Industrial robotic assembly process modeling using support vector regression

The process parameters of high precision robotic assembly have to be tuned in order to deal with part variations and system uncertainties. Some methods such as design-of-experiment, artificial neural network and genetic algorithms have been proposed to optimize these parameters offline. However, these parameters have to be retuned for different batches due to part variations, which increases the production cost and lowers the manufacturing efficiency. Therefore new methods have to be developed to solve the problem. Because of the complexity of high precision assembly process, it is challenging to build a physical model to establish the relationship between an assembly process and its process parameters. Therefore we propose an assembly process modeling method based on support vector regression that constructs a model by observing the relationship between the assembly parameters and assembly output. The effectiveness and accuracy of the support vector regression based algorithm are further demonstrated by experiments using a robotic valve body assembly process in automotive manufacturing. The results show that the proposed method is capable of modeling complex assembly processes.