<p>With the rapid development of Internet of Things and artificial intelligence technologies, the massive data generated by mechanical equipment has driven the fault diagnosis technology into the “big data” era. The analysis, diagnosis, and prediction of these data have become crucial for ensuring the smooth and safe operation of mechanical equipment. In recent years, traditional convolutional neural networks (CNNs) and Transformer-based models have been widely used in industrial fault diagnosis. This paper proposes a new lightweight fault diagnosis framework, Sim-ConvFormer, to address the issues of high model complexity and stringent hardware requirements. The Sim-ConvFormer framework integrates SimAM and External Attention. SimAM (A Simple, Parameter-Free Attention Module for Convolutional Neural Networks) enhances the model’s sensitivity to fine-grained signal variations, capturing locally significant features at different scales. Unlike self-attention, External Attention enhances generalization by computing the affinity between input features and two external memory modules shared across the dataset, thereby capturing global contextual information. The synergy of these technologies not only preserves their individual advantages but also enhances the model’s robustness and accuracy in handling various faults. Experiments on three different mechanical systems demonstrate Sim-ConvFormer’s superior fault diagnosis performance, particularly in terms of model lightness and diagnostic robustness compared to existing Transformer-based methods and CNN-based methods. These results indicate that Sim-ConvFormer is an effective fault diagnosis framework suitable for deployment in resource-constrained industrial environments.</p>