Abstract
During the construction of TBM tunnels, a substantial quantity of rock debris is generated, leading to significant land occupation and environmental pollution. Recycling rock debris into construction materials and other resources emerges as a viable solution to these problems. To realize the continuous classified storage and disposal of tunnel rock debris, this research explores the four-level processing network, establishes an objective function for evaluating the recycling value of tunnel rock debris during TBM tunneling, and grades the recycling value by calculating the weight and similarity of their performance indicators (uniaxial compressive strength, content of acicular and flattened particles, mud content, and crushing index) through the TOPSIS method. Through correlation and weight analysis, we identify five key characteristics, i.e. cutterhead torque, tool penetration, cutterhead thrust, advancing rate, and support shoe pump pressure, to conduct real-time perception of the recycling value level of rock debris. Leveraging a comprehensive database that encompasses both tunnel rock debris performance indicators and TBM tunneling parameters, perception models are constructed using different machine learning algorithms. After Bayesian hyperparameter optimization, the perception models based on CART, SVM, KNN, and ANN demonstrate accuracies of 67.5%, 80.0%, 82.5%, and 83.8% respectively. Notably, the hyperparameter optimization significantly enhances the accuracy of the ANN perception model. When applying the optimized ANN-based rock debris recycling value grade perception model to TBM tunnel engineering, the tested perception accuracy rate stands at 83.3%, demonstrating its effectiveness and potential for practical applications. This approach provides valuable guidance for the graded storage and efficient recycling of tunnel rock debris and helps to alleviate the pollution problem.