An Application of Neural-Network-Based Visual Servoing System to Image Path Marking and Precision Seam Welding with Wide Strip Image

• Main Authors: Chen,Sheng-Sing, 陳勝興
• Other Authors: Peng,Shou-Tao
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/c68f66

碩士 === 南臺科技大學 === 機械工程系 === 106 === In this study, we aim at developing a Neural-Network-Based Visual Servoing system to carry out Laser welding for irregular weld joint profiles. We propose an image processing scheme capable of visually tracking the weld seam and providing a good estimate on the weld gaps. The resulting image information is then displayed on the monitor. The estimate is also converted into CAM command through the Neural-Network Visual Servoing system to perform Laser welding and make a dynamical adjustment for the Laser spot size. A X-Y-Zoom servo table equipped with machine vision is used to verify the proposed scheme. Our study adopts the image processing methods of ‘wide line image detection’ and ‘line segment extraction’ to perform a path estimate with less deviation from its centerline. The concept of ‘prediction-measurement-correction’ in Kalman filter is employed as the strategy for the visual path tracking and marking. The mapping between the image information and the CAM coordinate of the servo table is achieved by training Self-Organizing Maps (SOM) and Radial Basis Function (RBF) Neural Network. Among them, the SOM network stores the Visual Servoing mapping between the target point image and the gesture position of the servo table. The SOM network also implements the identification of the Image Jacobian matrix. The RBF neural network is used as a Zoom learning framework to automatically adjust the laser spot size. This study features the application of a class of "wideband" image processing technology and neural network-like learning framework to accomplish the task of Laser welding tracking and spot resizing required in the precision Laser welding process.