Data-driven Model Based Online Fault Detection Using OMP-ERR
- Resource Type
- Conference
- Authors
- Zhou, Guangze; Luo, Zhong; Zhu, Yunpeng; Gao, Yi; Wang, Zhiao
- Source
- 2022 4th International Conference on Industrial Artificial Intelligence (IAI) Industrial Artificial Intelligence (IAI), 2022 4th International Conference on. :1-6 Aug, 2022
- Subject
- Communication, Networking and Broadcast Technologies
Computing and Processing
Engineering Profession
Power, Energy and Industry Applications
Robotics and Control Systems
Signal Processing and Analysis
Transportation
Fault diagnosis
Employee welfare
Fault detection
Computational modeling
Matching pursuit algorithms
Feature extraction
Data models
NARX model
System identification
Online fault detection
OLS algorithm
OMP algorithm
- Language
Model based online fault detection often conducted by extracting features from models driven by system input and output data under various working conditions. The efficiency of online system modelling is therefore significant to improve the performance of online fault detections. In this study, a novel fast data-driven modelling approach, known as the OMP (Orthogonal Matching Pursuit)- ERR (Error Reduction Ratio) method is proposed to improve the efficiency of online fault detections. The new system identification method is motivated by noticing that the traditional OMP algorithm is much faster but usually less accurate than the OLS (Orthogonal least squares) algorithm in the identification of system NARX (Nonlinear Auto-Regressive with Exogenous inputs) models. The problem is first illustrated by the identification of a Single Degree of Freedom (SDoF) system. After that, the OMP-ERR algorithm is developed to improve the NARX modelling efficiency for the purpose of system model-based online fault detections. A case study on the crack detection of a cantilever beam shows that the new approach is over 10 times faster than the traditional OLS modelling process, demonstrating the promising applications of the new approach in online fault detections in engineering practice.