针对传统故障诊断方法在应对复杂工况和多样化故障模式时面临准确性和可靠性方面的问题，提出一种基于瞬时傅里叶变换（Short-Time Fourier Transform, STFT）和注意力时序卷积神经网络（Attention Temporal Convolutional Neural Network, ATCNN）的滚动轴承故障诊断方法。在对滚动轴承故障特征分析基础上，利用STFT提取滚动轴承的振动信号作为特征参数并构建故障诊断样本，通过在卷积神经网络（Convolutional Neural Network, CNN）中嵌入卷积块注意力模块（Convolutional Block Attention Module, CBAM）和长短期记忆网络（Long Short-Term Memory, LSTM），增强对滚动轴承故障的特征提取能力和序列建模能力，实现故障诊断准确性。以凯斯西储大学轴承数据集为例，与基准CNN、CNN-LSTM和CNN-CBAM的故障诊断方法进行对比分析，结果表明:基于STFT和ATCNN的滚动轴承故障诊断方法在多样化故障模式下取得了显著的性能提升，能够更准确地识别滚动轴承的故障类型。

Addressing the issues of accuracy and reliability that traditional fault diagnosis methods face when dealing with complex working conditions and diverse fault modes, this paper proposes a rolling bearing fault diagnosis method based on Short-Time Fourier Transform (STFT) and Attention Temporal Convolutional Neural Network (ATCNN). Building on the analysis of rolling bearing fault characteristics, STFT is employed to extract vibration signals as feature parameters to construct fault diagnosis samples. By embedding the Convolutional Block Attention Module (CBAM) and Long Short-Term Memory (LSTM) network into the Convolutional Neural Network (CNN), the method enhances the feature extraction and sequence modeling capabilities for rolling bearing faults, thereby improving diagnostic accuracy. Using the Case Western Reserve University bearing dataset as an example, a comparative analysis with benchmark methods including CNN, CNN-LSTM, and CNN-CBAM was conducted. The results demonstrate that the proposed STFT-ATCNN-based rolling bearing fault diagnosis method achieves significant performance improvements under diverse fault modes, enabling more accurate identification of rolling bearing fault types.