Detection of events causing pluggage of a coal-fired boller: A data mining approach

This paper presents an approach to analyze events leading to pluggage of a boiler. The proposed approach involves statistics, data partitioning, parameter reduction, and data mining. Two independent data mining algorithms have been applied to detect both static and dynamic relationships among the process parameters. The multi-angle data mining approach increases the ability to locate rare events as well as the reliability of the results. The proposed approach has been tested on a 750 MW commercial coal-fired boiler affected with an ash fouling condition that leads to boiler pluggage, thus resulting in unscheduled shutdowns. The cause of the boiler pluggage is not known. The rare event detection method presented in the paper identifies several critical time-based data segments that are indicative of the boiler pluggage. The events define a set of general guidelines that when followed should reduce the likelihood of boiler pluggage. The knowledge extracted by the data mining algorithm is an important component of an intelligent alarm system.