Point cloud segmentation is a critical step in tunnel construction for detecting overbreak and underbreak regions and measuring their volumes. However, due to the irregular contours of tunnels with overbreak and underbreak regions, achieving high segmentation accuracy is challenging, especially for objects near the contours that may cause interference. To address this issue, a Transformer-based model is proposed. The model employs a self-attention mechanism to capture long-range dependencies across the global domain and integrates a local information fusion module to combine local geometric and feature context, leveraging the inherent point distribution in 3D space. By adopting a DGCNN-like architecture, the model enhances its feature representation capabilities. Experiments were conducted on a constructed 3D point cloud dataset representing overbreak and underbreak tunnels, comparing the proposed model with DGCNN and Point Transformer. The results demonstrate that the proposed model outperforms the others in terms of inference speed, computational resource requirements, and segmentation accuracy, achieving an mIoU of 75.8% and showing significant performance improvements. This model not only provides technical support for the excavation of overbreak and underbreak regions but also enables point cloud segmentation for lined tunnels, facilitating visualized construction and enhancing engineering quality and efficiency.