An Investigation of Using Manifold Learning To Diagnose Machinery Faults

• 1、School of Mechanical Engineering, Xi\'an Jiaotong University
• 2、School of Science, Xi\'an Jiaotong University

Abstract：Manifold learning can effectively find out the intrinsic characteristics of high-dimensional data sets and is a promising method which can be employed to identify and analyze the fault mechanism of a complex mechatronic system. In this paper, we investigate the feasibility of applying manifold learning to fault diagnosis deeply and discuss how the characteristics of a signal sampling system will influence the performance of a manifold learning technique. Based on the theoretical analysis and simulated experiments, the following conclusions can be drawn. When the characteristics of the signal sampling system maintain relatively stable, manifold learning can tolerate the existence of the systemic nonlinearity and zero-offset to a certain extent. In order to make a manifold learning algorithm achieve good performance, it requires that the collected data should be easily reconstructed into a high-dimensional space and the dissimilarity between them can be measured.

Keywords： Fault diagnosis Manifold learning Isomap algorithm Signal sampling system