Abstract: The matching accuracy of the revolute pair is directly affected by its matching clearance.Considering the energy loss caused by collision in the operation of the revolute pair with clearance,a thermal image monitoring method was proposed for clearance of revolute pair with words frequency features.To effectively obtain the abnormal heat distribution,the feature thermal image extraction method of revolute pair with clearance based on the Euclidean-channel separation was proposed.Words frequency matrix of feature thermal image was extracted by constructing the bag model of visual words,and words frequency feature of words frequency matrix was extracted using multi-layer perceptron.The convolutional neural network embedded with attention mechanism was used to extract the feature from feature thermal imaging of the revolute pair with clearance.Through integrating the words frequency feature into the thermal image feature,a clearance monitoring method based on feature fusion was proposed,which could overcome the problem of low accuracy of clearance monitoring using single thermal image feature.Finally,a revolute pair clearance fault simulated experiment platform was built to verify the effectiveness of the revolute pair clearance monitoring method based on fusion features.

Key words: fusion feature, feature thermal image, revolute pair clearance, words frequency feature, convolutional neural network

摘要： 转动副的配合间隙直接影响着其配合精度,考虑到含间隙的转动副运行时会由于碰撞产生能量损耗,提出一种融合词频特征的转动副热成像间隙监测方法。为了有效获取异常热量分布,提出基于欧式-通道分离的含间隙转动副特征热图像提取方法。通过构建视觉词袋模型提取特征热图像的词频矩阵,基于多层感知机提取词频矩阵的词频特征。采用嵌入注意力机制的卷积神经网络对含间隙转动副的特征热成像提取特征。将词频特征融入热图像特征,提出基于特征融合的转动副间隙监测方法,克服采用单一热图像特征进行间隙监测准确率不高的问题。最后搭建转动副间隙故障模拟实验台,验证了融合特征的转动副间隙监测方法的有效性。

关键词: 融合特征, 特征热图像, 转动副间隙, 词频特征, 卷积神经网络