International Journal of Advanced Research in Computer and Communication Engineering

Abstract: Predictive Maintenance (PM) refers to the utilization of various forms of data for the timely anticipation of system failures. The objective is to schedule maintenance, instead of performing it according to a fixed pattern or after failure, thus maximizing uptime and resource efficiency. Modern PM systems are becoming popular within smart manufacturing as they contribute to automation and operational efficiency improvement, particularly when deployed near the equipment that generates and suffers the data. These models harness diverse forms of sensor data generated by the manufacturing process, and include techniques such as survival, reliability and hazard function estimation; time series analysis; and statistical, machine learning and deep learning methodologies. Such techniques are able to recognise and model ‘normal’ machine behaviour when sensors are not broken, and thus may fail to anticipate faults that lead to abnormal physical behaviours detectable by sensors. Such issues are of growing concern within smart factories, where maintenance modelling needs to remain accurate even when the operating environment or underlying machine behaviour changes. PM model deployment and operationalisation can thus be challenging, and requires effective instrumentation, data engineering and model management around these techniques, especially when real-time, low-latency inference at the edge is necessary.
Despite the challenges, there is considerable ongoing research work applying predictive maintenance solutions in production environments. Promising demonstrations have been reported across multiple domains, including automotive, semi-conductor, mining, electronics, and food processing. Use cases span A- and B- lines of automotive assembly, anticipation of failures in cooling units of A-lines, cooling balance modelling for wafer manufacturing and plasma etching, board cleaning process deviation alerting, run-to-failure estimation in semiconductor and electronic assembly, electric drive system on-condition servicing planning, web break prediction for textile manufacturing, and applying predictive and prescriptive analytics to food processing.

Keywords: Predictive maintenance,Smart manufacturing,Industrial Internet of Things (IIoT),Machine learning models,Condition-based monitoring,Equipment failure prediction,Time-series analysis,Sensor data analytics,Remaining useful life (RUL),Anomaly detection,Digital twin technology,Edge computing,Big data analytics,Fault diagnosis,Industry 4.0.

Downloads: | DOI: 10.17148/IJARCCE.2022.111261