国外文献 > 1.轴承 > 5. 监测与诊断 > 故障诊断
关键词搜索：
| 标 题： |
A Bearing Health Monitoring Method Based on Interpretable Improved GoogLeNet
基于可解释改进的GoogLeNet的轴承健康监测方法 |
| 作 者： |
Siying Chen
Liangji Chen
Zixuan Chen |
| 关键词： |
Fault diagnosis [故障诊断]
GoogleNet network [ GoogleNetNetNet 网络]
Grid search method [ 网格搜索方法]
Feature mapping [ 特征映射] |
| 出版年： |
2023 |
| 文集名： |
2023 IEEE International Conference on Sensors, Electronics and Computer Engineering (ICSECE 2023)
2023年IEEE国际传感器、电子与计算机工程会议（ICSECE 2023） |
| 会议名： |
2023 IEEE International Conference on Sensors, Electronics and Computer Engineering (ICSECE 2023)
2023年IEEE国际传感器、电子与计算机工程会议（ICSECE 2023） |
| 届 次： |
|
| 摘 要： |
Problems such as the need for a priori knowledge, expert experience, feature interpretation and difficulty in signal feature extraction exist in the process of rotational fault diagnosis of industrial machinery. A diagnosis method based on progressive grid search optimized GoogLeNet is proposed, firstly, a two-dimensional wavelet real frequency transform is performed on the collected noisy data to constitute an image data set, and then a noise test is performed on several networks to obtain a noise-resistant The GoogleNet network, which has the strongest performance, is then used to obtain a well-trained bearing fault diagnosis model by combining the GoogleNet network with progressive grid optimisation and parameters, and finally the data is acquired to test the network model and the grid is interpretable using Grad-cam feature mapping. After the test, it was found that the model after parameter optimization has a better correct rate than the model with empirically selected parameters, and its fault recognition accuracy is higher and more robust.
在工业机械旋转故障诊断过程中，存在需要先验知识、专家经验、特征解释以及信号特征提取困难等问题。本文提出一种基于渐进网格搜索优化的GoogLeNet诊断方法：首先对采集的噪声数据进行二维小波实频变换，构建图像数据集； 然后对多个网络进行噪声测试，选出抗噪性能最强的GoogleNet网络； 接着将该网络与渐进网格优化及参数相结合，训练出一个性能优良的轴承故障诊断模型； 最后通过Grad-cam特征映射技术对网格进行可解释性分析，验证网络模型的性能。实验结果表明，经过参数优化后的模型比采用经验参数选择的模型具有更高的准确率，其故障识别精度更高且更具鲁棒性。 |
| 中文内容为网络AI翻译，专业术语未必准确，仅供参考。 |
京公网安备 11010202009788