Real-Time Prediction of Disc Cutter Wear in Low-Abrasive Rocks: Integrating Physico-Mechanical Properties and Signal Processing Features Through Machine Learning Methods

Abstract

Tunnel boring machine (TBM) is a popular choice for mechanical excavation due to its efficient, safe, and cost-effective tunnelling capabilities compared to traditional methods. One of the factors that can impact TBM performance is the wear of disc cutters. The wear of the cutting discs can lead to a reduction in their cutting ability, resulting in slower excavation rates, increased power consumption, and increased wear on other TBM components. In this research, sound and vibration signals, along with physical and mechanical characteristics, were used as a real-time method to determine disc wear during the cutting of low-abrasive rocks. For this purpose, the features extracted from the sound and vibration signals recorded during the cutting process were compared with the amount of disc wear. It was observed that with the progress of disc wear, the sound signal decreases, and the vibration increases. Finally, three machine learning methods, including decision tree (DT), random forest (RF), and extreme gradient boosting (XGBoost), were employed to analyse disc wear. The fivefold cross-validation approach was utilized to assess the predictive accuracy of the models. The XGBoost model achieved an R² value of 0.9424, making it the most accurate model for predicting the wear of the disc cutter. The DT and RF models attained an accuracy of R² = 0.8379 and R² = 0.8941, respectively. The method presented in this study can estimate the wear of the disc in real-time and suggest the right time to replace the disc.