模糊函数图像与概率神经网络在柴油机气阀故障诊断中的应用

Application of Ambiguity Function Images and Probabilistic Neural Networksto Fault Diagnosis of Diesel Valve Train

本文将柴油机缸盖表面振动信号的模糊函数结果在频偏 时延相平面上用灰度图表示出来,得到一系列模糊函数图像。对此图像进行归一化处理,降低维数,再采用概率神经网络对模糊函数图像进行分类,从而将气阀机构的故障诊断转换为模糊函数图像的分类识别。试验结果表明,利用模糊函数图像和概率神经可以取得很好的诊断结果,识别正确率可达95%,当训练样本比较充足时,识别正确率可达100%。

Ambiguity functions of the vibration acceleration signals, which were acquired from a cylinder head, were calculated and then expressed in a series of grey images. These images were classified into 8 kinds with probabilistic neural networks (PNN) after they were normalized to a low dimension. Then the process of fault diagnosis of valve train was changed to the classification of ambiguity function images. The experimental results show that a very high rate of correct recognition (about 95%) can be obtained by using ambiguity function images and PNN. When the training samples are abundant enough, the recognition correct rate can even be as high as 100 %.