Cutting performance optimization of laboratory scale rotary rock cutting head by application of machine learning methods

This thesis investigates the cutting performance of a laboratory-scale cutting head equipped with conical picks and develops a data-driven approach to reduce specific energy and machine-frame vibration. Unrelieved cutting was carried out under a full factorial 3×3×3 design, resulting in 243 tests. Power demand, traverse speed, bi-axial vibration (X and Z), and actual depth of cut were recorded. The X direction, aligned with cutting, displayed the highest vibration, and two thresholds (≈86 mm/s and ≈141 mm/s) separated vibration behavior in line with material strength. Supervised machine learning models were trained to relate operating parameters to specific energy and vibration. Tree-based ensembles predicted specific energy with high accuracy, yielding cross-validated R² values of 0.93–0.97, with depth of cut identified as the most influential factor. Vibration prediction achieved R² values of about 0.32–0.65, with depth primary and rotational speed secondary. A constrained Particle Swarm Optimization (PSO) routine provided feasible parameter recommendations, supporting adaptive and vibration-aware selection of operating settings.