Fault diagnosis method of self-validating metal oxide semiconductor gas sensor based on t-distribution stochastic neighbor embedding and random forest

Peng Xu; Kai Song; Yinsheng Chen; Guo Wei; Qi Wang

doi:10.1063/1.5090142

Fault diagnosis method of self-validating metal oxide semiconductor gas sensor based on t-distribution stochastic neighbor embedding and random forest

Rev Sci Instrum. 2019 May;90(5):055002. doi: 10.1063/1.5090142.

Authors

Peng Xu¹, Kai Song¹, Yinsheng Chen², Guo Wei¹, Qi Wang¹

Affiliations

¹ Department of Automatic Testing and Control, Harbin Institute of Technology, Harbin 150001, People's Republic of China.
² Department of Measurement and Control Technology and Instrument, Harbin University of Science and Technology, Harbin 150001, People's Republic of China.

PMID: 31153286
DOI: 10.1063/1.5090142

Abstract

The metal oxide semiconductor (MOS) gas sensor plays an important role in the machine olfactory system, and the accuracy of the measured value affects the performance of the system. Because of the material characteristics of MOS gas sensors, the sensors are prone to be faulty under the condition of long-time working. Therefore, it is necessary to identify the faults of MOS gas sensors online to improve the maintainability and reliability of the machine olfactory system during the measuring process. The self-validating technology can improve the reliability of sensors. Combining with self-validating sensor technology, a fault diagnosis method for the MOS gas sensor based on t-distribution Stochastic Neighbor Embedding (t-SNE) and random forest (RF) is proposed in this article. The trailing effect of t-SNE is used to enhance the separability of the extracted fault features, and the fault feature set is utilized to construct a RF multifault classifier. To verify the effectiveness of the proposed method, a gas monitoring experimental system is designed and developed to obtain sufficient experimental samples and compose simulation data of different fault types. The simulation experimental result shows that compared with the other methods, the proposed method has higher fault diagnosis accuracy, which is up to 99.78%.