Early fault detection and isolation in coal mills based on self-organizing maps

Abstract

Classical approaches to the fault detection and isolation usually require extensive plant-modeling and statistical analysis of the measured signals and their residuals versus the developed model. In this paper, alternative simple model-free approach is proposed. Real-time data are preprocessed and self-organizing map is trained and used for the reliable isolation of the most frequent mill fault — output fuel-mixture drop due to the coal-stuck in the input bunker. Proposed approach is successfully verified on the real-time data-sets from the coal mills in thermal power plant “Nikola Tesla B”, Serbia.