Due to continuous high dynamic load of industrial robots, rotate vector (RV) reducers may be more prone to wear and pitting and there will be abnormal vibration and sound caused by incipient faults. However, many experienced experts can identify the health state of machines by the sensitivity of their hearing to some special sounds. To simulate human hearing and visualize the sound heard by experts when RV reducer is running, Mel-filter bank is designed and Mel-spectrograms are introduced to show the sound energy distribution of RV reducers under various conditions. A spliced feature is constructed to describe the texture and boundary of the energy distribution. Moreover, a sound-vibration fusion approach is proposed to fuse the spliced features of sound and vibration for accurately identifying of RV reducer health states. Single-joint RV reducers experiments are performed to obtain sound and vibration data sets under different health states. The identification results of 50 runs show that the average accuracy of the proposed sound-vibration spectrogram fusion method is 93.83 %, greatly preferable to traditional methods. It is demonstrated that the proposed method is useful for robot condition monitoring and fault identification in industrial field.

中文翻译：

---

工业机器人RV减速器声振谱融合诊断方法

由于工业机器人持续高动态负载，旋转矢量（RV）减速器可能更容易出现磨损和点蚀，并且会出现由早期故障引起的异常振动和声音。然而，许多经验丰富的专家可以通过听觉对某些特殊声音的敏感度来识别机器的健康状态。为了模拟人类听觉，可视化专家在RV减速器运行时听到的声音，设计了梅尔滤波器组，并引入梅尔频谱图来显示RV减速器在各种条件下的声能分布。构造拼接特征来描述能量分布的纹理和边界。此外，提出了一种声音-振动融合方法，融合声音和振动的拼接特征，以准确识别 RV 减速器的健康状态。进行单关节RV减速器实验，获得不同健康状态下的声音和振动数据集。 50次运行的识别结果表明，所提出的声振谱融合方法的平均准确率为93.83%，大大优于传统方法。实验结果表明，该方法对于工业领域机器人状态监测和故障识别具有一定的实用价值。