A belief rule-based evidence updating method for industrial alarm system design. (December 2018)

• Record Type: Journal Article
• Authors: Xu, Xiaobin; Xu, Haiyang; Wen, Chenglin; Li, Jianning; Hou, Pingzhi; Zhang, Jing
• Is Part Of: Control engineering practice. Volume 81(2018)
• Page Start: 73
• Page End: 84
• Publication Date: 2018-12
• Subjects: Alarm system design -- Uncertainty processing -- Evidence theory -- Belief rule base -- Data-driven design; Automatic control -- Periodicals; 629.89
• DOI: 10.1016/j.conengprac.2018.09.001
• Languages: English
• ISSNs: 0967-0661

Abstract

Abstract: This paper presents a belief rule-based evidence updating method for industrial alarm system design, concentrating on handling uncertainties of process variable. Firstly, Sigmoid function-type thresholds are designed to transform the sampled value of a process variable to the corresponding alarm evidence with the form of belief degrees about "Alarm" and "No-alarm". Secondly, a linear updating strategy of evidence is introduced to combine the current alarm evidence with historical evidence such that the fused evidence can provide more accurate alarm decision support. In the process of evidence updating, the belief rule inference is used to determine the combined weights of the current and historical evidence by modeling the reliability degree data of alarm evidence. The proposed method adopts the knowledge and data-driven idea without knowing the precise probabilistic characteristics of the monitored process variable. Hence, in industrial practice it may be more available and flexible than the traditional probability-based design methods. Finally, a typical numerical experiment and an industrial case show the proposed method has better comprehensive performance than some typical probability-based methods, binary classifiers, and the original evidence updating methods. Highlights: A novel S-type threshold is presented to transform a process variable to the alarm evidence. The characteristic of lossless information transformation is proved. Data-driven belief rule inference is proposed to enhance the accuracy of the evidence-based alarm decision. The effectiveness of the method is demonstrated by experiment and industrial case.