DNN–GA–RF prediction model for rock strength indicators based on sound level and drilling parameters

This study explores the estimation of rock properties using sound levels and drilling parameters recorded during quantitative drilling. Utilizing an indoor digital drilling test system and acoustic monitoring device developed by the authors, the research involved setting varied drilling parameters for vertical drilling of various rocks and rock-like materials. The equivalent continuous A-weighted sound pressure levels were measured during the drilling process. Simultaneously, the compressive strength of the drilled rock and rock-like samples was determined in the laboratory. A regression prediction model was developed using drilling rate (V), rotational speed (N), torque (M), propulsive force (F), ground stress magnitude (σ), and equivalent continuous A-weighted sound pressure level (Leq A) as inputs and compressive strength Rc as the output. To enhance prediction accuracy, the model integrates the feature extraction capabilities of traditional random forest (RF) and deep neural network (DNN) methodologies while incorporating a genetic algorithm (GA) for feature selection and parameter optimization. Analysis of the results indicates that the prediction accuracy of the DNN–GA–RF (deep neural network–genetic algorithm–random forest) model, based on sound level and drilling parameters, significantly surpasses traditional methods such as ridge regression, support vector machine, and RF. This finding highlights the potential of the equivalent continuous A-weighted sound pressure level as an effective tool for estimating rock properties during drilling. Furthermore, the model’s integrated learning-based approach to predicting rock strength indicators offers an innovative concept and methodological basis for the nondestructive assessment of rock properties.