Abstract:

Realtime judgment and early warnings of tunnel face surrounding rock conditions during tunnel boring machine (TBM) tunneling are paramount in intelligent rock mass identification and safe and efficient TBM tunneling. A few onsite images of rock muck are collected from some TBM projects, i.e. Zhuxi Reservoir diversion tunnel. First, the correspondence between rock muck and rock mass integrity is summarized. Second, according to the characteristics of rock muck images, a new segmentation algorithm based on the improved marked watershed transform is proposed to extract and classify the muck images. Finally, the accuracy of the method is verified using 160 pieces of onsite rock muck images under different degrees of rock mass integrity. The results show that the average accuracies of A, B, and C image classification are 96.3%, 94%, and 86.7%, respectively, indicating that the method can effectively realize automatic analysis and recognition of rock muck, providing an early warning for TBM tunneling in adverse geology. 

Key words: TBM, marked watershed transformation, rock muck, image segmentation, image recognition