Rolling bearing fault warning method based on adversarial autoencoder and adaptive threshold
To cope with the difficulties of fault warning in current engineering practice,such as the challeng-es of constructing sensitive feature combinations,scarcity of complete fault samples,and inaccurate warning threshold settings,etc.,a rolling bearing fault warning method based on Adversarial Autoencoder(AAE)and adaptive threshold was proposed.Firstly,the preprocessed normal sample spectrum data was utilized as AAE training data for autoencoder network and adversarial network training,and the autoencoder reconstruction er-ror was calculated and the coding network was retained;Then,the low-dimensional features obeying the prior distribution was extracted layer by layer using the encoder,and the health indicator was constructed by com-bining the reconstruction error and similarity measure,and the probability density distribution of the health indicator was fitted based on the beta distribution to determine the threshold adaptively;Finally,the test data was processed by the same steps and compared with the threshold to discriminate abnormalities.The pro-posed method was verified by using two types of rolling bearing datasets,and the experimental results show that the proposed method has excellent fault warning performance and adaptability,and can realize early warning of weak fault.
rolling bearingfault warningadversarial autoencoderhealth indicatorbeta distributionadaptive threshold
万方数据
维普
