一种基于多尺度多流形投影的故障检测方法  

A fault detection method based on multiscale multi-manifold projections

作　　者：郭金玉 赵素华 李元 GUO Jinyu;ZAHO Suhua;LI Yuan(College of Information Engineering,Shenyang University of Chemical Technology,Shenyang,Liaoning 110142,China)

机构地区：[1]沈阳化工大学信息工程学院,辽宁沈阳110142

出　　处：《燕山大学学报》2025年第2期155-166,188,共13页Journal of Yanshan University

基　　金：国家自然科学基金资助项目(62273242);辽宁省教育厅科学研究资助项目(JYTMS20231516)。

摘　　要：针对工业过程数据具有高维度、多尺度和非平稳性等特点,本文提出一种基于多尺度多流形投影(multiscale multi-manifold projections,MSMMP)的故障检测方法。该方法在传统多流形投影(multi-manifold projections,MMP)算法的基础上引入多尺度分析。首先,利用小波变换将过程数据分解为不同的深度;其次,在每个深度上分别建立MMP模型,并使用均方误差对模型进行评判,选择最佳分解深度;最后,建立具有最佳分解深度的MSMMP模型,通过对比统计量和控制限进行故障检测。将该故障检测方法应用于连续搅拌釜式反应器和田纳西-伊斯曼(Tennessee Eastman,TE)过程,验证该方法的可行性和有效性。MSMMP解决了传统算法在处理高维数据时的困境,可保持数据的结构信息完整,有效地捕获数据的多尺度特征,提高了降维后数据的表达能力。A fault detection method based on multiscale multi-manifold projections(MSMMP) is proposed in this paper for industrial process data with high dimensionality,multiscale and non-stationary characteristics.The method introduces multiscale analysis to the traditional multi-manifold projections(MMP) algorithm.First,the process data are decomposed into different scales using wavelet transforms.Second,the MMP model is built at each scale and the model is judged using the mean squared error to select the best decomposition scale.Finally,a MSMMP model with an optimal decomposition scale is developed and fault detection is performed by comparing the statistics and control limits.The fault detection method is applied to a continuous stirred tank reactor and the Tennessee Eastman(TE) process for simulation analysis to verify the feasibility and effectiveness of the method.MSMMP solves the dilemma of traditional algorithms in dealing with high-dimensional data,keeps the structural information of the data completely and captures the multiscale features of the data effectively,and improves the expressive ability of the data after dimensionality reduction.

关 键 词：多流形投影 小波变换 多尺度分析 故障检测 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]