文献类型：期刊文献

中文题名：基于故障相关变量因果关系分析的工业过程故障根因诊断

英文题名：Fault's Root Causes Diagnosis of the Industrial Process Based on the Causality Analysis of Fault Related Variables

作者：孙丽华[1];王文静[1];朱宏涛[1];姜庆超[1]

机构：[1]华东理工大学信息科学与工程学院

年份：2024

卷号：51

期号：6

起止页码：1035

中文期刊名：化工自动化及仪表

外文期刊名：Control and Instruments in Chemical Industry

收录：CSTPCD

基金：国家自然科学基金(批准号:62322309)资助的课题。

语种：中文

中文关键词：故障诊断;根因分析;故障相关变量;可解释相关性分析

外文关键词：fault diagnosis;root cause analysis;fault correlation variable;interpretable correlation analysis

摘要：提出了一种数据驱动的故障根因分析方法,旨在通过数据处理和可解释相关性分析,准确识别系统故障的根本原因。首先,采用基于神经网络的变量挑选方法,通过深度学习技术自动挑选出与故障相关的关键变量,以提高数据的有效性。随后,运用沙普利加法解释模型对挑选出的变量进行可解释相关性分析,分析各个变量对故障发生时系统的贡献度。接着,引入偏交叉映射方法进行因果关系分析,探索变量之间的因果关系,找出导致故障的关键因素,定位和识别导致故障发生的根本原因变量。最后在两个案例应用中证明了该方法的适用性。
In this paper,a data-driven root cause analysis method for system faults was proposed to accurately identify the root causes of system faults through data processing and interpretive correlation analysis.In which,through making use of deep learning techniques,having a neural network-based variable selection method employed to automatically select key variables related to faults so as to enhance the efficacy of the data;and then,having the Shapley additive model used to interpretively analyze correlation of the variables selected and examine their contributions to the system during faults occur.In addition,the partial crossmapping method was introduced to causality analysis and the causal relationship among variables was investigated,including identifying the root causes which incurring the faults.This method's application in two cases proves its applicability.