Research On Engine Fault Diagnosis And Performance Prognosis

Engine is a kind of frequently used power machinery with complex mechanical structures and motion coupling modes.After long-term operation,many problems occur,such as physical aging of mechanical components,lack of lubrication,mechanical friction,high temperature overheating,overload and so on.These problems could easily cause frequent faults of engine sub-components and performance degradation of the whole machine,thus affecting the normal operation.Typically,cylinder misfire fault can intensify engines vibration,reduce output torque,worsen the exhaust emission,even cause irreversible damage to the engine.Consequently,the economic and environmental costs of related industrial production activities are increased dramatically.Based on above analyses,it is of great significance to research on engine faults diagnosis and performance prognosis.Aiming at engine typical faults and performance degradation process,this paper studies effective diagnostic methods and prognostics.By analyzing cylinder misfire faults,bearing components surface damage and performance degradation process,researching health state monitoring and evaluation methods.Specifically,signal acquisition experiments and data set validation tests are carried out to study methods for feature extraction,fault diagnosis,state recognition,and performance prediction.Meanwhile,the mapping relationship between engine-related sensing signals and component fault categories,health states of the whole machine.Consequently,engine fault diagnosis and performance degradation assessment can be realized successfully.The main research contents include:(1)For the misfire of engine cylinder components,a diagnostic method based on synchrosqueezing wavelet transform and XGBoost is proposed to solve the problem of low resolution in feature extraction and over-fitting in classification of existing diagnosis methods.Vibrational signals of a cylinder head are acquired through experiments for testing the performance of misfire in diesel engine at different speeds.Features of signals are extracted by using time-domain statistic methods and the synchrosqueezing wavelet transform.Then,the locally linear embedding method is adopted to reduce the dimensions of features.The XGBoost algorithm is finally applied to evaluate misfire situations.Comparative experimental results on different working conditions and evaluation methods show that the proposed method has outstanding generalization performance,robustness and ability for restraining model over-fitting.(2)For rolling bearings,the key components of engine rotating subsystem,a method for identify surface damage health state based on DNN with optimized structural parameters is proposed.Vibration monitoring signals of the rolling bearing are obtained,the shallow features are extracted by time-domain and frequency-domain statistical analysis.On this basis,stacked auto-encoders with optimized parameters are used to extract deep sensitive features,train and test model.The results show that the proposed method can achieve high recognition accuracy and avoid the complex manual feature extraction process.(3)Aiming at the irreversible phenomenon of engine performance degradation,performance prognosis of the whole machinery is studied.Based on data fusion theory and Bayesian model,a remaining useful life prognostic method is proposed.Multi-dimensional monitoring sensor data of the engine are selected and fused.Then,the health indexes and degradation state model of the engine are established.On this basis,the Bayesian estimation model for parameters of the state model is established.Finally,the parameters of the model are updated by using Markov chain Monte Carlo random sampling method,and the remaining useful life of the testing engines is predicted.The C-MAPSS turbofan engine simulation data set verifies the validity of the prognostic method.(4)Based on the above researches,the platform of intelligent diagnosis and performance prognosis for the engine is preliminarily developed,which can effectively realize the functions of signal data upload,data visualization,algorithm evaluation and result analysis.