为了实现轴承故障智能诊断, 对基于信息融合的机器人薄壁轴承故障智能诊断方法进行研究。首先, 采用声发射和振动传感器, 搭建了机器人薄壁轴承试验与多信息数据采集系统; 然后, 以薄壁单列角接触球轴承ZR71820为对象, 在轴承外圈、内圈和滚动体上分别制作点蚀、裂纹缺陷, 用正交试验法采集不同缺陷类型、不同当量载荷及不同转速状态下薄壁轴承在试验过程中的声发射和振动信号; 最后, 选取时域中均方根值和峭度指数及频域中均方根频率作为振动、声发射信号的特征参数, 分别进行了基于单一振动、声发射信号的薄壁轴承故障诊断, 并采用SOM与BP神经网络将试验过程中的振动和声发射信号的特征信息进行融合, 研究了基于信息融合的机器人薄壁轴承故障智能诊断技术。结果表明: 基于振动信号故障诊断的正确率为85.7%; 基于声发射信号故障诊断的正确率为81.0%; 基于BP神经网络信息融合故障诊断的正确率为93.5%; 基于SOM神经网络信息融合故障诊断的正确率为95.2%。基于SOM神经网络信息融合的薄壁轴承故障智能诊断比单用振动或声发射信号的诊断正确率分别高出9.5%和14.2%, 比用BP神经网络信息融合故障诊断的正确率高1.7%。

To realize the intelligent diagnosis of bearing faults, an intelligent fault diagnosis method for the thin-wall bearing of a robot based on information fusion was studied. First, a test and multi-information data acquisition system of the thin-wall bearing of a robot was built by acquiring acoustic emission and vibration acceleration signals. Then, data from acoustic emission and vibration acceleration signals detected during the test of thin-wall bearing under different fault types, equivalent loads, and rotational speeds were obtained using an orthogonal experimental method. A thin-wall single-row angular contact ball bearing (ZR71820) was used as the research object, and pitting and micro-crack defects were produced on the bearing outer ring, inner ring, and rolling bod. Finally, the root mean square value and kurtosis index in the time domain, as well as the root mean square frequency in the frequency domain, were selected as the characteristic parameters of the vibration and acoustic emission signals. Fault diagnosis of thin-wall bearings based on single vibration or acoustic emission signals was conducted. In addition, an intelligent fault diagnosis of thin-wall bearings were researched based on the fusion characteristics of acoustic emission and vibration acceleration signals using Self-Organization feature Map (SOM) and Back-Propagation (BP) neural networks. Experimental results indicate that the accuracies of fault diagnoses based on vibration signals, acoustic emission signals, and BP and SOM neural network information fusion are 85.7%, 81.0%, 93.5%, and 95.2%, respectively. The accuracy of intelligent fault diagnosis based on SOM neural network information fusion of the thin-wall bearing is 9.5%, 14.2%, and 1.7% higher than that of single vibration, acoustic emission signals, and BP neural network information fusion, respectively.