RMINE: A Rough Set Based Data Mining Prototype for the Reasoning of Incomplete Data in Condition-based Fault Diagnosis

Abstract

Condition-based fault diagnosis aims at identifying the root cause of a system malfunction from vast amount of condition-based monitoring information and knowledge. The needs of extracting knowledge from vast amount of information have spurred the interest in data mining, which can be categorized into two stages data preparation and knowledge extraction. It has been established that most of the current data mining approaches to fault diagnosis focus on the latter stage. In reality, condition-based monitoring data may, most of the time, contain incomplete, or missing data, which have an adverse effect on the knowledge or diagnostic rules extracted. Several approaches to deal with missing data can be found in literature. Unfortunately, all of which have serious drawbacks. In this paper, a novel approach to the treatment of incomplete data is proposed for the data preparation stage, while a rough set based approach has been developed to pre-process the data for the extraction of diagnostic rules. The two-stage data mining technique is implemented into a prototype system, RMINE, which also possesses a self-learning ability to cope with the changing condition-based data. A real industrial case study of a pump system is used to demonstrate the fault diagnosis process using RMINE. The result has shown the potential of RMINE as a general data mining prototype to condition-based fault diagnosis.