A Time-aware Data Clustering Approach to Predictive Maintenance of a Pharmaceutical Industrial Plant

Abstract

Predictive maintenance is one of the most active fields of study for Industry 4.0, as it is expected to significantly decrease the maintenance costs of the equipment. Often, it is not possible to accurately predict the deterioration of a component, as the reliability of predictive models strongly depends on the available sensory data and on the specific characteristics of the monitored component. In this paper, we present a clustering-based approach with the aim of predicting the time-aware evolution of the health status of a machine component in a pharmaceutical plant. The developed strategy allows to obtain a time segmentation of the component’s operational points, which are then clustered using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN). In particular, this approach has the advantage of being general and making use of a limited amount of features extracted from a single sensor signal. The proposed approach becomes attractive when the quantity of single sensory collected data is not sufficient to build a physical model capable of identifying changes in the system status.