Fault Relevant Variable Selection for Fault Diagnosis

Ruixiang Deng,Yunpeng Fan,Zhenbang Wang

doi:10.1109/access.2020.2970046

Abstract

In process monitoring, fault relevant variable selection and fault diagnosis are two important branches. But they are often discussed independently and scarcely integrated in research. To integrate them, a novel fault diagnosis algorithm based on fault relevant variable selection is proposed in this paper. The main contents are summarized as follows. For fault relevant variable selection, from normal state to fault state, the relative changes between variables and statistics are analyzed using least absolute shrinkage and selection operator (LASSO). In order to determine the optimal set of fault relevant variable, a fault reconstruction algorithm based on least angle regression (LARS) is proposed. The set of relevant variables is constantly updated until there is no abnormality after reconstruction. Finally, a monitoring strategy based on fault subspace is proposed. It can detect fault effectively and provide useful information for fault diagnosis. The effectiveness of the proposed algorithm is illustrated by some experiments.

Highlights

For improving the safety of process and the quality of product, the process monitoring technology has become a necessary portion in modern industrial processes
Yan and Yao proposed a faulty variable selection method based on least absolute shrinkage and selection operator (LASSO) [29]
The main purpose of this study is to develop an effective fault diagnosis algorithm based on fault relevant variable selection

Summary

INTRODUCTION

For improving the safety of process and the quality of product, the process monitoring technology has become a necessary portion in modern industrial processes. Owing to the great progress and extensive applications of sensor and computer technology, abundant operational information is collected in the running time This has effectively promoted the development of data-based methods for process monitoring. Yan and Yao proposed a faulty variable selection method based on least absolute shrinkage and selection operator (LASSO) [29] They did not use the historical faulty data in the modeling. The main purpose of this study is to develop an effective fault diagnosis algorithm based on fault relevant variable selection. To obtain the appropriate set of fault relevant variable, a fault-reconstruction method based on least angle regression (LARS) is proposed.

FAULT RELEVANT VARIABLE SELECTION

CASE STUDIES

Findings

CONCLUSION