A study on predictive maintenance using edge intelligence

Abstract

Production systems have been following a reactive type of approach for maintenance activities, but in recent times, the interest has shifted towards the development of more proactive approaches to avoid failures. This has resulted in increased indus trial data collection through the deployment of information and communication technologies. With this advent, there is also an increase in Predictive Maintenance (PdM) solutions. Developing Machine Learning (ML) models to perform PdM ac tivities have become a recent challenge and a popular research area under Industry 4.0. Therefore, this thesis takes its roots from this emerging area of data-driven de cision making for PdM with the support of key enabling technologies of Industry 4.0. This thesis aims to analyze high-dimensional data coming from Edge devices and investigate what type of predictive algorithms can be designed to implement PdM at the industrial level. Data collected by the Edge device from two Manufactur ing companies were analyzed and different ML models were developed for both cases. Further, a comparative study between the cases was presented to show the importance of faulty data to develop a better ML model for PdM. Finally, some recommendations were provided for the successful implementation of the ML model in both companies by using the advantages of the Edge device. This provides a concrete platform for future research in the area of handling missing data and im plementation of Edge Intelligence in a manufacturing setup.