Composite Fault Diagnosis for Rotating Machinery of Large Units Based on Evidence Theory and Multi-Information Fusion

Xavier Chiementin,Xiao Li,Naiquang Su,Zhiqiang Huo,Qinghua Zhang

doi:10.1155/2019/1982317

Abstract

Due to the complexity of the structure and process of large-scale petrochemical equipment, different fault characteristics are mixed and present multiple couplings and ambiguities, leading to the difficulty in identifying composite faults in rotating machinery. This paper proposes a composite faults diagnosis method for rotating machinery of the large unit based on evidence theory and multi-information fusion. The evidence theory and multi-information fusion method mainly deal with multisource information and conflict information, synthesize multiple uncertain information, and obtain synthetic information from multiple data sources. To detect faults in rotating machinery, the dimensionless index ranges of composite faults are first used to form a feature set as the reference. Then, a two-sample distribution test is applied to compare the known fault samples with the tested fault samples, and the maximum statistical distance is used. Finally, the multiple maximum statistical distances are fused by evidence theory and identifying fault types based on the fusion result. The proposed method was applied to the large petrochemical unit simulation experiment system, the results of which showed that our proposed method could accurately identify composite faults and provide maintenance guidance for composite fault diagnosis.

Highlights

Rotating machinery works in complex environments and has difficulty in separating the signal of faults in industrial petrochemical plants, thereby complicating the fault diagnosis decision [1,2,3]
Once the large unit presents problems, it needs to be entirely stopped for inspection, which can result in a huge economic loss. erefore, it is essential to quickly identify the fault signal and predict the fault types
Dimensionless indexes refer to the ratio of two quantities with the same dimensions. e basic idea behind the dimensionless algorithm is to achieve the eliminating dimension of the two-dimensional ratios that are based on the probability density function, so the dimensionless indexes are not affected by the frequency and amplitude of the mechanical signal in the fault diagnosis [38, 39]. e dimensionless algorithm is defined as follows: ζx

Summary

Introduction

Rotating machinery works in complex environments and has difficulty in separating the signal of faults in industrial petrochemical plants, thereby complicating the fault diagnosis decision [1,2,3]. The scope of the dimensionless indexes of normal equipment and fault equipment is difficult to distinguish, which makes the decision more difficult To solve these problems, Xiong et al [22] proposed a genetic programming method based on dimensionless indexes in the time domain, which has achieved positive results in rotating machinery classification. (i) Two-sample distribution of the known fault samples is compared with the tested fault samples, and the maximum statistical distance (ii) Multi-dimensionless information is fused, and the fault types can be identified according to fused results (iii) e method we proposed is verified with a large petrochemical unit simulation experiment system and is shown to effectively improve fault identification e rest of this paper is organized as follows: Section 2 describes the process of fault diagnosis and theoretical basis: dimensionless algorithm, two-sample distribution test, and multi-information fusion.

Proposed Method

Validation Experiment

Findings

Discussion of Experimental Results