神经网络模糊聚类方法在故障诊断中的应用

在旋转机械故障诊断研究中，对大量样本的有效分类至关重要。传统的贴近度法计算繁锁，分辨率低，且领带于大量实验测试和经验案例的总结。根据故障诊断中样本数据无须精确要求的特点，可以对样本数据作模糊化处理，进一步采用自组织竞争神经网络对样本进行自动聚类，速度快，准确性好，具有较高的智能特性。实际应用表明，这种模糊神经网络完全满足使用要求。

A study of fuzzy cluster of samples by neural network in fault dianosis

In the research of fault diagnosis of the rotary machine, it is very important to collect large sums of samples rapidly and accurately. In the past, apt degree is generally used to classify samples, which is slow, not accurate enough and depends on great quantities of practical data. Two methods have been adopted to improve the process of classification. One is to fuzzify sample data, the other is to make use of the excellent automatic cluster ability of Kohonen self organized mapping neural network. Experiments show that the improved method satisfies results.