基于多域特征及多传感器网络融合的滚动轴承故障诊断

Fault Diagnosis for Rolling Bearings Based on Multi-Domain Features and Multi-Sensor Networks Fusion

将幅值域无量纲参数和时频域信息熵作为概率神经网络的特征向量,构建多传感器系统概率神经网络的初级诊断网络,并利用概率神经网络累加层输出结果构建Dempster-Shafer证据理论的mass函数,通过Dempster-Shafer证据理论进行决策级融合诊断。将该方法用于滚动轴承故障模式分类,并通过实验室及现场实例验证了该方法的可行性与有效性。

The dimensionless parameters of amplitude domain and information entropy in time - frequency domain are taken as feature vector of probabilistic neural network, and the primary diagnosis network of multi - sensor system probabilistic neural networks is constructed. The mass functions of D - S evidence theory are built using output of accumulation layer of probabilistic neural network. The fusion diagnosis of decision level is carried out by D - S evidence theory. The method is applied to fault pattern classification for rolling bearing, and the feasibility and effectiveness of method are verified through examples of laboratory and worksite.