A time-series probabilistic preventive maintenance strategy based on multi-class equipment condition indicators

Abstract

This paper focuses on the condition-based maintenance of a high gravity reactor, which is used to process the emissions from the production process in petrochemical industries. Three indicators, i.e. the current, temperature, and amplitude, are monitored together to indicate whether the equipment has failed. Based on these indicators, this paper focuses on the cumulative probability of equipment operating indicators corresponding to the empirical distribution function and calls it the p-value. To improve the prediction accuracy, a time series preventive maintenance strategy based on the appropriate p-value is designed to determine whether each indicator exceeds the threshold. Then, considering various indicators comprehensively, a method aiming to reduce and minimize the costs is proposed. To test the robustness of the method, the results of the prediction accuracy and the cost reduction level based on different threshold values are given. The empirical application shows that the method can judge whether each indicator exceeds the threshold under a threshold change. Moreover, as the threshold increases, the advantages of the probabilistic approach are more obvious. Compared with the conventional strategy, the probabilistic approach can greatly reduce the costs.

Keywords:

• High gravity reactor
• condition-based maintenance
• time series
• prediction
• cost reduction
• data-driven

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by the NSFC [grant numbers 71872033, 71831003], the 2020 LiaoNing Revitalization Talents Program [grant number XLYC2007061], the LiaoNing Education Department [grant number LN2020Q13], and the Research Project of Dongbei University of Finance and Economics [grant number 20200090].