A signal based approach for condition monitoring and predictive maintenance of a capsule filler machine

The need to increase manufacturing systems productivity and reduce their downtimes has led researchers to investigate and develop Predictive Maintenance practices. One of the major challenges to cope with Predictive Maintenance is related with the health assessment and analytics (i.e.: diagnostic and prognostic methods): the identification of the incoming faults is difficult and hard to deploy in complex automated machinery operating in real life conditions. At state of the art, established approaches are based on locating specific sensors as near as possible to the potential failure. However, such approach is expensive and often hard to realize due to machine topologies. The present work deals with a fault detection signal based approach, which analyses the vibrations of a complete pharmaceutical capsule filler machine and detects the signature of a fault on a critical stage to build a pattern threshold for Predictive Maintenance. The main novelty and strength of the proposed engineering method is that the detection can be achieved despite the sensor position and in presence of many sources, as it is in real life industrial environments. An industrial case study, on a pharmaceutical capsule filler is presented