【摘要】 为了控制焊接机器人焊缝成形的质量，提出了一种基于改进蚁群算法（ACO）优化BP神经网络的焊缝成形预测模型，实现对焊缝成形尺寸的控制。首先通过Otsu优化Canny算子的方法提取焊接过程中熔池图像的数据样本，然后用BP神经网络来进行训练预测。为了优化初始权阈值，引入ACO优化BP；针对蚁群陷入局部最优的情况，引入遗传算法（GA）中的交叉变异，利用适应度值来确定选择概率的特性，从而加快迭代速度，避开蚁群初期的收敛慢问题，提升预测模型的性能。最后通过与传统BP、GA-BP和ACO-BP的预测实验对比，发现改进后的预测模型准确度高、稳定性好。更多还原

【Abstract】 In order to control the weld forming quality of welding robot, a prediction model of weld forming was proposed based on improved ant colony algorithm(ACO) optimizing BP neural network to control the weld forming size. Firstly, the data sample of molten pool image in welding process was extracted by the method of Otsu optimizing Canny operator. Then BP neural network was used to train and predict. In order to optimize the initial weight threshold, ACO was introduced to optimize BP.In the case of ant colony falling into local optimum, cross-variation in genetic algorithm(GA) was introduced, and fitness value was used to determine the characteristics of selection probability, so as to speed up the iteration speed, avoid the problem of slow convergence in the initial stage of ant colony and improve the performance of prediction model. Finally, compared with the traditional BP, GA-BP and ACO-BP prediction experiments, it is found that the improved prediction model has high accuracy and good stability.更多还原