Asset health management utilizing batch multivariate pattern analysis

Abstract

Budgetary and performance demands have led many power utility businesses to focus on some form of centralized fleet monitoring. Advanced pattern recognition (APR) is a common tool used for monitoring the major assets involved in the production of electricity in this manner. While pattern recognition techniques can focus on real-time steady-state operation and ease the challenge of monitoring hundreds or thousands of pieces of equipment, many assets can incur damage during the start-up and shutdown transient conditions that are not as commonly watched. This damage leads to failure over time, but signs of the damage may not be apparent during steady-state operation. A multivariate pattern analysis was designed to identify anomalies specific to start-up and shutdown data of these assets, along with any other batch production process with defined start and end parameters. This method can be used across all types of processes and equipment. An early warning case study was conducted with a major power utility to validate the technique on a forced outage caused by a steam turbine bearing failure. Data was provided for two units with multiple coast downs over several months. This case study examines normal operation for monitored variables, such as temperatures and vibration readings, with the goal of detecting an individual variable’s deviation from acceptable conditions and identifying the overall impact each has on the process. The study shows how this technique may be used to increase the early detection of faults and can be used in conjunction with other APR and diagnostic tools to help a business improve overall asset health management.