Fault Diagnosis Under Variable Operating Conditions for Rotating Machinery Based on MKPCA and Domain Adaptive DBN

Under varying operating conditions, the operating state of rotating equipment is influenced by multiple factors, presenting the characteristics of complex and changeable data distribution and high nonlinearity. Deep belief networks (DBN) possess powerful feature learning capabilities, making them suitable for fault diagnosis under such complex conditions. However, traditional DBNS have limitations such as structural design relying on experience and being prone to getting stuck in local optima. At the same time, the adaptability to the differences in distribution of data under different working conditions is insufficient, resulting in limited accuracy of fault diagnosis and model generalization ability. To cope with these issues, a fault diagnosis method based on improved DBN is presented in this paper. Firstly, a multiple kernel principal component analysis (MKPCA) is proposed. The radial basis function(RBF) kernel and the polynomial function kernel are combined, and the weights of the kernel functions are dynamically adjusted according to different characteristics to improve the processing ability of KPCA for complex data. Secondly, the Particle Swarm Optimization algorithm (PSO) is introduced to improve the network structure and parameters of the DBN. Then, the shortcomings of traditional DBN structure design that relies on experience and is prone to fall into local optimum is overcame, achieving adaptive optimization of network structure and parameters. In addition, considering the differences in the distribution of data under different working conditions, this paper combines semi-supervised domain adaptation with DBN and introduces the multiple kernel maximum mean difference (MK-MMD), forcing DBN to learn general features, reducing the distribution differences between domains, and transferring the knowledge learned in the source domain to the target domain, thus solving the problem of insufficient generalization ability of the model caused by changes in working conditions. Finally, through experimental verification, the results indicate that the method introduced in this paper significantly enhances the accuracy and generalization ability of fault diagnosis of rotating equipment under varying operating conditions.