Machine Failure Diagnosis by Combining Software Log and Sensor Data

Abstract

Many studies have been conducted in the manufacturing industry to support the cause analysis and early recovery of production line shutdowns caused by machine failures. However, methods such as simple anomaly detection are not effective against large machines with complex behavior. In this study, we propose a method for such machines to show the estimated causes of failure by combining log text files and sensor data, which record software behavior and hardware status, respectively. The proposed method is twice as accurate as methods with only software logs or sensor data, and achieves explainability of the results.