In a paper published in the journal Scientific Reports, researchers investigated the prediction of jumbo drill rate of penetration (ROP) in underground mining using regression and machine learning (ML) methods.
Study: Predicting Jumbo Drill Rate of Penetration in Underground Mining. Image Credit: I Made Pasek/Shutterstock
They employed multilayer perceptron (MLP), support vector regression (SVR), and random forests (RF) with the rock mass drillability index (RDi) for rock mass characterization. The high accuracy of SVR in the training and testing phases highlights its effectiveness in optimizing drilling processes and reducing operational costs.
Related Work
Past work in underground excavation underscores the pivotal role of drilling in the efficiency of the drill and blast method, a key factor in cost and time management. Factors like geological conditions and machine characteristics affect drilling performance, with controllable and uncontrollable parameters at play.
Despite efforts using empirical and statistical models, accurately predicting the ROP remains challenging due to complex rock-property relationships. Traditional regression models face limitations, prompting the exploration of artificial intelligence (AI) techniques like ML.
Empirical Model Development
The team aimed to develop an empirical model using ML algorithms to predict the ROP of jumbo drills in underground mines, and they collected real field data from seven mines in Iran. Six parameters related to drilling and rock properties focused on the RDi as a comprehensive indicator of rock mass characteristics were considered.
Data collection involved gathering information on rock discontinuities, laboratory testing of rock properties, and classification of drilling zones based on RDi ratings. Correlation analysis revealed a linear relationship between RDi and ROP, with variability attributed to machine operating parameters.
Data Preprocessing Essentials
The analysts emphasize the criticality of rigorous preprocessing of raw data to ensure accurate model training, which is essential for addressing noise and outlier data. Failure to remove outliers and reduce noise can hinder the model learning process and prolong training time.
In this phase, collected data undergoes meticulous review and analysis, with 80% allocated for training and 20% for testing, randomly selected. Two critical actions are taken: first, data analysis and verification of accuracy and precision, and second, data matching to prevent scattering and consolidate data within specific intervals.
Non-quantitative data are handled using various techniques, with improper scaling addressed to avoid misestimating variable significance in regression analysis. The study employs the interquartile range (IQR) method, a widely used outlier labeling technique, to handle extreme values before processing, ensuring robust data analysis.
After the data cleaning and normalization process, the next crucial step is model validation. This step is of utmost importance as it provides comprehensive insights into the model's performance. Five statistical performance metrics are employed: determination coefficient (R2), variance accounted for (VAF), mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE). These metrics play a pivotal role in assessing prediction accuracy and error margins, which are crucial for evaluating the effectiveness of the model in computational mechanics.
Analytical Comparison
The study uses nonlinear regression analysis and ML methods to develop an ROP model for jumbo drills. Initial regression analyses explore basic functions to establish correlations between independent variables and the ROP, revealing the complexity of factors influencing drilling speed. Consequently, the team develops and compares nonlinear regression models, ultimately identifying the most accurate model.
Meanwhile, ML methods, including artificial neural networks (ANN), support vector machines (SVM), and RF, are employed to address the complexity of drilling processes. Hyperparameter grid searches are conducted to optimize model performance, resulting in selecting the most suitable network architecture and transfer functions. The developed models are evaluated based on their performance indices and compared to determine their effectiveness in predicting the ROP.
Furthermore, sensitivity analysis uses the cosine amplitude method to identify the most influential factors affecting the ROP. The strengths of relations between the rate of penetration and input parameters are analyzed, revealing the significant impact of various factors on drilling speed. Among these factors, the RDi emerges as the most influential parameter, highlighting the critical role of rock mass characteristics in determining drilling efficiency. Overall, the study provides valuable insights into regression and ML approaches for modeling the ROP of jumbo drills in underground mining operations.
Rock Mass
The study highlights the effectiveness of RDi in estimating rock mass characteristics and its impact on ROP studies. The dataset underwent preprocessing, including outlier detection and replacement, normalization, and train-test data splitting. ML hyperparameters were optimized using grid search, leading to the identification of optimal parameters.
Evaluation metrics, such as MAE, MAPE, RMSE, R2, and VAF, were utilized to assess model accuracy, indicating that ML models outperformed regression models due to their ability to capture complex nonlinear relationships. Particularly, the SVR method exhibited superior accuracy. Furthermore, a comparison with the Shen model revealed the inadequacy of traditional mathematical models in accounting for rock mass discontinuities, emphasizing the superiority of computational intelligence-based models in predicting ROP accurately.
Conclusion
To summarize, this study investigated the factors influencing the ROP in drilling processes and developed predictive models using regression and ML techniques. Results showed that ML algorithms outperformed regression models in predicting ROP, particularly SVR. Sensitivity analysis revealed the significant contribution of various parameters to ROP, with the rock ass characteristic being the most influential.
In comparison, the developed models show promise for estimating jumbo drill performance in challenging rock conditions; limitations exist, particularly in saturated rock environments and less jointed rock masses. Nonetheless, these models offer potential for drilling automation in underground mines, paving the way for enhanced efficiency and cost savings.
Machine Learning Boosts Earthquake Prediction Accuracy in Los Angeles