Research On Fault Diagnosis Method Of Rolling Bearings Based On Multi-source Domain Heterogeneous Model

As a key component in rotating machinery equipment,rolling bearings are widely used in industrial production,and once a failure may lead to serious economic losses and even casualties.Due to their complex specifications and working conditions,the working environment is harsh and difficult to disassemble,resulting in insufficient labeled data.At the same time,the vibration data distribution of rolling bearings under different specifications and working conditions is quite different.And in practical applications,there are often multiple similar rolling bearing data set resources.Therefore,it is of great significance to effectively use multiple source domain vibration data to realize the condition identification of rolling bearings with different specifications and working conditions.Taking deep learning and multi-source domain transfer learning as the core technology,and introducing an emerging meta-learning,a rolling bearing fault diagnosis method based on multi-source domain heterogeneous model transfer is proposed to realize the fault status identification of rolling bearings under different specifications and working conditions.To solve the problem of the sparse labeled data under certain specifications and working conditions,and the large difference in the distribution of rolling bearing vibration data under different specifications and different working conditions,which is insufficient to train an effective fault diagnosis model,an improved heterogeneous model transfer learning method is proposed.Fourier transform is used to obtain the time-frequency spectra of the rolling bearing vibration signals.A variety of labeled data under different specifications and conditions are used as the multi-source domain,and a small amount of labeled data in other specifications and conditions are used as the target domain.The multiple source domains data are used to train Res Net-34 deep networks,SAdam optimizer is introduced to speed up the network convergence speed,the parameter transfer strategy of heterogeneous model is improved by using evolution strategies model agnostic meta learning,so that it can adaptively determine the level and content of knowledge transferred to the target domain.Addressing the problem of the resources of multiple similar data sets are not fully utilized,It is proposed to transfer the source domain knowledge to the VGG-16 deep networks to obtain multiple target domain models,the extracted features are input into the same extreme learning machine in turn to achieve feature fusion,the classification results can be obtained through the extreme learning machine,and finally rolling bearing fault diagnosis model can be established.Experimental results show that the proposed method can achieve the transfer diagnosis between rolling bearings under different specifications and working conditions,and has a higher accuracy.