基于特征选择与概率神经网络的轴承故障诊断研究

Study on the Bearing Fault Diagnosis based on Feature Selection and Probabilistic Neural Network

为提高航空发动机轴承故障诊断精度,应用距离评估准则和概率神经网络分类技术,提出了一种基于特征选择与概率神经网络的轴承故障诊断方法。首先,利用轴承故障试验数据,提取得到14个时域特征和13个频域特征,构成故障诊断多域特征集;其次,为提高分类效率,降低各特征量间的耦合特性对分类结果的影响,应用基于距离评估的特征选择方法,筛选得到分类性能更好的特征参数;在此基础上,应用概率神经网络方法进行了轴承故障诊断研究。应用轴承模拟故障实验数据进行验证,结果表明,与BP神经网络和支持向量机诊断方法相比,PNN方法诊断精度更高;同时由于采用了特征选择,诊断效率和精度又得到进一步提高。

To improve the aero - engine fault diagnosis accuracy grade, by using the DET and PNN classi- fication techniques, a bearing fault diagnosis technique based on feature selection and PNN is put forward. Firstly, the bearing fault test data are extracted to form the multi -domain fault diagnosis feature set composed of 14 time -domain features and 13 frequency -domain features. Secondly, to increase classification efficiency and reduce the influence on classification result from coupling characters between features, the feature selection technique based on DET is applied to obtain feature parameters which can be classified easily. On this basis, the PNN technique is applied to carry on research of bearing fault diagnosis. The bearing simulation fault experiment data is applied for verification, the results prove that compared with diagnosis techniques of BP neural network and support vector machines, the PNN is higher in the respect of diagnosis accuracy grade. Meanwhile, the efficiency and accuracy grade of diagnosis are further improved for the reason of employing feature selection technique.