Novel approach for industrial process anomaly detection based on process mining

Abstract

Anomaly detection plays a critical role in ensuring the quality and safety of industrial processes. Process mining, as an emerging technology, has proven effective in extracting knowledge and process rules inherent in process events. However, industrial time series data possess characteristics such as high noise, and data redundancy, posing challenges for accurately assessing system anomalies using traditional data-driven methods. To address these challenges, this paper proposes a Trend-Value Fusion Process Mining (TVPM) approach based on the representation of industrial process events to tackle anomaly detection in industrial processes. Firstly, TVPM introduces a fusion measure that combines data trends and mean-value to provide a comprehensive description of operating events in industrial processes. This fusion measure serves as a vital tool for capturing the complex details and features of industrial events. Additionally, a novel process discovery algorithm is developed to extract behavioral patterns and sequence information from events. Finally, a process abnormal state detection model is constructed by leveraging the comprehensive and in-depth information obtained from TVPM. The proposed method is applied to an industrial coal gasification process, yielding satisfactory results.