⚡ electrical engineering

Integrated perception, decision, and control for tricone blasthole drilling: System development and field validation

This paper presents an intelligent control system that integrates formation perception, decision-making, and control for tricone blasthole drilling, demonstrating through field trials a 33.6% increase in penetration rate, a 45.3% reduction in energy consumption, and effective abnormal-condition detection.

Original authors: pu liu, zujie zou, dong fan, binjie he, longpeng wang

Published 2026-10-23

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: pu liu, zujie zou, dong fan, binjie he, longpeng wang

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Deep in the open-pit mines that feed the world's energy, a massive machine known as a rotary drill rig bores holes into the earth. These holes, called blastholes, are the starting point for blasting rock to extract coal or ore. The quality of these holes determines how efficiently the rock breaks apart later, which directly impacts the safety and productivity of the entire mine. For decades, the success of this operation has relied heavily on the skill and intuition of a human operator sitting in the cab. As the drill bit moves from soft soil into hard rock, or encounters a hidden fracture, the operator must constantly guess how to adjust the machine's speed and pressure. If the guess is wrong, the drill can stall, the bit can break, or the hole can be drilled too slowly, wasting time and energy. The challenge has always been that the rock underground is unpredictable, and the machine's response to it is complex, making it difficult to automate without a system that can truly "feel" the ground and react instantly.

A team of researchers from the CCTEG Xi'an Research Institute and the China University of Mining and Technology has developed a new way to solve this problem. They created an intelligent control system that acts as a unified brain for the drill rig, linking three critical functions: sensing the rock, deciding what to do, and controlling the machine's movements. Unlike older systems that treated these tasks separately, this new approach connects them in a continuous loop. The system constantly watches multiple signals from the drill, such as how fast the bit is turning, how much force is pushing it down, and how much power the motors are using. It uses this data to identify the type of rock it is currently drilling through and to estimate how easy or difficult that rock is to break. Based on this understanding, the system automatically selects the best speed and pressure settings, and then commands the machine to follow those settings precisely, even as the ground conditions change beneath the bit.

To test this system, the researchers took it to a real open-pit coal mine and put it through its paces on a ZY250E drill rig. They did not just run the machine in a computer simulation; they drilled 322 actual blastholes, covering a total distance of over 5,400 meters. The environment was harsh, with temperatures dropping to minus 25 degrees Celsius and strong winds, providing a rigorous test for the technology. The results showed that the system could identify the type of rock with an accuracy of more than 85 percent. More importantly, it could detect when the drill was about to get stuck—a dangerous situation known as drill sticking—by noticing subtle changes in the machine's behavior before the problem became critical. When the system sensed a sticking event, it reacted within about two seconds, issuing an alarm and stopping the drill to prevent damage.

The impact on efficiency was significant. In simulations involving hard rock, the system's recommended settings increased the speed at which the drill advanced by 33.6 percent compared to manual operation by a human operator. At the same time, the energy used to drill each meter of hole dropped by 45.3 percent. In the field tests, the average drilling speed reached 35 meters per hour, which is higher than the rig's standard design target. For a typical hole that is 18 meters deep, the machine finished the job in about 13 minutes less time than before, representing an efficiency gain of more than 30 percent. The holes drilled by the system were also precise, with the diameter staying within 3 millimeters of the target and the depth varying by less than 1 percent.

The researchers also demonstrated that the system could handle the unexpected. When the drill encountered a sudden change in the rock, the system adjusted its recommendations instantly. If the rock became harder, the machine slowed its rotation and increased its downward pressure to maintain stability. If the rock softened, it sped up to maximize progress. This coordination happened automatically, without the operator needing to intervene. The system also maintained a separate safety watch that operated independently from the normal drilling controls. This safety branch monitored the machine for signs of trouble, such as a sudden drop in speed or a spike in motor torque, which are early warnings of a stuck drill. When these signs appeared, the system intervened immediately, proving that it could protect the equipment and the operator from harm in real-world conditions.

While the study was limited to one specific type of drill rig and one mine, the results suggest that integrating perception, decision-making, and control into a single system is a viable path forward for mining automation. The system proved that it is possible to move beyond simple automation, where a machine just follows a fixed set of rules, to a smarter approach where the machine understands its environment and adapts to it. By combining the ability to identify rock types, choose the right settings, and react to danger, the researchers have shown that drilling can be made faster, safer, and more efficient. The work confirms that with the right technology, the uncertainty of the underground world can be managed with a level of precision that human operators alone cannot consistently achieve.

Technical Summary: Integrated Perception, Decision, and Control for Tricone Blasthole Drilling

Problem Statement
Open-pit tricone blasthole drilling requires a coordinated closed loop of formation perception, parameter selection, actuator control, and abnormal-condition detection to ensure efficiency and safety. Conventional drill rigs typically implement these functions independently, which limits information transfer and coordinated response. In manual operations, lithology assessment, parameter selection, and handling of abnormal conditions rely heavily on operator experience. While existing research has advanced individual components—such as formation perception using machine learning, safety supervision using temporal detection, and robust control for specific subsystems—these capabilities are rarely integrated into a unified framework. This separation is particularly limiting for air-assisted tricone drilling, where formation estimates must dynamically update setpoints, actuators must track these setpoints despite uncertainty, and a safety layer must override normal control during anomalies.

Methodology
The authors developed an intelligent control system organized around a perception–decision–control framework, validated through offline simulation and field trials on a ZY250E tricone drill rig. The system architecture comprises two parallel branches: an adaptive normal-drilling loop and an independent safety-supervision branch.

Formation Perception:
- Lithology Identification: Utilizes a Random Forest–Multilayer Perceptron (RF–MLP) fusion classifier. Multisource time-series data (feed speed, rotary speed, motor torques) are processed through two branches: statistical feature extraction (mean, variance, peaks) and structural feature extraction using non-Euclidean graph embedding (directed visibility graphs).
- Drillability Assessment: Following classification, a continuous quantitative evaluation method characterizes formation drillability, updated every 3 seconds using a 5-second sliding window, to support online parameter adjustment.
Operating-Parameter Decision Making:
- Employs a K-Nearest-Neighbor (KNN) case-matching method. A historical case base is constructed offline using high-quality data (efficiency, energy consumption, stability).
- In the online stage, current formation drillability and rig state form a query vector. Similar historical cases are retrieved, and their operating parameters are weighted and fused to generate recommended feed and rotary speeds, updated every 3 seconds.
Drilling-Parameter Tracking Control:
- A robust $H_\infty$ controller regulates the feed and rotary systems.
- Formation hardness variations are modeled as bounded uncertainties within the drill-string dynamics. The controller ensures stable tracking of the recommended setpoints while attenuating the effects of formation changes and external disturbances.
Abnormal-Condition Detection and Safety Intervention:
- Operates independently of the normal control loop to detect events like drill sticking.
- Uses a temporal classification method integrating a Gaussian Mixture Model–Hidden Markov Model (GMM–HMM) and a Bidirectional Long Short-Term Memory–Conditional Random Field (BiLSTM–CRF).
- The system extracts multiscale temporal features (trend, fluctuation, complexity) from multisource data. When the estimated probability of drill sticking exceeds a 70% threshold, the system triggers immediate drilling-stop and alarm commands.

Key Results
The system was evaluated over 322 blastholes (5,426.7 m of drilling) in an open-pit coal mine under varying geological conditions and environmental factors (temperatures down to −25°C).

Parameter Decision Performance: In offline simulations of a representative hard formation, the algorithm-recommended parameters increased the mean penetration rate by 33.6% compared to manual operation and reduced relative energy consumption per unit advance by 45.3%. The improvement was attributed to better coordination between feed and rotary speeds rather than simply increased power output.
Field Validation (Normal Drilling): The system achieved a mean penetration rate of 35 m/h (exceeding the rig's design value of 30 m/h). For a typical 18-meter hole, the mean operating time decreased by approximately 13 minutes, corresponding to an efficiency increase of more than 30%. Lithology identification accuracy exceeded 85%.
Field Validation (Abnormal Conditions): The system detected 34 of 38 drill sticking events, achieving a detection rate of 89.5% with a mean response time of approximately 2 seconds.
Engineering Metrics: Hole-diameter deviation remained within ±3 mm, and depth deviation did not exceed 1%. Remote response time was below 200 ms with packet loss under 1%.

Significance and Claims
The paper claims to demonstrate the feasibility of integrating formation perception, parameter decision-making, tracking control, and safety intervention into a unified, field-scale tricone drilling system. The study establishes that a coordinated perception–decision–control framework can effectively handle the dynamic uncertainties of heterogeneous rock formations and provide timely safety interventions.

The authors note that the validation was limited to a single mine and the ZY250E rig. They explicitly state that application to other sites or equipment would require retraining or recalibration of the perception and decision models using site-specific data, as well as renewed system identification and control-parameter tuning. The paper concludes that cross-site and cross-platform generalization should be evaluated in future work.

Technical Summary: Integrated Perception, Decision, and Control for Tricone Blasthole Drilling

More like this