Tricone rigs stop running four loops apart: perception, decision and control close on 322 blastholes at 35 m/h
Synopsis
On open-pit tricone rigs, formation perception, parameter decision, actuator control and stuck-pipe monitoring have usually run as separate functions with limited information transfer. A CCTEG Xi'an team closed them into one loop: multisource drilling data identify lithology and drillability online, case matching recommends feed and rotary speeds, robust control tracks the setpoints, and a separate safety branch watches for drill sticking. Across 322 blastholes and 5426.7 m of drilling, lithology accuracy exceeded 85%, 34 of 38 sticking events were detected, and mean penetration rate reached 35 m/h.
Interpretation
Formation perception, parameter decision, actuator control and abnormal-condition monitoring are organized into one continuous loop, with an adaptive path for normal drilling and an independent safety branch for events such as drill sticking. Rigs have usually implemented these functions independently, which limits information transfer and coordinated response. Field trials ran on a ZY250E tricone rig at an open-pit coal mine, covering 322 blastholes and 5426.7 m of drilling.
The system identifies lithology and quantifies drillability online from feed speed, rotary speed and the two motor torques, then issues recommended feed and rotary speeds. Lithology uses a random forest and multilayer perceptron fusion classifier, drillability is described as a continuous quantity rather than a discrete label, and parameter recommendation uses k-nearest-neighbor case matching over a historical case base screened for efficiency, energy per unit advance and stability. Models were trained on blastholes 1–150, which also built the case base; blastholes 151–200 served for offline simulation and 201–322 for online field validation.
Recommended values become setpoints for robust control, which treats formation-hardness variation as bounded uncertainty so actual feed and rotary speeds stay close to the recommendations. In an offline simulation on a representative hard formation, algorithm-recommended parameters raised mean penetration rate from 10 mm/s to 13.4 mm/s and cut relative energy consumption per unit advance from 2.365 to 1.29, compared with manually specified field parameters. That comparison comes from a formation whose drillability values were mainly distributed between 2.3 and 2.6, and neither feed-motor nor rotary-motor torque rose substantially.
Drill sticking is detected by a separate branch that fuses a Gaussian mixture hidden Markov model with a bidirectional LSTM conditional random field and issues stop and alarm commands once probability crosses a 70% threshold. Of 38 field sticking events, 34 were detected, a field detection rate of 89.5%, with a mean response time of about 2 s. Mean penetration rate reached 35 m/h, above the rig's 30 m/h design value; for a typical 18-m hole, mean operating time fell by about 13 min and drilling efficiency rose by more than 30%.
Perspective
Validation covers one open-pit coal mine and one ZY250E tricone rig, across 322 blastholes and 5426.7 m of drilling, with ambient temperatures down to −25 °C and maximum wind speed of 10.8 m/s.
Validation was confined to one mine and one rig model, and performance may vary with available field data, formation conditions and equipment characteristics; another mine would need retraining or recalibration of the perception and decision models, another rig would need renewed system identification and control tuning, and cross-site and cross-platform generalization remains to be evaluated.