<p>To ensure the safe operation of the underground belt conveyor and accurately predict the mechanical properties of its key structures, as well as to enhance the intelligent monitoring capabilities of the belt conveyor, we propose a digital twin-driven online monitoring method for assessing the performance of the belt conveyor structure. Key technologies were investigated using the chain wheel as a case study. Initially, a five-dimensional digital twin model was constructed based on the operating characteristics of the underground belt conveyor to establish a comprehensive digital twin framework for this system. Subsequently, finite element simulation technology was employed to develop simulation models for both the chain wheel and belt chain, with data collected under various operational conditions. An XGBoost (Extreme Gradient Boosting) agent model was then constructed using this simulation dataset. To validate its predictive performance, five agent models were selected for comparative verification; results indicate that our XGBoost agent model exhibits superior accuracy, fitting degree and stability compared to other models. Finally, an online monitoring system for evaluating the structural performance of the belt conveyor was implemented using a laboratory test bench to confirm feasibility and present novel insights into intelligent health monitoring practices.</p>