基于迁移学习的钢丝绳断丝定量检测方法

A quantitative testing method for broken wires in steel wire ropes based on transfer learning

针对目前钢丝绳断丝定量检测的问题,利用深度卷积神经网络强大的特征提取能力,提出一种基于迁移学习的钢丝绳断丝定量识别方法。通过连续小波变换将原始断丝漏磁信号转换成时频图。将预训练网络GoogLeNet的低层参数直接迁移,使用标记好的时频图对网络高层进行参数调整,得到最终的目标模型。通过内外部断丝试验验证了所提出的定量识别模型的效果,将传统的BP(back propagation)神经网络与所提出的方法进行对比。结果表明:基于迁移学习的断丝定量识别方法能准确区分钢丝绳的内外部断丝故障,分类准确率达到了97.2%;与传统BP神经网络相比,所提出的方法对各种断丝具有更好的识别性能。

To address the current problems of quantitative testing of broken wires for steel wire ropes,using the powerful feature extraction ability of deep convolution neural network,a quantitative identification method based on transfer learning was proposed.The original magnetic flux leakage signals were converted into time-frequency images by the continuous wavelet transform.The low-level parameters of the pre-trained network GoogLeNet were directly transferred,and the labeled time-frequency images were used to adjust the parameters in the high level of the network to obtain the final target model.The effectiveness of the proposed quantitative recognition model was verified by the experiments of internal and external broken wires.The traditional BP(back propagation)neural network was used as a comparison.Results show that the quantitative identification method based on transfer learning can accurately classify the internal and external broken wires of the wire rope with an accuracy rate of 97.2%.Compared with the traditional BP neural network,the proposed method has better recognition performance for various broken wires.