大功率盘形激光焊焊缝背面宽度预测

发布时间：2018-04-02 03:03

本文选题：激光焊接　切入点：焊缝宽度预测　出处：《光学精密工程》2017年09期

【摘要】：提出了通过视觉传感获取焊接过程中的焊接特征信息并利用神经网络模型预测焊缝背面宽度的方法。利用大功率盘形激光器焊接了低碳钢SS400焊件,在焊接过程中改变焊接功率、焊接速度和焊接路径,并利用两台高速摄像机同步获取焊件正面和侧面出现的焊接特征信息。对获取的图像进行色彩空间转换、分层、滤波去噪和空域图像处理,提取飞溅、熔池和金属蒸气等焊接特征信息,观察焊接路径对各个特征的影响。最后,建立了一个三层的LMBP(LevenbergMarquardt Back Propagation)神经网络模型,将提取的特征信息作为输入量,预测焊缝的背面宽度。结果显示:当熔透不稳定或出现未熔透状态时,LMBP神经网络拟合度大于0.83,最大训练误差均值为0.002 8mm,最大实际误差均值为0.225 6mm。试验结果表明所建立的预测模型具有良好的准确性和稳定性。
[Abstract]:The method of obtaining welding characteristic information in welding process by visual sensor and predicting back width of weld by neural network model is proposed. Low-carbon steel SS400 welds are welded by high-power disk laser, and the welding power is changed during welding process. Welding speed and welding path, and using two high-speed cameras to simultaneously obtain welding features of the front and side of the welding parts. Color space conversion, stratification, filtering and denoising, spatial image processing, extraction of splash, Welding characteristic information such as molten pool and metal vapor are observed and the influence of welding path on each characteristic is observed. Finally, a three-layer LMBP(LevenbergMarquardt Back propagation neural network model is established, which takes the extracted feature information as input. The results show that the fitting degree of LMBP neural network is greater than 0.83, the mean of maximum training error is 0.002 mm, and the average of maximum actual error is 0.225 6 mm. The prediction model has good accuracy and stability.
【作者单位】：广东工业大学机电工程学院;
【基金】：国家自然科学基金资助项目(No.51675104) 广东省科技计划基金资助项目(No.2016A010102015) 广州市科技计划基金资助项目(No.201510010089)
【分类号】：TG456.7

【相似文献】