论文部分内容阅读
针对采用SOM网络进行多故障诊断时,要求多故障模式相似且不包含标准故障输出的限制,提出将SOM网络与可拓理论相结合的多故障诊断方法。首先采用SOM网络对训练样本进行聚类,得到故障模式及其聚类中心。然后针对每种故障模式的每个特征构造在聚类中心处取得最大值的关联函数,并以各特征的关联函数值为基础,设计多故障评价指标实现多故障诊断。最后采用汽轮发电机组振动信号的频谱数据对算法进行验证,结果表明该方法能够正确识别待诊断样本的单故障和多故障模式,具有可行性。
For multi-fault diagnosis based on SOM network, multi-fault diagnosis method is proposed, which is similar to multi-fault mode and does not contain the limitation of standard fault output. Firstly, the SOM network is used to cluster the training samples to get the failure modes and their clustering centers. Then, for each feature of each failure mode, the correlation function which maximizes at the cluster center is constructed. Based on the correlation function values of each feature, the multi-fault evaluation index is designed to realize multi-fault diagnosis. At last, the algorithm is validated by using the spectrum data of the vibration signal of the steam turbine generator. The results show that this method can identify the single fault and multi-fault mode of the sample to be diagnosed correctly, which is feasible.