Study on the machine-learning based system for detecting abnormal pressure drops in hydraulic press machines

Due to declining working-age populations in some countries, manufacturing and production sites are increasingly leveraging digital technologies to boost efficiency and labor productivity. In response to this trend, we have developed a system that swiftly assesses the operational status of machinery to optimize production efficiency within manufacturing companies, thus shortening the time needed for machine inspections and repairs. Our system, a machine learning-based approach to failure detection, specifically targets pressure drops in hydraulic press machines. We installed vibrational acceleration sensors on the cylinders—the press machine’s primary components—and collected continuous signal data. By modeling normal operations using standard deviation, crest factor, and maximum signal values, we can detect deviations and temporal changes in the data that indicate failures and anomalies. This allows for the proactive prediction and monitoring of potential failures.