基于多传感器融合与深度学习的工业轴承源域故障诊断

Fault Diagnosis of Industrial Bearings in the Source Domain Based on Multi-Sensor Fusion and Deep Learning

摘要: 针对工业轴承在复杂工况下故障特征提取难、传统诊断方法泛化能力不足的问题，本文提出一种融合多传感器数据与深度学习的源域故障诊断方法。首先对源域台架实验数据进行重采样、小波去噪与窗口采样预处理，构建时域、频域、时频域及包络域多维度特征集；随后设计三种传感器数据场景(All、Connect、Only DE)，并构建卷积神经网络(CNN)、CNN-LSTM及CNN-LSTM-Attention三种深度学习模型，通过Accuracy、Precision、Recall及F1-score等指标进行性能评估。实验结果表明：关键传感器融合场景(DE + FE)下的纯CNN模型表现最优，故障诊断准确率达96.61%，其中内圈故障召回率100%、外圈故障召回率94.1%、滚动体故障召回率80%，有效解决了类别不平衡问题。该方法通过多域特征协同与传感器数据优化，为后续跨工况迁移诊断提供了高精度源域基准模型，为工业轴承故障诊断的工程化应用提供了技术支撑。

Abstract: A novel source-domain fault diagnosis method integrating multi-sensor data with deep learning is proposed in this paper to address the challenges of extracting fault features under complex operating conditions of industrial bearings and the limited generalization capability of traditional diagnostic approaches. Initially, the source-domain bench test data undergo preprocessing, including resampling, wavelet denoising, and window sampling to construct multi-dimensional feature sets spanning the time domain, frequency domain, time-frequency domain, and envelope domain. Three sensor data scenarios (All, Connect, Only DE) are subsequently designed, and three deep learning architectures—Convolutional Neural Network (CNN), CNN-LSTM, and CNN-LSTM-Attention—are constructed. Performance is evaluated using metrics such as Accuracy, Precision, Recall, and F1-score. Experimental results demonstrate that the pure CNN model under the key sensor fusion scenario (DE + FE) achieves optimal performance, with a fault diagnosis accuracy of 96.61%. Specifically, it attains 100% recall for inner race faults, 94.1% for outer race faults, and 80% for rolling element faults, effectively addressing the class imbalance problem. By leveraging multi-domain feature collaboration and optimized sensor data integration, the proposed method establishes a high-precision source-domain benchmark for subsequent cross-condition transfer diagnosis, thereby offering technical support for the engineering application of fault diagnosis in industrial bearings.