Fan mill state estimation based on acoustic signature analysis

Two new, real time, noninvasive techniques for the detection of states of impact plates in fan mills at thermal power plants, using acoustic transducers as sensor elements, are proposed in this paper. Both methods rely on analysis of recorded acoustic signals in the time and frequency domains. One method uses a linear dimension reduction procedure and the state of the impellers is assessed by analyzing statistical distance as a metric. The second method uses a subtractive clustering technique to determine the cluster centers in multidimensional space and introduces the Euclidean distance ratio as a metric to estimate the amount of wear of the impellers. These data-driven methods are tested on real acoustic signals recorded at the thermal power plant TEKO Kostolac A1 in Serbia and shown to be effective in an extremely noisy environment. A comparison of the methods is made bearing in mind the efficiency and computational complexity of the algorithms.