Cutter wear detection in shield tunneling has become an operational priority, as tool condition directly governs excavation efficiency, project costs, and tunnel safety. With cutter replacement accounting for significant downtime in many projects, reliable wear monitoring has shifted from a preference to a necessity.

Direct Measurement Approaches

Eddy current sensors, widely deployed on disc cutter rings, can identify five distinct damage states—normal wear, eccentric wear, tip blunting, chordal wear, and insert fracture—while also enabling rotational speed calculation to detect cutter jamming. Image analysis offers another pathway, with high-resolution cameras and convolutional neural networks automating wear region identification and progression tracking.

Indirect Detection via Operational Signals

Vibration analysis, with sensors installed on the shield body rather than the cutterhead, has demonstrated clear correlations between vibration patterns and wear progression. Acoustic emission combined with machine learning classifiers has achieved wear-level prediction accuracy exceeding 95%. Parameter fusion models integrate penetration rate, cutterhead speed, thrust, and torque into an "equivalent wear" index for real-time quantitative assessment.

Intelligent Prediction and Transfer Learning

Deep learning models have outperformed traditional empirical formulas. A CNN-LSSVM hybrid achieved 99.4% accuracy in damage state classification, while the MVSAPNet prototypical network delivered 91.87% accuracy in real-world projects. To address model generalizability challenges across different geological conditions, Transformer-based transfer learning with domain-adversarial neural networks has been developed, validated in the Qingdao Jiaozhou Bay Second Subsea Tunnel project.

Outlook

The field is moving toward edge computing and digital twin integration for closed-loop monitoring and autonomous cutter-change decision-making, reducing downtime and enhancing tunneling safety under complex geological conditions.