Explainable Process Monitoring Based on Class Activation Map: Garbage In, Garbage Out

Abstract

Process monitoring at industrial sites contributes to system stability by detecting and diagnosing unexpected changes in a system. Today, as the infrastructure of industrial sites is advanced due to the development of communication technology, various and vast amounts of data are generated, and the importance of a methodology to effectively monitor these data to diagnose a system is increasing daily. As a deep neural network-based methodology can effectively extract information from a large amount of data, methods have been proposed to monitor processes using this methodology to detect any system abnormalities. Neural network-based process monitoring is effective in detecting anomalies but has difficulty in diagnosing due to the limitations of the black-box model. Therefore, this paper proposes a process monitoring framework that can detect and diagnose anomalies. The proposed framework performs post-processing based on the class activation map to perform the diagnosis of data that are considered outliers. To verify the performance of the proposed method, experiments were conducted using industrial public motor datasets, demonstrating that the proposed method can effectively detect and diagnose abnormalities. © 2020, Springer Nature Switzerland AG.