An accurate prediction method of robot machining deformation error under force-position data fusion

Robots, with their high flexibility and extensive working space, have emerged as a new manufacturing paradigm following traditional machine tool production. However, due to factors such as the weak joint rigidity of their serial structures, time-varying dynamic processing loads, and workspace heterogeneity, significant workpiece deformation errors occur during machining. Force-position data directly reflect the physical interactions of the machining process and contain deeper feature information. To address this, a prediction model of robotic machining deformation error based on Autoencoder-Temporal Convolutional Network (AE-TCN) force-position data fusion is proposed. The model leverages AE to achieve multidimensional feature fusion of raw data and TCN with causal dilated convolutions to enable efficient parallel computation while capturing the temporal dependencies in the robotic machining process. The model demonstrated exceptional performance in linear milling tasks on flat plates. Four generalization tasks were designed to validate its effectiveness and strong generalization capabilities: parameter differentiation, trajectory differentiation, workpiece profile differentiation, and workpiece profile-parameter differentiation. Comparative analyses confirmed the proposed method's robustness and adaptability. Additionally, owing to the post-measurement nature of workpiece deformation errors and the lightweight design of the model, it outperforms other complex models by enabling efficient inference for online tasks, making it highly significant for real-time applications.