dClink: A data-driven based clinkering prediction framework with automatic feature selection capability in 500 MW coal-fired boilers

The coal-fired boiler (CFB) often suffers from corrosion, slagging, fouling, and clinkering; among them, clinkering is one of the critical problems, which is addressed in very few literature. Most of the existing research does not focus on clinkering prediction for 500 MW CFBs, and the features are selected based on the operators' experience. Hence, to overcome human errors and improve the safety and reliability of thermal power plants, we propose a data-driven-based predictive maintenance method, dClink, that uses automated feature selection for clinkering prediction from real-time process parameters of a 500 MW CFB. An automatic filter-wrapper -based feature selection model is proposed to select and rank the most important process parameters for clinkering. Root-cause analysis of clinkering is also performed so that the vital parameters can be controlled to prevent clinkering. The Support Vector Machine detected clinkering-fault with 99.8% accuracy. Further, the designed Decision Tree Regression model predicted clinkering with Root Mean Squared Error of 0.0103 and Mean Absolute Error of 0.0107 during the testing phase. The method has been validated using randomly selected data from the given dataset. It has been proved that the proposed method can detect and accurately predict clinkering to prevent accidents, low efficiency, and financial loss.