基于特征提取的皮革裁剪机轴承故障诊断

Fault Diagnosis of Leather Cutting Machine Bearings Based on Feature Extraction

针对典型的皮革裁剪机轴承故障问题,提出一种基于改进ResNet网络的故障诊断方法。该方法在传统的ResNet网络中添加多尺度特征提取模块与CBAM模块,以提高对轴承故障特征的提取和诊断能力。结果表明,基于改进ResNet网络的皮革裁剪机械轴承故障诊断随着迭代次数的增加,在训练集上的分类精度不断提高,分类损失值不断降低,最后趋近于0;改进的ResNet网络在测试集上表现较好,未出现过拟合现象,性能优越且具有稳定性。在消融试验中,改进ResNet网络相较于原始ResNet网络、仅使用CBAM注意力模块改进的ResNet网络,分类准确率分别高出526%和059%,说明改进方法有效提升了网络性能。本改进ResNet网络在10折交叉验证法中,训练与测试的准确率始终保持在985%以上,相较于DNN、PreCNN、未改进ResNet与AlexNet等热门分类网络,测试准确率分别提高了4284%、1439%、1132%和396%。由此得出,本方法对多种类型的轴承故障诊断都较为准确,具有一定的实用性。

A fault diagnosis method based on improved ResNet network was proposed for typical leather cutting ma-chine bearing faults This method adds multi-scale feature extraction modules and CBAM modules to the traditional ResNet network to improve the ability to extract and diagnose bearing fault features The results show that with the in-crease of iteration times,the classification accuracy on the training set of leather cutting machinery bearing fault diag-nosis based on improved ResNet network continuously improves,the classification loss value continuously decreases,and finally approaches 0 The improved ResNet network performs well on the test set without overfitting,with superior performance and stability In the ablation experiment,the improved ResNet network showed a classification accuracy increase of 526%and 059%compared to the original ResNet network and the ResNet network improved only with CBAM attention module,indicating that the improved method effectively improved network performance In the 10 fold cross validation method,the accuracy of training and testing of this improved ResNet network remained above 985%Compared with popular classification networks such as DNN,PreCNN,unimproved ResNet,and AlexNet,the testing accuracy increased by 4284%,1439%,1132%,and 396%,respectively From this,it can be con-cluded that this method is accurate for diagnosing various types of bearing faults and has certain practicality.