Given the growing electricity demand, modern power plants'' operation must be ever more efficient and reliable. The steam boiler, which converts thermal energy into electricity, is one of the most significant thermal power plants (TPP). Approximately 60% of boiler outages result from a boiler tube failure; such failure can significantly affect the safe and economical operation of thermal power facilities. Early detection and prediction of boiler tube leakage can help schedule shutdowns and reduce maintenance and labor costs.
In a thermal power plant, waterwall tube leakage is one of the most frequent causes of tube failure in TPPs. Corrosion, erosion, and fatigue are the general phenomena that cause a decrease in tube wall thickness, which ultimately leads to explosion and leakage in the event of failure. Therefore, in the last decade, numerous attempts have been made to detect boiler waterwall tube leakage. The idea of e-maintenance based on multi-variate algorithms was recently introduced for intelligent fault detection and diagnosis in TPPs. E-maintenance mostly relies on standard process control variables (sensors data) for leak detection and localization. Generally, an overwhelming amount of data is collected in a power plant, which makes data processing difficult; it also contains redundant and irrelevant information due to highly localized, redundant sensors. This may affect the performance of multi-variate algorithms as these algorithms are highly dependent on the number of input sensors. This creates a need to develop an accurate and precise method of determining the optimal sensor arrangement for detection of boiler tube leakage. Therefore, this work proposes a machine learning-based model integrated with an optimal sensor selection scheme to analyze boiler waterwall tube leakage.
In this study, a machine learning-based model integrated with an optimal sensor selection scheme is introduced for the application of boiler waterwall tube leakage detection. In the first part, optimal sensor selection is performed via correlation analysis to select the most sensitive sensors necessary to detect waterwall tube leakage. In the second part, different supervised machine learning algorithms are utilized for boiler waterwall tube leakage detection. In the end, two data scenarios (TPP real plant operating data and Physics-based simulated data) are employed to validate the proposed model''s effectiveness. The results indicate that the proposed model can successfully detect waterwall tube leakage with improved accuracy vs. other comparable models.