国外文献 > 1.轴承 > 7. 智能方法与计算 > 神经网络
关键词搜索：
| 标 题： |
A Deep Intelligent Hybrid Model for Fault Diagnosis of Rolling Bearing
滚动轴承故障诊断的深度智能混合模型 |
| 作 者： |
Zhao, Xiaoqiang
Luo, Weilan |
| 关键词： |
Rolling bearing [滚动轴承]
Fault diagnosis [ 故障诊断]
Feature fusion [ 特征融合]
Convolutional neural network [ 卷积神经网络]
Network optimization [ 网络优化]
CONVOLUTIONAL NEURAL-NETWORK [ 卷积神经网络]
NOISY ENVIRONMENT [ 嘈杂的环境]
MACHINERY [ 机械]
ENSEMBLE [ 整体] |
| 出版年： |
2023 |
| 期刊名： |
Journal of Vibration Engineering & Technologies
振动工程与技术杂志 |
| 年卷期： |
2023, vol.11, no.2 |
| 摘 要： |
Background In the actual industry, due to the small amount of rolling bearing fault data in rotating machinery and the equipment operation under variable operating conditions, there are often serious data imbalance and weak fault signals easily drowned in strong noise, which make it difficult to extract feature information of the bearing health status, leading to a low accuracy of fault diagnosis. Purpose The purpose of this paper is to establish a fault diagnosis model based on deep learning, which can efficiently extract the feature information of vibration signals under the condition of data imbalance, and has good fault diagnosis results. Methods First, batch normalized convolution and batch normalized dilated convolution are used to build a High-performance Double Pyramidal Fusion (HDPF) module, which increases the perceptual field to extract more adequate and effective multi-scale complex feature information. Batch normalization is used to adjust the feature distribution and eliminate the feature distribution problem caused by data imbalance, the problem which data imbalance causes feature distribution discrepancies is eliminated. Second, a deep Adaptive Residual Decision Fusion (ARDF) module is used to adaptively calibrate the weak feature information, thus improving the adaptive feature learning performance for variable working conditions. Finally, an exponential moving average technique is used to optimize the overall performance of the proposed model by iteratively updating the hyperparameters. Results In the noise environment experiment, the diagnosis result of the proposed model for rolling bearing faults is 99.70%, and the diagnosis result for gearbox faults is 99.10%. By comparing with other mainstream intelligent diagnostic models, the experimental results show that the proposed method has better advancement and superiority. Conclusion In this paper, a deep intelligent hybrid model for rolling bearing fault diagnosis is proposed, which can achieve good fault diagnosis results under strong noise environment and solve the problem of data imbalance sample under complex working conditions.
背景 在实际工业中，由于旋转机械轴承故障数据量较少，且设备在变化的运行条件下工作，常出现严重的数据不平衡问题，导致微弱的故障信号容易被强噪声淹没，从而难以提取轴承健康状态的特征信息，造成故障诊断准确率较低。目的 本文旨在基于深度学习建立一种故障诊断模型，能够在数据不平衡的条件下高效提取振动信号的特征信息，并取得良好的故障诊断效果。方法方法方法 首先，采用批归一化卷积和批归一化膨胀卷积构建高性能双金字塔融合（HDPF）模块，扩大感知域，以提取更充分、有效的多尺度复杂特征信息。通过批归一化调整特征分布，消除因数据不平衡引起的特征分布不均问题，从而解决数据不平衡导致的特征分布偏差。其次，采用深度自适应残差决策融合（ARDF）模块对弱特征信息进行自适应校准，从而提升模型在变化工作条件下的自适应特征学习性能。最后，通过迭代更新超参数，利用指数加权平均技术优化所提模型的整体性能。实验结果表明，在噪声环境下，该模型对滚动轴承故障的诊断准确率为99.70%，对齿轮箱故障的诊断准确率为99.10%。与其他主流智能诊断模型相比，实验结果显示出本方法具有更高的改进性和优势。结论：本文提出了一种用于滚动轴承故障诊断的深度智能混合模型，能够在强噪声环境下实现良好的故障诊断效果，并有效解决复杂工况下数据样本不平衡的问题。 |
| 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