The design, control, simulation and animation of robotic systems heavily depend on dynamic modeling. A variety of studies have explored different dynamic modeling methodologies applied to diverse robotic mechanisms. Artificial neural networks (ANNs) have proven their value in engineering design in recent years, enhancing the understanding of complex mechanisms as well as shortening experimental periods and decreasing related expenses. This study investigates the application of various neural network algorithms for the analysis of a custom-designed three-link planar revolute–prismatic–revolute (RPR) robotic arm mechanism. Initially, the Euler–Lagrange equations of motion for the RPR mechanism are derived. Joint accelerations are then computed under different mass configurations of the robotic links, resulting in a dataset comprising 204 joint acceleration samples. Six distinct neural network models are subsequently employed to perform regression analysis on the collected data. The primary objective of this study is to analyze the relationship between joint accelerations and varying link masses under constant joint torques and forces, while its secondary aim is to present a representative application of neural networks as regression learners for the dynamic modeling of robotic mechanisms. The approach outlined in this study allows users to select appropriate neural network algorithms for use in specific applications, considering the wide range of available algorithms. Link mass variations and their effects on joint accelerations are investigated, establishing a basis for the modeling of robotic dynamics using regression-based neural networks. The results indicate that the optimizable neural network algorithm produces the best regression accuracy results, although the other models maintain similar performance levels.
Loading....