL prediction of the aircraft-engine.After experimental verification,the model proposed in this paper achieves different degrees of improvement in prediction accuracy under both evaluation metrics.