Anomaly Scoring Model for Diagnosis on Machine Condition and Health Management

Abstract

The reliability of manufacturing equipment is critical for ensuring the productivity and energy efficiency of a manufacturing facility. An unexpected machine breakdown may lead to unexpected downtime, disruption of manufacturing schedule, lower production efficiency, higher operation and maintenance cost. The recent development in machine learning and artificial intelligence enables data-driven Predictive Maintenance (PdM) by means of perceiving the dynamics of manufacturing systems and abstracting them into learnable features to provide a better interpretation of machine failures or unplanned downtimes. PdM, often translated to Prognostics and Health Management (PHM), aims to continue the optimal/normal operation of manufacturing systems. Often, vibration is used as a proxy of an early indicator of impending failure. In this study, tri-axial acceleration data collected from the two different machines are utilized. PdM-based strategies for machine condition monitoring and smart scheduling of equipment maintenance using an anomaly scoring model are discussed for two critical elements in a manufacturing system: 1) Chiller 2) Compressor. An anomaly scoring model is developed to extract meaningful information from the vibration data.