Abstract:

To address the problems of lagging fault response and insufficient accuracy in the traditional operation and maintenance mode of mine main hoists, a fault diagnosis model for mine main hoists based on BO-XGBoost was constructed, and the SHAP method was integrated to improve the model interpretability. The Bayesian Optimization (BO) algorithm was used to optimize the hyperparameters of the eXtreme Gradient Boosting (XGBoost) model. Based on the monitoring data from an experimental mine main hoist, the XGBoost model combined with the SHAP attribution analysis method was adopted to identify the key influencing factors and their action mechanisms. The results show that compared with the baseline XGBoost model, the BO-XGBoost model increases accuracy by 4.1%, reduces log loss by 41.9%, and shortens model training time by 80.1%. Compared with traditional decision tree, random forest and LightGBM algorithms, the BO-XGBoost model improves precision by 26.5%, 11.9% and 13.6%, respectively, demonstrating excellent test accuracy.Wire rope tension, lower sheave temperature and motor voltage are the three key causal factors of faults. Different fault types are affected by different factors; for instance, excessively high main shaft vibration, motor temperature and excessively low hoisting speed provide greater positive gain for main shaft fault prediction. Three-factor interaction analysis reveals the dominant role and influence patterns of various factors during wire rope faults. The probability of wire rope faults is mainly dominated by tension, motor current and hoisting speed. Excessively low tension or current significantly increases the risk, and rising hoisting speed further aggravates the fault probability, whereas lower sheave temperature has a weak influence.

Key words: Bayesian Optimization-eXtreme Gradient Boosting (BO-XGBoost), SHapley Additive exPlanations (SHAP), mine main hoist, fault diagnosis, cause location