Evaluation Of Rolling Bearing Performance Degradation By Combining Stacked Denoising Auto Encoder With Broad Learning

In modern production machinery,rolling bearings are the key parts to determine the mechanical health.Real-time monitoring and performance degradation evaluation of rolling bearings is of great significance,and timely evaluation and judgment of bearing status is very important.Aiming at the problem that the complex structure of traditional deep learning cannot realize the rapid fault diagnosis and performance degradation evaluation of bearings,this paper discusses how to use the emerging broad learning to realize the intelligent fault diagnosis and performance degradation evaluation of bearings based on data drive,and verifies the feasibility of the proposed method through experiments.The main research contents of this paper include:(1)Vibration signal analysis of early bearing failure points.Firstly,the characteristics of bearing degradation signal were analyzed by ensemble empirical mode decomposition.Then,the characteristic set sensitive to bearing degradation state was preliminarily screened by constructing characteristic evaluation index.Finally,multi-optimal minimum entropy deconvolution and Hilbert envelope were used to demodulate,and the envelope spectrum of early fault points of bearings was obtained,so as to judge the early fault points of rolling bearings.(2)A fault diagnosis model of rolling bearing based on Feature engineering combined with fuzzy broad learning system(FE-FBLS)was established.Firstly,considering the influence of features on the diagnosis model,a feature engineering method was designed to extract the features of bearing vibration signals.Then,the feature engineering was combined with the fuzzy broad learning system.After parameter training of the diagnostic model,the experimental data of three groups of different bearings were used to verify the model.Finally,through comparative analysis with other methods,the proposed FE-FBLS model can diagnose bearing faults in the three data sets within 0.3 seconds,and the accuracy rates are 96.43%,100%and 100%,respectively,which proves the robustness and feasibility of the proposed model.(3)A rolling bearing performance degradation evaluation model based on stack denoising auto-encoder and fuzzy broad learning system(SDAE-FBLS)was established.Firstly,the multi-domain features of bearing life data were extracted,and the extracted features were used as training features to train the stack noise reduction auto-encoder.Then,the feature signals reconstructed by the stack denoising auto-encoder are input into the mapping feature layer of fuzzy broad learning to form a performance degradation evaluation model with both width and depth.Finally,the degradation state of bearings was judged by the set performance degradation degree index,and the performance degradation of bearings was evaluated by using the lifetime bearing data.The experimental results showed that the constructed model could judge the early failure point of the selected three groups of bearings as the 533 point,the 1716 point and the 453 point.