A Novel Fault Diagnosis Method for Acceleration Sensor Utilizing IEM-Based LLE and WKELM

Abstract

In this study, fault diagnosis method for acceleration sensor utilizing information entropy measurement-based LLE and weighted kernel extreme learning machine (IEMLLE-WKELM) is proposed for fault diagnosis for acceleration sensor. First of all, this article proposes an information entropy measurement-based locally linear embedding (IEMLLE) algorithm to reduce the features of acceleration sensor data. The IEMLLE algorithm is a dimensionality reduction algorithm based on information entropy measurement. The discrimination of the distribution of sample data of the different classes based on IEMLLE is higher than that based on locally linear embedding (LLE) algorithm. Moreover, this article proposes a weighted kernel extreme learning machine (WKELM) algorithm, among which the use of kernel functions instead of hidden layer random feature maps containing activation functions is beneficial for improving the nonlinear processing ability and robustness of weighted extreme learning machine, and the chaos particle swarm optimization (CPSO) algorithm is proposed to optimize the penalty factor and the kernel parameter of weighted kernel extreme learning machine. The experimental results show that IEMLLE-WKELM is the higher fault diagnosis accuracy for acceleration sensor than LLE-WKELM, LLE-ELM, and principal component analysis-extreme learning machine (PCA-ELM).