Blind-Kriging based natural frequency modeling of industrial Robot

Abstract

High-precision assembly conditions tend to necessitate consideration of the vibration modes of industrial robots. The modal characteristics of complex systems such as industrial robots are highly nonlinear. It means that mechanics experiments and finite element methods (FEM) to evaluate such features are usually expensive. Surrogate models combined with simulation-based design are widely used in engineering issues. However, few investigations apply surrogate models to industrial robots' modal analysis. We propose a practical scheme, i.e., the Blind-Kriging (KRG-B) based natural frequency prediction of the industrial robot, utilizing the Latin Hypercube Sampling (LHS) technique. A reliable dataset with 120 samples is generated for surrogate models based on the FEM. Then, the fourteen surrogate models with different optimization algorithms are evaluated to identify the optimal model for the natural frequency. In addition, the accuracy and robustness of the optimal surrogate model are investigated under different training samples. KRG-B model has better robustness (good fitting accuracy for both higher and lower order modes) and higher computational efficiency (1.133 s, the shortest time among all models). The proposed scheme mapping robot's joint angle and the natural frequency offers a valuable basis for further studying dynamic characteristics in industrial robotics.