Harmony theory yields robust machine fault-diagnostic systems based on learning vector quantization classifiers

This contribution describes an algorithm to improve the ability of a learning vector quantization (LVQ) classifier in machine fault diagnosis. By adding a harmony model to the LVQ classifier the proposed method can construct an input-output mapping based on human knowledge and stipulated input-output vector pairs. Knowledge atoms from harmony theory are used to encode the knowledge of various machine fault patterns by capturing the probability distributions of input features during the training process. Therefore, the class boundaries of various fault patterns are made more distinguishable, and the capability of classification is enhanced. Moreover the summation of all the deviations generated from the input vectors and weights during the classification process can be better discriminated; therefore, the chance of misclassification caused by a few dominant distorted features is reduced. This proposed approach has been tested an classifying various faults obtained from a tapping machine, against other popular neural-network-based classifiers. The results from a series of experiments have demonstrated that this hybrid approach is promising, and particularly useful in classifying input features inherent with overlapping distributions and high uncertainty in the class boundaries. Copyright (C) 1996 Elsevier Science Ltd