Nonlinear machine fault detection by semi-supervised Laplacian Eigenmaps

Abstract

A semi-supervised Laplacian Eigenmaps algorithm for machine fault detection is proposed. The purpose of the algorithm is to efficiently extract the manifold geometric characteristics of nonlinear vibration signal samples, and to determine fault classification of operating equipment so that the accuracy of fault detection can be improved. The data acquisition and pre-processing of the vibration signal is firstly implemented from monitoring equipment, then hybrid domain feature is obtained, and the initial sample set can be built. This is followed by implementing the semi-supervised Laplacian Eigenmaps algorithm so that the sensitive nature characteristics of manifold can be obtained from the device initial sample set. In order to establish the intelligent diagnostic model, the Least square Support vector machine (LS-SVM) is then adopted, which fault diagnosis and decisions can be achieved in the feature space of the low-dimensional manifold. The experiment results of using the IRIS data, gearbox and compressor fault data show the proposed method has more advantage when compared with the PCA and Laplacian Eigenmaps on improving the accuracy of fault detection.