基于ART和Yu范数的聚类方法在齿轮故障诊断中的应用

针对传统聚类方法需预先指定类别个数而导致应用受限的问题,提出一种基于ART和Yu范数的聚类方法,可自适应地确定类别个数。通过对齿轮无标记故障样本的诊断分析对该方法进行验证。从多个角度提取反映故障信息的特征参数集,利用距离区分技术对其进行优选,并结合ART的机制和基于Yu范数的聚类技术,对齿轮故障类别进行诊断分析,并与Fuzzy ART方法的诊断结果进行比较。结果表明,该方法可以有效地对齿轮故障进行区分,且效果优于Fuzzy ART方法。

Application of clustering method based on ART and Yu norm to gears fault diagnosis

As the traditional clustering method needs to determine the number of classes in advance, a novel clustering method based on adaptive resonance theory (ART) and Yu norm that can self-adapt to determine the number of classes is proposed and validated by the diagnostic analysis of unlabeled faulty samples of gears. A feature parameter set that presents the fault-related information is extracted from different symptom domains, and some optimal features are selected by the distance discriminant technique. Having combined the merits of ART and Yu norm-based clustering method, the proposed clustering model is employed to diagnose the fault conditions of gears and found to be able to effectively classify the faulty samples of gears, having better diagnosis performance than the fuzzy ART.