Modeling and Simulation of Three Dimensional Weld Pool Reconstruction by Stereo Vision

Online monitoring of weld pool and weld bead geometry has been a daunting challenge because of the harsh environment, slow processing speeds and difficulty of acquiring quality images during welding. The dimensions of the weld bead are the most fundamental requirement of a qualified weld, and monitoring the weld pool and weld bead geometry can be instrumental in ensuring the weld meets these requirements. Additionally, such a sensor system can one day lead to controlling the process parameters to provide much needed closed loop control of the welding process. Recent advantages in imaging technology, stereo vision algorithms and computing power have made stereo vision more competitive for providing such an online three dimensional reconstruction of welding scenes and this research has been exploring this possibility. Ideally, real world data would be used but there is currently no three dimensional measurement system that can provide a detailed description of the whole three dimensional surface of the weld pool and relationship to the workpiece. Therefore, this research has generated synthetic images from known models. By generating a simple model of the welding process and synthesizing images complete with noise and lighting characteristics, a better understanding of the entire process is developed and comparisons are made among the various algorithms. Finally a scaled reconstruction of the synthetic weld pool is made and compared to the original model for error analysis.