دورية أكاديمية
基于改进深度残差收缩网络的旋转机械故障诊断.
| العنوان: | 基于改进深度残差收缩网络的旋转机械故障诊断. (Chinese) |
|---|---|
| Alternate Title: | Fault diagnosis of rotating mechanical based on improved deep residual shrinkage network. (English) |
| المؤلفون: | 杨正理, 吴馥云, 陈海霞 |
| المصدر: | Journal of Mechanical & Electrical Engineering; Mar2023, Vol. 40 Issue 3, p344-352, 9p |
| Abstract (English): | In order to solve problems of the degradation and over fitting phenomenon of rotating machinery vibration signals in the process of multi-layer deep learning, the low accuracy of fault diagnosis caused by data sample with noise, and the model inclination in training caused by imbalance data samples, an improved deep residual shrinkage network（DRSN） based fault diagnosis method for rotating machinery was presented. Firstly, the multi-fault long time series data samples were processed by matrix, and the multi-dimensional gray scale fault samples which were easily accepted by the model were obtained. For the mechanical aging process of rotating machinery from normal state to fault state, multi-point random sampling method was used to construct the whole life cycle data sample for fault diagnosis. Then, the residual term, attention mechanism and focus loss function were introduced to construct a multi-layer deep residual contraction network to diagnose the vibration signals of rotating machinery on the basis of convolutional neural network (the residual term reduced the feature loss of sample data in the training process and avoided the model degradation and over fitting, the attention mechanism and soft thresholding automatically set the noise threshold to reduce the impact of noise on fault diagnosis accuracy, the focus loss function modified the orientation of model training and improved the efficiency and sensitivity of model training）. Finally, the model was verified by using the sample of rolling bearing database. The results show that the multi-layer DRSN model has no obvious degradation phenomenon in the training process, and can maintain good training efficiency and fault diagnosis accuracy, avoid noise interference effectively, and correct the bias of model training on unbalanced data sets. Comparing with other models, the average fault diagnosis accuracy of the multi-layer DRSN model is improved by about 1%～ 6%. [ABSTRACT FROM AUTHOR] |
| Abstract (Chinese): | 旋转机械振动信号在多层深度学习过程中会出现退化和过拟合现象, 同时含噪数据样本也会使模型故障诊断正确率偏低, 数据样本不平衡会引起模型训练具有倾向性, 针对以上一系列问题, 提出了一种基于改进型深度残差收缩网络 (DRSN) 的旋转机械故障诊断方法.首先, 对多故障、长时间序列数据样本进行了矩阵化处理, 得到了模型容易接受的多维度灰度图故障样本; 针对旋转机械从正常状态到故障状态的机械老化过程, 采用了多点随机采样方法, 构建了全寿命周期数据样本, 用于后续的故障诊断; 然后, 在卷积神经网络(CNN)的基础上, 通过引入残差项、注意力机制和焦点损失函数, 构建起了多层深度残差收缩网络, 对旋转机械进行了故障诊断(其中, 残差项降低了训练过程中样本数据的特征损失, 避免了模型的退化和过拟合; 注意力机制和软阈值化自动设置噪声阈值, 降低了噪声对故障诊断精度的影响; 焦点损失函数修正了模型训练的倾向性, 提高了模型训练效率和灵敏性); 最后, 利用滚动轴承数据库样本对模型的性能进行了实例验证.研究结果表明: DRSN 模型在训练过程中没有出现明显的退化现象, 能够始终保持较高的训练效率和故障诊断精度, 有效避免了噪声干扰, 在不平衡数据集上修正了模型训练的倾向性.与其他模型相比较, DRSN多层模型的平均故障诊断精度提高约1％～6％. [ABSTRACT FROM AUTHOR] |
| Copyright of Journal of Mechanical & Electrical Engineering is the property of Mechanical & Electrical Engineering Magazine and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| قاعدة البيانات: | Complementary Index |
كن أول من يترك تعليقا!