Fault diagnosis of belt weigher using the improved DENCLUE and SVM

Abstract:

A method of on-line fault detection and diagnosis based on the modified DENCLUE clustering and partial binary tree support vector machine (SVM) is proposed for on-line fault diagnosis problem of bulk weighing equipment—electronic belt weigher. Firstly, in view of the fault data varying with equipment flow, a modified DENCLUE clustering algorithm is designed to realize the online fault detection by isolating the fault data after the clustering analysis of the real-time data. Secondly, the density estimation method in DENCLUE algorithm is introduced into the support vector machine, and then an improved BTSVM, in which the separability measure and binary tree structure is built based on the similar density within class and between class, is presented to recognize the detected fault on-line. The improved BTSVM is also verified the superiority by the standard dataset. Finally, the proposed online fault detection and diagnosis model is verified more suitable for the online fault detection and diagnosis of bulk weighing equipment by the array belt weigher experiments.