基于多维高斯贝叶斯的机械设备失效/故障智能诊断及参数影响分析

Smart Failure/Fault Diagnosis and Influence Analysis for Mechanical Equipment with Multivariate Gaussian Bayesian Method

采用多维高斯贝叶斯机器学习方法研究机械设备的失效/故障智能诊断方法,基于极大似然分析基础上,提出一种基于“马氏距离”估算的关键特征参量影响分析方法,用于评价各特征参量对失效/故障分类的影响。该方法用于两个机械设备的失效/故障数据库的智能诊断分析,在获得高诊断识别率(失效/故障模式正确识别率分别达到96%和86%)的同时,也识别了影响失效/故障分类的关键特征参量。分析表明特定的失效/故障模式往往取决于少数关键特征参量,而不确定的失效/故障模式的关键特征参量往往呈分散分布,关键特征的分散性会影响多维高斯贝叶斯分类器的诊断识别率。该方法可用于机械设备的失效/故障的智能识别与关键特征参量的智能诊断,也为失效/故障的影响因素分析指明方向。

The failure/fault of mechanical equipment is diagnosed with multivariate Gaussian Bayesian classifier.Based on the maximum likelihood methodology,a novel influence analysis model based on"Mahalanobis distance"estimation is proposed.The method is then applied to two datasets for the mechanical equipment failure/fault mode identification.The results show that the proposed method obtains high diagnostic recognition rate(failure/fault mode recognition rate in two cases are 96%and 86%,respectively),as well as principal attributes that contribute to specific failure/fault modes.It is found that the specific failure/fault mode mainly depends on a few characteristic parameters,while the unspecified failure/fault mode always involves several diverse characteristic parameters.The desperation of key parameters will cause the unsatisfactory result of multivariate Gaussian Bayesian classifier.The model proposed in this paper is helpful for the intelligent diagnose of failure/fault mode of mechanical equipment and the analysis of key parameters that contribute to specific failure/fault mode,and so as to provide guidance for failure/fault reasoning.