Research of visual image gray moment applying for weld robot trajectory control

Abstract

The new strategy for weld robot trajectory control has been research in the paper. The control strategy is based on gray moment of the weld visual image. Firstly, the 6-DOF weld robot experimental system has been setup. Some weld experiments have been caught out. Visual camera snaps 500 welding images in every welding experiment. The procession region of weld images has been determined. Median filtering and gray enhancement operations have been performed to the procession region. Then the values of first-order gray moment after normalization processing and its seam tracking errors have be obtained, which can be used to be the training and testing data. On this basis, a radial basis function network composing of three layers has been designed. The values of weld visual image first-order gray moment k j have used to be input parameter, and the errors of weld robot trajectory control y p j to be output parameter. Then the setup radial basis function network can be trained by the captured first 400 groups of visual image data, and the radial basis function network model for weld robot trajectory error prediction can be setup. Finally, the other 100 groups of data have been used to be test the setup radial basis function neural network prediction model. The verification experiment consequence showed that the trajectory prediction errors y p j by the radial basis function network model is basically coincident with the tracking measured errors y j . The error of the setup model is ±0.04 mm, which may be applied for weld robot trajectory control in real time.