The Industrial Internet of Things (IIoT) extends IoT into the industrial sector, enabling real-time data collection through interconnected devices. However, the increasing sophistication of cyberattacks poses significant security threats. Traditional network intrusion detection systems (NIDS) often fail to identify advanced attack patterns due to the reliance on manual feature design. To address this, we propose the MCAN-BIFT intrusion detection system, consisting of the Multi-Scale Convolutional Attention Network (MCAN) and the Bidirectional Interactive Fusion Transformer (BIFT). The system extracts attack features through one-dimensional convolution, enhances them using a multi-scale convolutional attention hybrid module (MS-CAHM) with channel and spatial attention mechanisms, and feeds them into BIFT module, which combines interacting information from two directions and merges them at the fusion layer to generate the final output. Evaluated on the Edge-IIoTset dataset, the model achieved a test accuracy of 97.30% across 15 classes, outperforming recent studies on IIoT intrusion detection.