Enabling predictive maintenance in high-speed manufacturing machines through critical parameter identification with sensor data

Abstract

Predictive maintenance is a proactive maintenance process based on the permanent monitoring and evaluation of machine data, using techniques and models from the fields of data science, machine learning and mechanical engineering. The aim is to increase equipment reliability and availability, minimize maintenance costs, avoid unexpected downtime and costly repairs, resulting in higher operational efficiency and productivity. This study analyses a high-speed, high-throughput machine producing paper and board products under critical environmental conditions such as temperature and humidity. There, short machine stops have a enormous impact on overall equipment effectiveness (OEE). However, the reasons for the machine downtimes are unclear. Hence, the critical parameters of the process were investigated to determine the relationship between machine stops and sensor data.

In addition, three machine learning models were trained and evaluated with the data collected from the sensors.

Through data analysis, it was found that the vacuum signals and torque peaks were responsible for the short machine stops. These, in turn, were interfering with systems that apply fluids. These results are the beginning of a roadmap for predictive maintenance of high-speed manufacturing machines

Keywords: Machine Learning, Predictive Maintenance, Sensor Signals, High-Speed Manufacturing, Industry 4.0