Adaboost algorithm in the frame of predictive maintenance tasks

The paper proposes a method for state change detection in rotary machines. The procedure relies on recorded signals spectrogram analysis and features appropriate for texture classification in digital images. Final classification in terms of machine health is performed using Adaboost algorithm. The application of this algorithm is proposed for thermal power plant fan mills whose impact plates are damaged during the coal grinding process. The main goal of the procedure is to detect the amount of wear of the impellers and to determine the appropriate timing for unnecessary maintenance tasks. Real acoustic signals for verification of this algorithm are recorded in thermal power plant Kostolac A1 in a coal grinding subsystem.