Introduction

The modern economy relies heavily on the production of minerals. However, the mining industry globally consumes a significant amount of energy with very low efficiency, especially during the comminution process, to reduce the size of the mined material. In extracting hard rocks, the initial step involves blasting techniques to break the ore mass into different fragment sizes, commonly known as blast fragmentation. In subsequent operations like crushing and grinding, these fragments undergo further reduction into finer particles1,2,3. These processes, collectively called comminution, are critical in mineral processing, and grinding is generally the stage that consumes the most energy in mineral processing4.

Overall, the mining operations have very low energy efficiency. For example, rock drilling has about 10% energy efficiency, rock blasting about 6%, rock crushing about 3–5%, and grinding about 1%5–7. Furthermore, rock comminution (including crushing and grinding) is associated with substantial energy consumption, representing approximately 4% of the world’s total energy usage, 53% of the total energy used in mining, and 67% of total mining expenses. In comparison, the energy consumption related to drilling and blasting is considerably lower, constituting only 2% of the total mining energy and 5% of mining costs8,9. This inefficiency in energy usage and high energy consumption leads to significant energy wastage, causing the mining industry to lag behind other industrial sectors in terms of energy efficiency. Figure 1 illustrates energy distribution in different parts of the mining processes.

Fig. 1
figure 1

Energy distribution in different mining processes8,10.

Full size image

As shown in Fig. 1, from the point of view of energy consumption, rock crushing and grinding operations include a significant part of the total energy consumption of mining and are among the most consuming industrial operations in the world. These operations not only have a substantial impact on energy usage but also on mine productivity. Today’s primary challenge in mining is the abundance of low-grade and widely dispersed mineral reserves. Reducing these low-grade reserves to liberate valuable minerals requires substantial energy input during crushing. Consequently, there is a pressing need to develop solutions that minimize energy consumption in mineral comminution operations. Based on the information presented, it can be inferred that optimizing the rock comminution process is essential for enhancing productivity and conserving energy in mining operations from extraction to processing11,12.

Since the 1970 s, it has been established that rock comminution through blasting has significant effects on subsequent operations such as loading, haulage, crushing, and grinding13. As a result, numerous studies have been conducted worldwide to enhance the efficiency of rock comminution operations. Many mines have achieved more significant energy savings and improved operational efficiency in grinding by employing a higher specific charge (powder factor) in blasting. One widely adopted approach is Mine to Mill (MTM) optimization. Researchers at the Julius Krutschnitt Mineral Research Center in Queensland, Australia, have been developing mine-to-mill optimization since 1998. “Mine to mill” is a holistic approach to extracting and processing minerals to reduce energy consumption. This approach considers the entire mining operation, from blasting to milling, to optimize reducing the size of mineral particles. The MTM approach has been successfully implemented in gold, copper, and lead-zinc mines globally, resulting in an 18% increase in operational throughput and a 20% reduction in cost14,15. Optimizing the entire mining operations using the MTM approach is highly complex, and it requires consideration of rock mass characteristics, blast energy, and energy consumption in downstream processes. Numerous studies have shown that optimizing blasting parameters can significantly lower overall mining and mineral processing costs. This is achieved by improving rock fragmentation and reducing energy consumption in downstream operations16,17,18,19. Additionally, efficient Mine-to-Mill processing of low-grade complex ores requires not only energy optimization but also effective waste management and interdisciplinary collaboration. These efforts aim to enhance the efficiency of rock breakage from blasting to grinding20. Obtaining the physical and mechanical characteristics of the rock mass necessary for this approach is challenging due to its time-consuming and costly nature. One solution is the use of indirect methods, such as analyzing drilling operation data, to determine the geomechanical characteristics of the rock mass21. The Monitoring While Drilling (MWD) approach, which utilizes drilling data, has become increasingly popular in recent years due to its systematic nature, cost-effectiveness, and real-time capabilities22. This approach will be detailed in the next section.

Monitoring while drilling (MWD)

Minerals vary widely in terms of resistance and structural characteristics. Core drilling and geophysical methods are commonly used to analyze core samples obtained from the desired reserves to assess the strength and structural properties of rock mass for drilling projects. However, these methods have limitations, such as a restricted number of samples and incomplete coverage of the entire mine. To improve the accuracy of geological data, it is necessary to increase the number of samples and information required to describe the rock mass accurately. Achieving this is not only time-consuming but also very expensive. The MWD method aims to assess the characteristics of the rock mass based on drilling data22. Figure 2 provides an overview of the MWD method.

Fig. 2
figure 2

Overview of the MWD.

Full size image

The rock mass description involves determining various geological and mechanical parameters such as uniaxial compressive strength (UCS), Young’s modulus, and tensile strength23. MWD is a technique that aims to provide a real-time and relatively accurate description of rock mass characteristics by examining drilling variables such as weight on the bit, torque, vibrations, rotation speed, penetrating rate, and specific energy24. Based on recent studies, MWD data can be utilized for mineral description, rock type classification, estimation of physical and mechanical rock properties (such as compressive strength, hardness, and weathering degree), blast optimization, mine planning, design, and production control. The collected data can also be utilized in the “mine-to-mill” approach. MWD data contains two main types of information about the rock mass:

  1. 1.

    Properties of intact rock.

  2. 2.

    Structural features (discontinuities).

The MTM approach primarily utilizes information related to the properties of intact rock, which can help measure rock comminution efficiency during blasting, crushing, and grinding operations (Because the size of the particles in the communication process is much smaller than the size of the blocks in the rock mass)25,26. This method can significantly reduce the high uncertainty associated with field data. MWD systems measure drilling performance data in real-time using sensors. Although it has long been recognized that geological conditions can be significantly evaluated through MWD data, these data are rarely used to characterize the rock mass in activities such as blast design or the MTM approach. Nevertheless, it is commonly understood that drilling parameters are related to the physical and mechanical characteristics of the rock mass. In general, Fig. 3 shows the results obtained from the MWD technique.

Fig. 3
figure 3

Results from the MWD technique.

Full size image

To meet the needs of the mining cycle and adapt to changes in data collection and processing techniques, it is necessary to use new methods to extract valuable geological and drilling performance information from MWD data. This requires a comprehensive understanding of the drilling process and establishing the connection between its operational parameters and the physical and mechanical characteristics of the rock mass27. Researchers have recently utilized acoustic and vibration waves generated during the drilling process to predict rock masses’ physical and mechanical properties28. The intended waves in these studies are acoustic and vibration waves produced from drilling rocks, which are used for collection, recording, processing, and analysis for various purposes. The data obtained through the analysis of acoustic and vibration signals can provide essential information to advance the “mine-to-mill” approach in mining operations. Due to the non-destructive nature of these methods, they can be applied in various mining operations without interrupting ongoing activities, allowing for real-time evaluations29. The following section will discuss the application of Acoustic Emission Methods in drilling operations.

While the concept of MTM emphasizes integrated optimization of the entire mining and comminution chain (from drilling and blasting to loading, crushing, and milling) this study focuses specifically on the drilling stage, as a fundamental and influential component of the overall process. The findings presented here provide a foundation for understanding how rock properties affect drilling performance, which can subsequently influence downstream efficiency.

Application of acoustic emission methods in the drilling operation

The rock drilling operation is a crucial stage in mineral extraction, and various factors can affect its performance. These factors can be categorized into three groups: parameters related to the drilling machine specifications, operational parameters, and parameters associated with the physical and mechanical characteristics of the rock mass. Figure 4 illustrates the parameters that affect drilling operations.

Fig. 4
figure 4

Key factors influencing drilling operations.

Full size image

Parameters influencing drilling operations can be categorized into controllable and uncontrollable. Controllable parameters can be identified and managed by assessing various conditions. In contrast, uncontrollable parameters are determined by natural geological factors and cannot be modified by human intervention. Therefore, developing a comprehensive method to simultaneously evaluate these parameters’ effects is crucial for optimizing drilling operations.

Drilling parameters, including the properties of the rock and operational drilling conditions, are crucial for the efficiency of subsequent mineral production processes. These parameters affect the quality and quantity of the extracted materials and directly impact the optimization of downstream operations, such as blasting, loading and haulage, primary and secondary crushing, and milling. For example, the rate of penetration (ROP) can provide insights into rock hardness, which is essential for designing blast patterns and selecting appropriate crushing equipment. Furthermore, data obtained from drilling can be utilized to monitor and optimize each of these stages30,31.MWD is a valuable source of real-time geotechnical data that can significantly enhance the efficiency and integration of mine-to-mill operations.

The Mine-to-Mill (MTM) optimization framework aims to enhance the efficiency of the entire mining value chain, from drilling to milling, by systematically improving each stage of operation. A key aspect of this approach is drilling, which not only initiates the excavation process but also provides essential data that inform decisions related to blasting, loading & haulage, crushing, and grinding. Drilling performance is influenced by both controllable parameters and uncontrollable factors that depend on intrinsic geological features, including rock hardness, structural characteristics, porosity, and the extent of weathering. Additionally, machine-specific attributes like bit type, equipment size, and engine capacity significantly affect drilling outcomes. Thus, a comprehensive optimization strategy must evaluate both controllable and uncontrollable factors to maximize drilling efficiency and its downstream impact. Drilling parameters serve not only as operational inputs but also as diagnostic indicators of rock mass characteristics, playing a crucial role in optimizing subsequent processes. Variables such as the ROP, torque, and vibration response correlate indirectly with rock strength, abrasiveness, and fracture patterns—properties vital for blast design, crusher selection, and energy modeling for comminution (Kanchibotla et al., 2002; Araneda et al., 2020)25,32,33,34. The integration of MWD systems has further enhanced the utility of drilling data, enabling real-time decision-making during operations and supporting predictive modeling efforts within the broader MTM framework. Although this study primarily concentrates on optimizing the drilling stage through controlled laboratory experiments and parameter analysis, the methodology establishes a foundation for scalable integration across the mining value chain. It emphasizes the strategic importance of drilling-derived data in optimizing each stage, from rock breakage to final particle size reduction. A key element of MTM optimization involves categorizing parameters into two main groups: (A) parameters related to rock mass characteristics, typically collected during drilling operations, and (B) parameters associated with the operational settings of downstream processes. Category A parameters (many of which can be estimated or inferred from MWD and drilling response data) provide the geotechnical context necessary for optimizing subsequent processes. Conversely, Category B parameters are operationally defined and usually derived from engineering design, planning, and control frameworks. The relevance and availability of these parameters across various MTM stages are detailed below. In the blasting phase, the goal is to achieve controlled fragmentation with minimal overbreak. Category A parameters, such as rock strength, density, structure, the presence of cavities, and the degree of weathering, can be partially inferred from drilling metrics like ROP and torque. These inputs help define blast geometry and energy distribution. Category B parameters (such as powder factor, block size, explosive type, hole condition, initiation sequence, and overall blastability) are determined during the design and execution stages34,35,36,37,38. When aligned with Category A data, these parameters enable precise control over fragmentation outcomes, which directly impacts downstream efficiency. In the loading and haulage stage, optimization focuses on reducing cycle time and matching equipment capacity with material characteristics. Rock density and swell factor (both Category A parameters) affect the shovel fill factor and truck payload efficiency and can be estimated using drilling and core sample data. Category B variables in this stage include equipment compatibility, haul distance, and operational constraints such as bench height and pit geometry39,40,41,42,43. Utilizing accurate rock mass data facilitates better scheduling, fleet assignment, and cost control. During the crushing phase, the objective is to maximize throughput while minimizing energy consumption. Rock-specific properties such as strength, density, and abrasivity—central to Category A—can be inferred through MWD outputs and drilling response profiles. These insights are critical in selecting crusher types (jaw, cone, tertiary), defining throughput targets, and adjusting crusher settings to manage varying material hardness and fragmentation size44,45,46,47. Category B parameters, including crusher configuration and specifications for feed and output size, are essential design elements that must be integrated with rock properties for optimal performance. In the milling stage, optimization targets grindability and circuit efficiency. Key rock parameters can be partly inferred from drilling data such as ROP, torque, and MWD signals. These estimates support adjustments in Category B parameters like feed size, mill speed, and throughput to enhance performance.

Rock type plays a crucial role in optimizing MTM, as different lithologies respond differently during the drilling, blasting, crushing, and milling stages. Granite, known for its high strength and abrasiveness, poses challenges such as low penetration rates, increased wear on drill bits, and high energy demands during crushing. In contrast, marble has moderate strength and a uniform texture, allowing for smoother drilling and more predictable fragmentation, which enhances efficiency in subsequent processes. Travertine, with its high porosity and weaker structure, enables rapid drilling with minimal energy input. However, its irregular fracture patterns and lower bulk density can lead to variability in crushing and milling performance. Understanding these differences in behavior is essential for adjusting operational parameters and improving overall process efficiency48,49,50,51,52.

Figure 5 illustrates a summary of the impact of MWD data on various downstream operations.

Fig. 5
figure 5

Parameters influencing drilling operations and their impact on downstream operations.

Full size image

As clearly stated in Fig. 5, ROP is the most important drilling parameter, directly affecting all mining operations. It indicates drilling efficiency and rock characteristics, influencing blast design, crushing, and milling performance. Optimizing ROP by adjusting drilling parameters improves productivity, reduces energy use, and ensures consistent material quality. Maintaining optimal ROP enhances safety, lowers costs, and maximizes resource utilization throughout mining operations.

Drilling methods are chosen based on geological conditions, rock types, and operational needs. The primary drilling systems include rotary, percussion, and rotary-percussion methods, each designed for specific rock types and conditions. The rotary drilling method utilizes a rotating drill bit that grinds and removes material to form boreholes. It is highly efficient in soft to medium-hard rocks and is commonly employed in mineral exploration and resource extraction. The rotary action ensures a high penetration rate, particularly in less resistant materials. It is preferred in mining and civil engineering projects for its precision and adaptability to different geological formations. Percussion drilling, or hammer drilling, involves a heavy drill bit repeatedly lifted and dropped, delivering significant impact force to fracture and break up the substrate. This striking action allows for faster drilling in harder rocks. However, the penetration rate is slower than rotary drilling, especially in soft rocks. Rotary-percussive drilling combines both rotary and percussion actions, offering versatility for drilling through various soil and rock types. It is particularly effective for medium-hard to hard rocks, where the combination of grinding and hammering enables efficient material removal and deeper penetration53,54. Rocks are generally categorized into four groups based on their Uniaxial Compressive Strength (UCS). Soft rocks, with a UCS of less than 20 MPa, allow for faster drilling with high penetration rates. Rotary drilling is the most efficient method for these rocks, facilitating quick material removal. Medium-hard rocks, ranging from 20 to 50 MPa in UCS, typically require rotary-percussive drilling systems, such as Top Hammer (TH) or Down-the-Hole (DTH) drills, which balance penetration rate and hole straightness54,55,56.

For hard rocks, with UCS between 50 and 120 MPa, percussion drilling, especially DTH, is most effective due to the hammering action that efficiently fractures tough materials compared to rotary drilling alone. Very hard rocks, with UCS greater than 120 MPa, present significant drilling challenges. Rotary-percussive drills are most effective in these conditions, as their combined rotary and percussion actions help maintain a suitable penetration rate even in extremely hard rock formations. When selecting a drill, several factors must be considered, including the UCS and geological characteristics of the rock. Rotary drilling is ideal for soft rocks, while harder rocks benefit from percussion or rotary-percussive drilling systems. The required drilling depth also influences drill choice. DTH or rotary-percussive drills are preferable for deeper holes due to their ability to maintain hole straightness and achieve greater depths. Operational factors such as energy availability, site accessibility, and project requirements are crucial in the selection process. Drilling efficiency is commonly evaluated by the ROP, which varies with rock type. Penetration rates are highest in soft rocks and decrease progressively through medium, hard, and very hard rocks, where the drilling rate diminishes significantly. The performance of percussive drills is often assessed in terms of energy transmission from the piston to the drill bit, and factors such as fuel consumption, overall production capacity, and maintenance needs must also be considered when selecting the optimal drill53,55,56.

As mentioned in previous sections, the initial step in the mine-to-mill approach involves enhancing the comminution process, which requires a thorough understanding of the rock mass characteristics.

Throughout the years, numerous studies have employed the acoustic emission method to assess the physical and mechanical characteristics of rock mass during drilling operations57. Jung et al. (1994) utilized acoustic wave characteristics to predict rock hardness and drillability. Their experiments revealed a correlation between rock hardness and the parameters of the emitted acoustic waves58. In 2003, Futo et al. conducted research to optimize rock drilling through acoustic signals. They developed a device to measure acoustic waves during drilling and discovered that these signals could indicate the type of rock, rendering them suitable for rock detection59. Vardhan et al. have also contributed significantly to this field, conducting various studies on acoustic wave applications in rock drilling. In 2007, they examined the acoustic waves emitted during drilling tests on rocks with varying compressive strengths, demonstrating a relationship between these waves and characteristics such as compressive strength and rock abrasivity60. Gradl et al. (2008) investigated the relationship between different drilling bits and the emitted sound waves by analyzing frequency data collected during drilling operations. Their findings indicated that distinct drills produce unique noise data61. These studies illustrate the potential for identifying drilling bit types and diagnosing operational issues through acoustic wave analysis during drilling processes. In 2009, researchers explored the relationship between drilling noise and rock properties, such as compressive strength, using laboratory-scale hammer drills. This study demonstrated that rock resistance could be predicted based on the sound level emitted during drilling operations62. The results suggest a clear linkage between sound levels produced during drilling and rock characteristics, as higher sound levels correspond to increased rock compressive strength. Kumar et al. (2010) examined the physical and mechanical properties of rocks and their relationship with emitted sound levels during drilling63. They attempted to establish mathematical relationships between various rock characteristics and parameters such as penetration rate, drill rotation speed, and emitted sound level64. Additionally, in 2011, researchers explored this relationship further using multivariate regression, leading to models that accurately predicted desired parameters65. The results indicate that the sound levels generated during drilling operations correlate with the rock’s characteristics, providing a basis for predicting the physical and mechanical parameters of the rock. In 2013, Kumar et al. estimated the physical and mechanical characteristics of various rocks using methods like multivariate regression and artificial neural networks based on the sound levels produced during drilling66. The input parameters included drilling speed, penetration rate, drill bit diameter, and the sound levels generated while drilling different rocks. The outputs encompassed uniaxial compressive strength, Schmidt number, density, tensile strength, and porosity of the rocks. Results indicated that with fixed drilling speed, bit diameter, and penetration rate, the equivalent sound level produced during drilling increases with rock density, hardness, and strength (including uniaxial compressive strength, Young’s modulus, and tensile strength) while decreasing with increasing porosity. In 2014, Karakus et al. investigated acoustic waves produced during the collision of drill bits with rock, examining their relationship with variations in drilling operational conditions and parameters. Their findings demonstrated a direct correlation between the amplitude of acoustic signals and drilling parameters, suggesting potential applications for estimating the depth of cut, weight on the bit, and torque using signal amplitude67. Subsequent studies continued to expand upon these concepts. In 2018, Yari et al. investigated the correlation between the physical and mechanical characteristics of igneous rocks and the acoustic signals emitted during drilling, finding that most rock properties could be predicted using the dominant frequencies of the emitted signals68. They also explored sedimentary rocks the same year, discovering reliable mathematical relationships between rock characteristics and the dominant frequencies in the emitted signals69. In 2019, they extended this research to carbonate rocks, further confirming that the acoustic signals emitted during drilling operations are also related to the physical and mechanical characteristics of these rocks70. Khoshouei et al. (2020) recorded acoustic signals during drilling operations of sedimentary, metamorphic, and igneous rocks, calculating the RMS values of these signals and comparing them with the mechanical properties of rock samples. Their results indicated that RMS values for each rock type have distinct ranges and can be utilized to estimate rock type and its physical and mechanical characteristics71. In 2021, they studied a method for predicting the geomechanical properties of hard rocks, performing statistical analyses on recorded sound and vibration signals to explore the relationship between rock properties and sound parameters. Their findings established significant correlations between rock characteristics and sound and vibration signals during drilling operations72. In 2021, Rafezi and Hassani studied the link between drilling signals and tricone bit wear in mining operations to enhance automation, efficiency, and safety. They developed a qualitative method for classifying bit wear states through real-time vibration and motor current signal analysis. Their research identified critical failure vibration frequencies, regardless of geological conditions, and explored how bit design parameters affect them, contributing to better bit wear monitoring and failure prediction73. In 2022, Khoshouei et al. developed a method for evaluating rock abrasivity using acoustic emission techniques during measurement while drilling (MWD). They analyzed acoustic and vibration signals from igneous rock samples and established a predictive equation, demonstrating the practicality of real-time assessments of rock properties during drilling operations28. In 2023, Rafezi and Hassani developed a method for monitoring tricone drill bit wear and predicting failures in surface mining. Their research analyzed in-situ vibration signals in the time-frequency domain to aid AI models for autonomous drilling. They identified wavelet packet energy distribution as a critical wear indicator and created tailored backpropagation artificial neural network (ANN) models to classify drill bit health and forecast failures9. In 2024, Zheng et al. investigated the processing of acoustic emission (AE) signals to better understand fracture mechanisms in rocks. They demonstrated a correlation between the dominant frequency distributions of AE signals and specific fracture modes. This approach allows for quantitatively identifying damage mechanisms, providing valuable insights for diagnosing failures in drilling operations74. Most recently, in 2024, Khoshouei et al. focused on monitoring while drilling (MWD) and optimizing drilling operations using acoustic emission techniques (AET). Their research investigated the potential of analyzing vibroacoustic signals generated during drilling to assess specific energy (SE). By capturing and analyzing acoustic and vibration signals across different domains, the researchers established correlations between vibroacoustic features and SE. These findings suggest a promising approach for developing an industrial monitoring system capable of detecting excessive energy consumption and indicating when a drill bit approaches the end of its useful life through vibroacoustic sensors on drilling equipment75.

The application of acoustic wave propagation methods during drilling operations has proven effective for estimating and predicting the physical and mechanical properties of rock masses. This approach, renowned for its accuracy, cost-efficiency, and real-time capabilities, has been widely embraced by researchers as a reliable tool. In the present study, acoustic emission techniques integrated with intelligent algorithms are employed to predict drilling penetration rates, primarily aimed at characterizing the rock mass, thus optimizing the overall mine-to-mill process. This study presents a novel approach integrating real-time drilling acoustic emissions with machine learning to enhance rock mass characterization within the Mine-to-Mill framework. Unlike conventional methods that rely on time-consuming and costly laboratory analyses, this approach enables in-situ estimation of rock properties, facilitating immediate adjustments in drilling and blasting strategies. The application of Support Vector Machine (SVM) modeling to acoustic and vibration signals provides a robust predictive tool for optimizing fragmentation, ultimately reducing energy consumption in comminution. This innovative methodology bridges the gap between drilling data acquisition and downstream process efficiency, offering a practical and scalable solution for cost-effective mining operations.

Methodology

30 rock samples, including granite, marble, and travertine, were selected to study the impact of penetration rate on vibroacoustic signals. These samples, representing various physical and mechanical properties, were collected from different mines and prepared for drilling tests (Fig. 6). After collection, the rock samples were meticulously cut into 10 cm cube-shaped blocks to facilitate the subsequent drilling tests.

Fig. 6
figure 6

tested rock samples with identification codes, types, and commercial names.

Full size image

Drilling experiments utilized a small-scale rotary drilling machine, recording three types of signals: (i) the acoustic signal from the interaction between the drill bit and the rock, captured with a condenser microphone; (ii) the ambient sound pressure level (SPL) measured using a digital sound level meter; and (iii) vibration signals from the drilling machine’s axis, monitored by a 3-axis accelerometer sensor. All tests were performed under standardized operating conditions to enhance the reliability of the vibroacoustic data, ensuring that the recorded signals reflected the intrinsic characteristics of the rock samples. Figure 7 illustrates the small-scale rotary drilling machine, signal recording equipment, and the drill bit used in the experiments, along with a summary of the specifications for the recording equipment and the operational conditions maintained.

Fig. 7
figure 7

Setup of small-scale rotary drilling machine and signal recording equipment.

Full size image

Acoustic signals during drilling arise from three primary sources: the drilling machine, the interaction between the drill bit and rock, and ambient noise. To predict the ROP, identifying the frequency range associated with operational parameters is essential. Acoustic signals were recorded under varying conditions, including ambient noise, drilling machine operation, and drilling with cooling fluid. The data were analyzed using Short-Time Fourier Transform (STFT), identifying frequencies above 50% of the maximum signal amplitude. Under ambient conditions, signals reached − 73 dB with 0 to 200 Hz frequencies. When cooling fluid was added, the range expanded to 1200 Hz. The drilling machine produced stronger signals, peaking at −53 dB and frequencies up to 1500 Hz, with 1500 Hz being the highest, unrelated to the drill bit-rock interaction. Signal analysis included time-domain peak values, STFT for time-frequency spectra, and FFT for isolating dominant frequencies, particularly in the 6500–10,000 Hz range, which correlated with operational changes. Vibration signals were recorded using a 3-axis accelerometer and pre-processed with a bandpass filter (2–20 Hz) to eliminate noise. Frequencies below 2 Hz were related to machine vibrations, while those above 20 Hz were considered noise. Time and frequency domain analyses and FFT were used to assess vibration data. Sound pressure levels were also measured, providing critical data for further study.

A total of 336 experiments were conducted on rock samples. According to the described methodology, the dataset contains 336 records, each featuring 23 input parameters and one output parameter representing each experiment’s penetration rate. The input parameters include various acoustic and vibration signal features, as illustrated in Fig. 8.

Fig. 8
figure 8

Statistical description of the input parameters used in the model.

Full size image

The acoustic signal parameters consist of the maximum acoustic signal power, the count of threshold-level crossings, crest factor, kurtosis, peak signal value, mean, root mean square (RMS), skewness, and standard deviation. These parameters were extracted separately for two frequency bands: 5512.5–8268.75 Hz and 8268.75–11,025 Hz. These frequency ranges correspond to two selected wavelet nodes relevant to rock drilling operations. The vibration signal parameters examined include the maximum vibration amplitude and the RMS of the vibration signals.

Results

Introduction to SVM method

The Support Vector Machine (SVM) is a supervised learning model that partitions data into distinct regions separated by a linear boundary76. This algorithm is handy for classification and prediction tasks, demonstrating remarkable performance in various domains, including text categorization, image recognition, and geotechnical applications. SVMs identify a hyperplane that divides data points within a dataset into two segments, maximizing the margin between different classes77. SVM aims to establish the optimal line or decision boundary that can effectively separate classes in an n-dimensional space, enabling accurate classification of new data points78. SVM is a generalized linear classifier based on the Vapnik–Chervonenkis (VC) dimension theory. It was initially developed by Vladimir Vapnik for linear models in 1963 and later expanded to accommodate non-linear training in 1995. Today, it is one of the most prevalent and advanced machine learning techniques79,80. One of the critical advantages of SVM is its ability to handle high-dimensional spaces and its effectiveness in scenarios where the number of dimensions exceeds the number of samples79. This characteristic makes SVM particularly suitable for predicting the ROP during drilling operations, as it can manage the complex relationships between acoustic signals and geological characteristics81. Furthermore, SVM’s use of kernel functions allows it to create non-linear decision boundaries, enhancing its flexibility in modeling intricate patterns in the data. In summary, SVM’s robustness, versatility, and ability to handle high-dimensional data make it a compelling choice for predicting penetration rates in drilling operations, particularly when integrated with acoustic signals, thereby optimizing the mine-to-mill process effectively82.

Model details and results

The Support Vector Machine (SVM) model employed in this study was configured with a set of carefully selected hyperparameters to predict the ROP based on acoustic signal features:

  • Kernel Function: A linear kernel was utilized, offering an efficient decision boundary for linearly separating the data.

  • Box Constraint (C): A box constraint value “1” was applied to balance margin maximization and classification accuracy. This setting allows the model to maintain flexibility while controlling the trade-off between underfitting and overfitting.

  • Kernel Scale: The kernel scale parameter was set to ‘auto’, enabling the algorithm to automatically adjust the scale of the input features, thus improving the model’s adaptability to different feature distributions.

  • Polynomial Order: Although a polynomial kernel of degree 3 was initially tested, the linear kernel was selected as the optimal configuration for this particular application.

  • Standardization: Input feature standardization was disabled (false), as it was determined unnecessary for the model’s performance in this case, given the linear nature of the selected kernel function.

  • Learners: The default SVM learner was applied using a one-vs-one classification scheme, which allowed the model to effectively distinguish between various rock types and accurately predict the ROP.

These hyperparameters were optimized using cross-validation techniques to ensure the model’s robustness and generalization capabilities without overfitting. In Fig. 9, the comparison between the SVM model’s predicted values and the actual measured data is presented. The figure illustrates how well the model captures the patterns in the data, demonstrating its predictive accuracy.

Fig. 9
figure 9

Comparison of actual and predicted drilling penetration rates using the SVM model.

Full size image

Performance evaluation of the SVM model

The performance of machine learning models, particularly in regression tasks like predicting the ROP, can be assessed using a range of statistical metrics. These metrics offer a comprehensive view of how well the model can capture the underlying patterns in the data and generalize to unseen examples. The critical evaluation criteria used in this study include:

  • Coefficient of Determination (R²): This metric quantifies the proportion of variance in the dependent variable explained by the independent variables. An R² value close to 1 indicates a firm fit, showing that the model effectively captures the variability in the data.

  • Mean Absolute Percentage Error (MAPE): MAPE evaluates the error in percentage terms, providing insight into the model’s relative prediction error. It is beneficial when comparing the performance across datasets with different scales, as it normalizes the error.

  • Root Mean Squared Error (RMSE): RMSE measures the magnitude of the prediction errors by calculating the square root of the average squared differences between predicted and actual values. This metric gives greater weight to more significant errors, making it more sensitive to outliers than other metrics.

  • Mean Absolute Error (MAE): MAE represents the average absolute differences between the predicted and actual values, offering a simple and interpretable metric for understanding the model’s overall accuracy. It is less affected by extreme values than RMSE.

  • Mean Squared Error (MSE): This metric calculates the average of the squared differences between predicted and actual values, penalizing more significant errors more heavily. MSE often focuses on minimizing substantial deviations in the model’s predictions.

These metrics were applied to evaluate the predictive performance of the SVM model in this study. Table 1 presents the results, showing the model’s accuracy during the training and testing phases.

Table 1 The performance evaluation results of the SVM model.
Full size table

Figure 10 presents a histogram of the prediction errors, providing a detailed assessment of the model’s error distribution. This histogram illustrates the difference between the actual values and the predicted values (𝑦_true − 𝑦_predicted), giving a comprehensive view of how the errors are distributed.

Fig. 10
figure 10

Histogram of Prediction Errors for the SVM Model.

Full size image

In Fig. 9,the x-axis represents the prediction errors, while the y-axis indicates the frequency of each error value. Most errors fall within the range of [−0.2, 0.1], demonstrating the model’s high accuracy. The distribution appears left-skewed, suggesting that the model tends to underestimate actual values slightly in some instances. Additionally, a few outliers beyond 0.3 and below − 0.3 indicate significant deviations in certain predictions. The mean error is close to zero, which shows minimal bias in the model’s predictions. The concentration of errors around zero reinforces the model’s reliability, though some larger errors may be attributed to data noise or the model’s inherent limitations. The analysis based on the histogram in Fig. 9 offers deeper insight into the error distribution, complementing the statistical metrics in Table 1 and confirming the model’s capability in accurately predicting ROP.

The performance evaluation of the SVM model highlights its strong predictive capabilities in estimating the ROP based on acoustic signals. The model achieves a high R² value of 0.976 on the training dataset, reflecting its ability to accurately capture the complex relationships between the input features and the ROP. Although the R² value on the test set drops to 0.808, it still demonstrates a robust generalization capacity, indicating that the model performs well on previously unseen data. Regarding error metrics, the MAPE, RMSE, and MAE values show that while there is a moderate increase in prediction error on the test set, the model remains within acceptable limits. Specifically, the test set MAPE of 29.52% suggests that, while imperfect, the model’s predictions are accurate enough to provide valuable insights. The relatively low RMSE and MAE values for the training and test sets further support the model’s ability to minimize significant errors, maintaining an impressive level of precision. These results indicate that the SVM model offers considerable potential for improving drilling efficiency through better prediction of ROP. Its performance, particularly on the training set, underscores the capability of machine learning techniques to model complex relationships in real-world datasets. Furthermore, the reasonable performance on the test set suggests that the model is well-positioned for practical applications, such as optimizing drilling operations and reducing costs associated with incorrect penetration rate estimations. This initial solid performance points toward the feasibility of using such a model for real-time decision-making in drilling operations, significantly optimizing the mine-to-mill process. In future studies, machine learning (ML) models could be leveraged to optimize various mining operations, including drilling, blasting, and crushing, by incorporating diverse input parameters such as geological data, material properties, and operational conditions. These models have the potential to enhance prediction accuracy through the use of advanced algorithms like deep learning and ensemble methods83,84,85,86,87. Furthermore, real-time data from monitoring systems, such as Measurement While Drilling (MWD), can be integrated into these models for dynamic adjustments during operations. Additionally, a sophisticated tool utilizing the latest technologies, including the Industrial Internet of Things (IoT) and cloud computing, could be developed to facilitate continuous monitoring and data collection throughout the mining stages. IoT devices can provide real-time feedback on operational parameters, optimize equipment performance, improve safety, and reduce operational costs, ultimately contributing to more sustainable and energy-efficient mining practices88,89.

Conclusion

Enhancing energy efficiency and reducing operational costs are fundamental challenges in mining, especially within the comminution stage, where crushing and grinding account for over 50% of energy consumption and more than 60% of total operational expenses. Given the significant impact of this stage on overall profitability, the Mine-to-Mill (MTM) approach has become a critical strategy for optimizing mining processes. The MTM framework underscores the importance of considering the entire value chain, particularly the influence of upstream operations such as drilling and blasting on the efficiency of downstream processes like comminution. Improving fragmentation through better blasting requires less energy during the crushing and grinding stages, leading to substantial cost savings and enhanced operational performance.

A critical factor in realizing MTM optimization is accurately characterizing the rock mass before blasting. The physical and mechanical properties of the rock directly influence how it will fragment and, in turn, the energy efficiency of the comminution process. Traditional methods of determining these rock properties are often costly, time-consuming, and need more real-time applicability. To address this, modern approaches such as Monitoring While Drilling (MWD) and acoustic emission monitoring offer promising alternatives by providing real-time data on rock conditions during drilling operations. These methods allow for the immediate adjustment of drilling and blasting techniques, enhancing the overall efficiency of the mining cycle.

This study analyzed acoustic and vibration signals emitted during drilling to predict rock mass characteristics. Integrating these signals with machine learning techniques, specifically the Support Vector Machine (SVM) model, demonstrated strong predictive capabilities in estimating drilling penetration rates, which are closely linked to rock properties. The model achieved impressive results, with R² values of 0.976 during training and 0.808 during testing. These figures indicate a high level of accuracy in capturing the complex relationships between operational parameters and rock characteristics. A more detailed performance evaluation further underscores the model’s reliability, showcasing MAPE values of 4.36% (training) and 29.52% (testing), RMSE values of 0.0486 (training) and 0.141 (testing), and MAE values of 0.021 (training) and 0.103 (testing).

The ability to predict rock properties using acoustic emission techniques, combined with the robustness of the SVM model, has significant implications for the MTM strategy. By providing real-time information on rock conditions, this approach enables more precise control over the blasting process, leading to better fragmentation and reduced energy consumption during comminution. The results of this study reinforce the potential of integrating advanced monitoring technologies and machine learning models to optimize mining operations, ultimately driving down operational costs while improving efficiency across the entire mine-to-mill process. Applying acoustic and vibration signal analysis and intelligent modeling techniques like SVM offers a powerful tool for improving the MTM approach. This study primarily focused on drilling and its impact on downstream processes, such as comminution. However, it is essential to recognize that fully realizing the potential of the Mine-to-Mill approach requires considering all stages of the mining process, including blasting and downstream operations like grinding and flotation. Future research will broaden this investigation to include these additional processes, leading to a more comprehensive optimization strategy for the entire mining cycle. The findings demonstrate the effectiveness of using real-time drilling data to predict rock mass properties, enhancing the overall energy efficiency of mining operations and supporting the long-term goal of reducing costs associated with the comminution process. As the mining industry continues to seek innovative solutions to its energy and cost challenges, this approach holds considerable promise for widespread adoption. This study demonstrates the effectiveness of integrating acoustic emission monitoring with machine learning for rock mass characterization; however, certain limitations must be acknowledged. The model’s accuracy is affected by variations in drilling conditions, equipment specifications, and geological heterogeneity, which may require additional calibration for broader applications. Furthermore, the dataset used in this study is restricted to specific drilling environments, highlighting the need for further validation across diverse geological settings. Future research should focus on expanding the dataset, incorporating additional rock properties, and exploring advanced deep-learning techniques to improve predictive accuracy.