Application of Over-Sampling Techniques and Fuzzy ARTMAP to Condition Monitoring of a Power Generation System

Condition monitoring is a process of assessing the health status of a system, process, or machine. Monitoring and identifying any potential fault can be conducted by leveraging measurements from the installed sensors that provide information on the state of the system. In this respect, machine learning models are useful for processing and analyzing the sensor data for fault detection. However, the imbalanced nature of these sensory data can cause misleading high accuracy scores. In this study, we employ an over-sampling method to tackle the imbalanced class problem. Specifically, both Synthetic Minority Over-sampling Technique (SMOTE) and Gaussian SMOTE are used to generate minority class samples. The balanced data set is used by the Fuzzy ARTMAP (FAM) model for fault classification. The effectiveness of the developed method is evaluated using a real-world circulating water system in a power generation plant. The results indicate that both SMOTE variants can improve the performance of FAM in detecting faults corresponding to operating conditions of the circulating water system for efficient power generation.