Zhenyang Lv, Yi Wang, Guangpeng Zhang, and Yongchang Li
Robot grinding, natural frequency prediction, HO-BP neural network model, processing posture selection
Industrial robots are increasingly deployed for precision processing tasks, yet their natural frequency exhibits significant variation with operational configuration. Identifying optimal postures within the workspace is thus critical for processing quality. Theoretical models relying on simplifications often yield inaccurate predictions, while experimental modal analysis cannot feasibly characterise all postures. This study addresses these limitations by developing a novel neural network model—the HO-BP neural network model— to predict natural frequencies for grinding robots, enabling data- driven posture optimisation. Our methodology establishes the joint angle–natural frequency relationship through theoretical analysis and conducts experimental modal characterisation via impact hammer testing across 270 configurations within the workspace. The rigorously validated HO-BP network subsequently predicts posture-dependent natural frequencies across the entire operational domain. By integrating these predictions with belt grinder operating frequencies, the framework generates natural frequency distribution maps that guide resonance-avoiding posture selection. Experimental validation confirms significantly reduced vibration amplitudes (≤40%) in model-identified safe zones, demonstrating the framework’s efficacy in enhancing grinding stability and surface quality.
Important Links:
Go Back
