Improving efficiency with orthogonal exploration for online robotic assembly parameter optimization

摘要

In this paper, an online robotic assembly parameter optimization method is developed, this method make industrial robots can assemble workpiece from unskilled to skilled just like human's learning behavior. With the development of the force sensor technology and force control technology, the industrial robots are used in high precision assembly tasks which are more complicated. In order to ensure the efficiency and success rate of assembly, it is necessary to select the appropriate assembly parameters, this problem is called robotic assembly parameters optimization. The traditional solutions are used by artificial methods, a lot of experiments are carried out to get the optimal parameters, which are very time-consuming and laborious. Especially when the production line changes, using the traditional solutions have to do heavy experiments again, it can't meet the requirements of today's flexible manufacturing requirements. This paper presents an online robotic assembly parameter optimization method, which is called Gaussian Process Regression surrogated Bayesian Optimization Algorithm based on the Orthogonal Exploration (OE-GPRBOA), this method can liberate the labor, does not require artificial participation. The algorithm can optimize the parameters autonomously, finally find the optimal parameters for robotic assembly. For GPR is suitable for processing high dimension, small size of sample and nonlinear complex regression problems, the proposed OE-GPRBOA method can be used for various assembly tasks. In this paper, peg-in-hole assembly experiments are performed. The proposed method also compared with design of experiments (DOE) method and GPRBOA method. Experimental results show that, the proposed OE-GPRBOA method has more efficiency to find the optimal assembly parameters, this method can generate big economic impact.