An Overview of Useful Data and Analyzing Techniques for Improved Multivariate Diagnostics and Prognostics in Condition-Based Maintenance

Abstract

The reliability of production machines gains in importance in today’s optimized and highly productive business environments. Unexpected machine breakdowns do not only lead to loss of production time and production outages but also to diminishing customer satisfaction due to deterioration in quality and declining availability of products. The condition-based maintenance (CBM) strategy aims at preventing these machine breakdowns through real-time monitoring of machine conditions. Sensor data are collected and analyzed using diagnostic and prognostic approaches to identify the type of fault and the remaining useful life. Identifying the reasons and time of breakdowns fosters improved planning of maintenance and spare parts demand, leading to higher machine reliability. In general, machine sensor data are regarded as a useful source of information to assess the machine’s operating condition. However, in some specific cases, the machine sensors lack the ability to correctly represent the health of the machine or the specific component under consideration. Therefore, additional information by further available data is required to improve diagnostic and prognostic techniques for more accurate and precise analysis. Current research focuses on the analysis of sensor data for condition-based maintenance, while other data like the operating history and environment temperature have only been considered to a limited extend so far. Hence, this paper gives an overview on potential data for machine health assessment and remaining useful life prediction in condition-based maintenance. Furthermore, corresponding approaches and techniques for fault diagnostics and prognostics are presented targeting the analysis of individual data sources as well as of multivariate settings featuring multiple integrated data sources.