A Systematic Survey of Hybrid ML Techniques for Predicting Peak Particle Velocity (PPV) in Open-Cast Mine Blasting Operations

Blasting operations in open-cast mines are crucial for mineral extraction but also come with significant environmental and structural risks. The peak particle velocity (PPV) generated during blasting is a key metric for assessing the impact of blasting vibrations on surrounding structures and the environment. Accurate PPV prediction is essential for optimizing blasting practices, reducing environmental damage, and ensuring structural safety. Traditional prediction methods face limitations in handling nonlinear relationships and high-dimensional data, while machine learning (ML) techniques, particularly hybrid ML methods, show great potential in PPV prediction. This article aims to systematically review the application of hybrid ML techniques in PPV prediction for open-cast mine blasting, exploring their advantages, challenges, and future research directions.

Source of the Paper

This paper is co-authored by Gundaveni Shylaja and Ragam Prashanth, both from the School of Computer Science and Engineering at VIT-AP University. The paper was accepted on February 18, 2025, and published in the journal Artificial Intelligence Review with the DOI 10.1007/s10462-025-11156-3.

Key Points

1. Advantages of Hybrid Machine Learning Techniques

Hybrid ML techniques combine traditional methods (e.g., decision trees and support vector machines) with advanced techniques (e.g., ensemble learning and neural networks), demonstrating superior performance in PPV prediction. Compared to traditional methods, hybrid models effectively reduce bias and variance, improving prediction accuracy. Particularly in handling nonlinear relationships and high-dimensional data, hybrid methods enhance model robustness and generalization through advanced feature engineering, ensemble learning, and optimization techniques.

Supporting Evidence: Studies show that hybrid and ensemble methods outperform other techniques in most cases, especially in surface blasting scenarios. For example, hybrid models exhibit higher reliability in PPV prediction, with both RMSE and R² values surpassing those of traditional models.

2. Limitations of Traditional Methods

Traditional methods (e.g., regression analysis) show significant shortcomings in handling the nonlinear relationships of blasting vibration phenomena. These methods typically consider only a limited set of parameters (e.g., explosive quantity, site-specific constants, and the distance between the monitoring station and the blast location), overlooking the multifaceted nature of blasting design parameters. For instance, traditional regression models perform poorly when dealing with complex blasting design parameters such as hole diameter, number of holes, spacing, stemming height, etc.

Supporting Evidence: Research indicates that traditional methods exhibit large errors in PPV prediction, especially in complex geological conditions. In contrast, ML models (e.g., CART, ML models, and MR) demonstrate higher accuracy and reliability in PPV prediction.

3. Applications of Machine Learning in Blasting Optimization

ML techniques show great potential in blasting optimization. For example, particle swarm optimization (PSO) and extreme learning machine (ELM) demonstrate high accuracy in flyrock prediction. Additionally, hybrid methods (e.g., KELM and FB-SVR) significantly improve prediction accuracy and robustness by leveraging the strengths of multiple algorithms.

Supporting Evidence: Studies show that hybrid methods excel in flyrock and PPV prediction. For instance, the GPR model achieves the highest R² and lowest RMSE values in PPV prediction, showcasing its strong capability in handling complex datasets.

4. Future Research Directions

Despite the significant potential of hybrid ML techniques in PPV prediction, several challenges remain. Future research should focus on the following areas: developing standardized datasets, improving model interpretability and scalability, extending research to underground blasting environments, integrating real-time adaptive systems, automating blasting design optimization, and addressing environmental and social impacts.

Supporting Evidence: Research shows that hybrid models perform well in handling complex geological conditions and dynamic blasting parameters but face challenges in data quality and model complexity in practical applications. Future studies should address these issues to promote the widespread adoption of hybrid ML techniques in mine blasting.

Significance and Value of the Paper

This paper systematically reviews the application of hybrid ML techniques in PPV prediction for open-cast mine blasting, exploring their advantages, challenges, and future research directions. By combining traditional methods with advanced techniques, hybrid models demonstrate higher accuracy and reliability in PPV prediction, offering new solutions for optimizing blasting practices, reducing environmental damage, and ensuring structural safety. The findings of this paper are significant for advancing the application of ML techniques in mine blasting and provide valuable references for future research.

Other Valuable Information

The paper also details specific applications of ML techniques in blasting optimization, such as PSO, ELM, and hybrid methods (e.g., KELM and FB-SVR). Additionally, it discusses the performance of ML models in handling complex geological conditions and dynamic blasting parameters, providing practical insights for real-world applications.

By systematically reviewing the application of hybrid ML techniques in PPV prediction for open-cast mine blasting, this paper highlights their potential in improving prediction accuracy, optimizing blasting practices, and ensuring structural safety. Future research should address challenges such as data quality and model complexity to promote the widespread adoption of hybrid ML techniques in mine blasting.