A convolutional neural network method to obtain the dynamic model parameters of a three-linear-axis Cartesian robot

Abstract Parametric identification plays a crucial role in improving the performance, accuracy, robustness and repeatability of Cartesian robots in advanced manufacturing processes applications, assembly and material handling. This paper presents an innovative convolutional neural network (CNN) architecture for the parametric identification of 18 critical dynamic model parameters in a three-linear-axis Cartesian robot. The dynamic model is obtained by the Euler–Lagrange motion equations from an analysis of lumped parameters such as friction, inertia, mass and stiffness of the design of a three-axis linear Cartesian robot. The proposed CNN architecture is trained and validated using experimental data, achieving a parametric identification accuracy of approximately 98%. The identified parameters are further applied to optimize the robot’s dynamic response, demonstrating improved trajectory tracking and reduced energy consumption in manufacturing tasks. This work bridges the gap between theoretical modeling and practical application, providing a robust framework for enhancing robotic system performance in advanced manufacturing environments.