MECHANICAL FAULT DIAGNOSIS BASED ON ADAPTIVE NEIGHBORHOOD PRESERVING ELM-AE

In order to solve the problems of prior knowledge dependence and insufficient data mining in machine learning fault diagnosis, a local preserving extreme learning machine automatic encoder based on adaptive neighborhood is proposed. Euclidean distance penalty factor was introduced into the original data space and the embedded representation space for paired samples to realize the similarity classification of data samples. A unified objective function was proposed, which could simultaneously learn data representation and correlation matrix, and a soft discriminative constraint was proposed to prevent overfitting. The experimental results show that the fusion learning association matrix and data representation method has the advantages of fast learning speed, strong generalization ability and high diagnostic accuracy.