A Fault Diagnosis and Reconfiguration Strategy for Self-Validating Hydrogen Sensor Array Based on MWPCA and ELM

• Main Authors: Kai Song, Peng Xu, Yinsheng Chen, Tinghao Zhang, Guo Wei, Qi Wang
• Format: Article
• Language: English
• Published: IEEE 2019-01-01
• Series: IEEE Access
• Subjects: Fault diagnosis; reconfiguration; hydrogen sensor; extreme learning machine; moving windows principle component analysis
• Online Access: https://ieeexplore.ieee.org/document/8805059/

The flammable and explosive property of hydrogen is the main danger in its safe use, storage and transportation. In this paper, a novel hydrogen monitoring system is designed based on the principle of semiconductor, catalytic combustion and heat-conducting gas sensors. Also, the gas sensor will inevitably fail due to the nature of gas sensitive materials in the long-time monitoring process. To ensure the accuracy and reliability of hydrogen concentration measurement, a novel fault diagnosis and reconfiguration strategy for hydrogen sensor array based on moving window principle component analysis and extreme learning machine (MWPCA-ELM) is proposed. Firstly, online multiple faults detection is carried out by using MWPCA. Once one or multiple faults are detected, the measured values of other fault-free sensors will be used to recover the faulty data in real-time by using ELM predictor according to the relevancy among the hydrogen sensors. Secondly, the hydrogen concentration is reconfigured seamlessly and accurately based on ELM under the condition of small calibration data sample. Finally, fault diagnosis is conducted by MWPCA feature extraction coupled with ELM multi-classifier. In order to illustrate the effectiveness and feasibility of the proposed fault diagnosis and reconfiguration strategy, a hydrogen concentration monitoring experimental system was established. The average relative error (ARE) of hydrogen concentration estimation is declined from 1.18% to 0.82% compared with the traditional regression methods. Particularly, the proposed fault reconfiguration model can recover the fault data even if the concentration is changed, and the accuracy of fault diagnosis is 100% within 250 samples.