Segment dislocation is a major issue in subway shield tunnels, and its detection is crucial for ensuring structural and operational safety. Existing methods often rely on local point clouds, failing to provide a comprehensive and precise representation of overall segment dislocation. Others assume an ideal cylindrical geometry, neglecting common non-uniform segment deformations. To address these limitations, we introduce a novel global deformation-aware approach for segment dislocation detection. The method first separates non-lining tunnel points to accurately reflect the tunnel lining structure and eliminate their adverse effects on segment joint extraction. This is achieved using an ellipse fitting residual statistics-based algorithm. Subsequently, circumferential joints are extracted and located by integrating adaptive intensity features with prior information, allowing the division of the tunnel lining into individual shield rings. Radial joints within each shield ring are then identified through a deep feature clustering algorithm. Finally, circumferential and radial dislocations are detected using a piecewise fitting approach for shield ring segments. The feasibility and effectiveness of the proposed method are verified using Rail-borne Mobile Laser Scanning (RMLS) point cloud data from Guangzhou Metro Line 8. Experimental results demonstrate that the method effectively detects dislocations under non-uniform deformation conditions, overcoming errors introduced by traditional methods that simplify segment geometry. Compared to conventional fault detection techniques, the proposed approach achieves improved accuracy and robustness.