Nonlinear automatic excavation control strategy for mining electric shovels using vision-assisted trajectory planning

Mining electric shovels are key excavation equipment in open-pit mines. To improve bucket-fill efficiency and trajectory-tracking accuracy during automatic excavation, this study proposes a vision-assisted nonlinear automatic excavation control framework. The main contribution is an integrated perception-planning-control method: (i) a binocular-vision pile-surface recognition module is used to reconstruct and fit the excavated material pile surface; (ii) a fifth-order-polynomial excavation trajectory is optimized by a volume-prediction model to improve the expected bucket filling rate; and (iii) a disturbance-observer-neural-network sliding mode controller (DOB-NN-SMC) is developed to compensate for external nonlinear excavation resistance and dynamic-model uncertainties. The RBF neural network approximates structured model uncertainties, whereas the disturbance observer estimates the remaining excavation disturbances online, thereby improving both robustness and tracking precision. Experiments on a mining-electric-shovel test bench show that the pile-surface recognition error is 1.28%–1.67%, the planned trajectory achieves an actual bucket-fill-factor error of 8%–10%, and the proposed controller reduces steady-state tracking errors by approximately 33.3%–45.2% for the hoist angle and 38.5%–53.8% for the crowd displacement compared with PID and conventional sliding-mode controllers under the tested operating conditions. These results demonstrate that the proposed framework can improve excavation trajectory generation and nonlinear trajectory tracking for automatic excavation of mining electric shovels.