Top And Bottom Weldment Surface Synchronous Detection For Penetration Status And Weldment Bottom Surface Weld Forming Prediction Algorithm During High-power Disc Laser Welding

Laser welding technology has been rapidly developed and is widely used in various fields of industrial manufacturing in modern times since it has lots of advantages such as high speed, high power density, large aspect ratio, welding of refractory materials and welding of dissimilar materials, the implementation of non-contact long-distance welding,and easy to control, etc.,. As the laser welding has high power density and laser spot diameter is very small, precision requirements of the welding process equipment and the weldment fix are high, otherwise there would be lots of welding defects, and more seriously,it would cause scrap pieces. In order to get a better weld quality, it is very important to have an accurate judgement of weld condition during laser welding process. The weld joint forming quality affects weld quality directly, and it is very necessary to have a better control of weld penetration to achieve a good weld joint forming quality during laser welding.By means of summarizing and comparing the existing methods of penetration detection,the dissertation focused on the real-time penetration detection technology in welding process.As the penetration condition is hard to be observed directly, laser welding experimental platform was established and visual sensors were used to capture the weld characteristics information. By the used of the experimental platform, the keyhole penetration condition has been observed directly, and other weld characteristics have been captured simultaneously.Pattern recognition was established to estimate and predict the weld penetration condition.And the established pattern recognition model was verified to be effective by forming inspect of weld joint after welding.There were three kinds of experimental platform have been established to research the relationship between weld penetration and weld characteristics, and the weld penetration impacts on weld formation and welding quality were analyzed. Laser welding of low carbon steel, laser-arc hybrid welding of stainless steel, laser-arc hybrid welding of stainless steel and low carbon steel, and laser welding of aluminum alloy 5052 and aluminum alloy 6061 were performed. For the laser welding of low carbon steel, the characteristics of both top surface and bottom surface of weldment, and both top surface and flank surface of weldment during laser welding were captured through the designed different experimental platform.Penetration conditions were analyzed by observing the weld width under different weld conditions. For the laser-arc hybrid welding of stainless steel, stainless steel and low carbon steel, and aluminum alloy laser welding, the flank surface and bottom surface weld characteristics were captured under different weld conditions. Penetration conditions were analyzed by the observation of the keyhole penetration condition on bottom surface directly during different weld conditions.Two high speed cameras were used as visual inspection system during laser welding.Monochrome high speed camera with infrared-visible light filter was used to capture the top surface characteristics of weldment, and color high speed camera with visible light filter was used to capture the bottom surface characteristics that were reflected by a mirror under the weldment. K-means clustering algorithm was used to process the color images, and wavelet algorithm was used to process the grayscale images. As keyhole condition was extracted directly, it was obvious that the weld conditions affected the penetration a lot, and the characteristics had very close relationship with penetration condition.Weld characterisctis were extracted from color images by pattern recognition algorithm,and LMBP (Levenberg-marquardt Back-Propagation) neural network and Bayes neural network were established for weldment bottom surface weld width prediction. Six characteristics were extracted from color images, and three characteristics were extracted from grayscale images. There were nine characteristics in total were set as in puts of LMBP neural network and Bayes neural network, and the predicted value of bottom surface weld width was set as out put. The results showed that the LMBP neural network and Bayes neural network were effective and accurate.Weld joint forming inspect was performed after laser welding. By analyzing and comparing the macrograph of different weld conditions, it was found laser power, weld route and weld speed affected penetration condition most. Then the microstructure of weld joint was observed and the Vickers hardness measurement was performed on weld joint. The Vickers hardness condition was weld section > parent metal > heat effect zone. The analysis of weld joint forming inspect verified the penetration pattern recognition and weldment bottom surface weld width predictionary were effective.