Deep learning that helps crews spot trouble before it becomes expensive
We developed a vision based analysis system that monitors the soil coming out of a tunnel boring machine and turns a simple video feed into a real time quality indicator. The model was trained on carefully labeled data and learned to recognize subtle shifts in material that often signal issues inside the cutter head. This gives operators an early warning window, letting them adjust settings or perform checks before a small irregularity becomes a full stoppage.
To keep hardware costs under control, we also optimized the model so it can run smoothly on non GPU devices. The result is a practical tool that reduces downtime, cuts operating costs, and gives TBM crews clearer insight into what is happening underground.