大功率激光焊背面焊缝宽度神经网络预测

Weldment back of weld width prediction based on neural network during high-power laser welding

针对焊接过程中熔透及焊缝背面成形难以直接检测的问题,通过焊件正面和侧面的传感特征信息,对焊件背面的焊缝宽度进行预测.用视觉传感器获取激光焊接过程中包含焊接特征信息的图像,对图像进行分割分层、模式识别和空域图像处理,准确提取焊接特征信息,发现焊接特征信息随着焊接路径的变化有着相应的变化趋势.建立包含两个隐含层的贝叶斯神经网络,用提取到的9组特征信息作为输入,对焊件背面焊缝宽度进行预测.通过10组焊件背面焊缝宽度的预测值与实际值的比较,验证了贝叶斯神经网络具有良好的预测能力,在焊缝不理想的状态下,也具有较好的预测能力.

In high-power laser welding process, it is hard to detect weld penetration conditions and back of weld shape directly. The width of back of weld was predicted by the sensing characteristics information of weld face and side surface. Visual sensors are used to capture images which contain weld characteristics information in laser welding process. Weld characteristics are extracted accurately through image segmentation, image hierarchical, pattern recognition and space image process. The extracted characteristics variation trends are corresponding to weld route change obviously. Bayes neural network that contains two hidden layers is established for back of weld width prediction of weldment, and the characteristics extracted from images are used as inputs. The compare results between prediction value and real value verified that the established Bayes neural network has good predictive ability, and better predictive stability even the weld is not ideal.