A NEW METHOD OF INTELLIGENT CONDITION EVALUATIONFOR MECHANICAL EQUIPMENT

摘要

In order to solve the problem of implementing intelligent condition evaluation for mechanical equipment(ICEME) with insufficient abnormal condition samples, a new method of ICEME, support vector data description(SVDD), is proposed. With this method, the abnormal condition (outlier objects) can be distinguished from normalcondition (target objects) when only the information of the normal condition class is available but nothing is knownabout the abnormal condition class. Applying this method to ICEME, we can correctly evaluate machine condition onlyusing normal condition signals. Compared with Fuzzy Mathematics, Expert System, Artificial Neural Network andGenetic Algorithm based traditional methods, the SVDD based method is unnecessary to preprocess the signals toextract their features and to obtain prior knowledge. The experimental results show that SVDD based method hasexcellent evaluation and clustering ability, and it represents a new approach to ICEME.