【摘要】 轴承故障检测在工业设备的健康监测与维护中具有重要意义，但传统方法在处理超长时序信号和复杂工况时，难以同时兼顾全局建模能力与计算效率，导致检测精度下降或推理速度受限。针对这一问题，提出了一种基于噪声自适应多变量变分模态分解（NA-MVMD）和自适应局部卷积注意力Transformer(ALCAT)的故障检测方法，以提高故障检测的准确性、鲁棒性及计算效率。其中，NA-MVMD通过引入噪声自适应机制优化多变量信号解耦，增强特征提取的稳定性和可靠性；ALCAT结合信号自适应分组（SAS）、卷积神经网络（CNN）、注意力机制（Attention）及Transformer结构，实现高效的局部与全局特征学习，从而提升故障识别精度和推理速度。为验证该方法的有效性，将NA-MVMD-ALCAT与多个先进模型进行对比。实验结果表明，该方法在各项性能指标上均显著优于对比方法，F1值达到0.9863，展现出优越的故障检测能力。此外，当窗口大小设定为1024时，该模型实现了检测精度与计算效率的最优平衡，能够高效处理超长时序数据。综上，NA-MVMD-ALCAT适用于在线轴承故障监测与智能运维系统，为工业设备健康监测提供了一种高效、精准的解决方案。更多还原

【Abstract】 Bearing fault detection plays a crucial role in the health monitoring and maintenance of industrial equipment. However, traditional methods struggle to balance global modeling capability and computational efficiency when processing ultra-long time-series signals and complex operating conditions, leading to reduced detection accuracy and limited inference speed. To address this issue, this paper proposes a fault detection method based on Noise-Adaptive Multivariate Variational Mode Decomposition(NA-MVMD) and Adaptive Local ConvolutionAttention Transformer(ALCAT) to improve accuracy, robustness, and computational efficiency of fault detection. Specifically, NA-MVMD introduces a noise-adaptive mechanism to optimize multivariate signal decoupling, enhancing the stability and reliability of feature extraction. Meanwhile, ALCAT integrates Signal-Adaptive Segmentation(SAS), Convolutional Neural Networks(CNN), Attention Mechanisms, and Transformer structures, enabling efficient local and global feature learning, thereby improving fault recognition accuracy and inference speed. To validate the effectiveness of the proposed method, we compare NA-MVMD-ALCAT with multiple advanced models. Experimental results demonstrate that the proposed approach outperforms all comparison methods across various performance metrics, achieving an F1 score of 0.9863, which highlights its superior fault detection capability. Moreover, when the window size is set to 1024, the model achieves an optimal balance between detection accuracy and computational efficiency, effectively handling ultra-long time-series data. Overall, NAMVMD-ALCAT is well-suited for online bearing fault monitoring and intelligent maintenance systems, providing an efficient and accurate solution for industrial equipment health monitoring.更多还原