BP神经网络与Sugeno模糊积分融合 的转子系统故障诊断

目的针对BP神经网络对转子故障诊断方法存在的局限性,提出一种融合Sugeno模糊积分和BP神经网络的转子故障轴心轨迹识别诊断方法。方法首先利用轴心轨迹图像的不变矩为特征向量,提取常见旋转机械转子故障特征,随后利用多个BP神经网络对故障类型进行识别,最终采用Sugeno模糊积分对BP神经网络识别结果进行决策,从而构建转子故障诊断模型,并应用于转子系统故障的诊断。结果通过机械故障仿真模拟实验平台采集了6种常见转子系统故障信号,利用matlab2012a软件编程建模仿真处理,试验表明,该模型有效地提高了转子系统多类别故障的识别正确率。同时,该方法对同一故障类型识别所需样本少,大大节省了数据获取和处理的时间。结论该方法提出并用于转子系统故障诊断中,诊断准确性高,可靠性强,利用样本数据量少,节约时间,对小样本数据的故障诊断有着良好的效果。

Fault Diagnosis of Rotor System Based on BP Neural Network and Sugeno Fuzzy Integral

Objective To propose a diagnosis method of rotor fault axis trajectory recognition based on Sugeno fuzzy integral and BP neural network to address the limitations of BP neural network for rotor fault diagnosis. Methods Firstly, the invariant moment of the axial trajectory image was used as the feature vector to extract the fault characteristics of common rotating machinery rotors. Then, multiple BP neural networks were used to identify the fault types. Finally, the Sugeno fuzzy integral was used to make decisions on the BP neural network recognition results, construct a rotor fault diagnosis model and apply it to diagnose faults of the rotor system. Results The six common rotor system fault signals were collected by the mechanical fault simulation platform. The matlab2012a software was used for programming, modeling and simulation. The test showed that the model effectively improved the recognition accuracy of multi-class faults in the rotor system. At the same time, the method required fewer samples for identifying the same fault type, which greatly saved data acquisition and processing time. Conclusion This method is proposed for fault diagnosis of rotor system. It has high diagnostic accuracy and high reliability. It requires less sample data and saves time. It has a good effect on fault diagnosis of small sample data.