摘要
现行的焊接工艺预测算法难以满足算法设计的适用性、可靠性以及高效性的原则。根据工程需要,建立了以电弧长度、焊接电流、焊接速度、送丝速度和保护气流量为输入,以堆焊后的熔宽、熔深和稀释率为输出的5-8-3结构的误差反向传播(BP)网络模型,利用粒子群算法(PSO)优化BP网络得到最优权值和阈值来预测X65钢板堆焊Inconel625镍基合金的焊后质量。结果表明,PSO+BP算法相比单一BP算法具有较高的准确性,比遗传优化BP网络(GA+BP)算法高效。与GA+BP算法相比,稀释率的平均误差分别为0.30和1.05,计算时间分别6 726 s和11 034 s,将PSO优化后的最优权值与Chebyshev直接法确定的权值对比,得出两个模型的权值基本吻合,说明PSO+BP算法预测堆焊质量过程中没有陷入局部最优解,具有准确、高效和可靠的优点,适用于堆焊质量的预测。
The current welding process prediction algorithm cannot satisfy the applicability, reliability and efficiency of algorithm design principles. According to the requirements of the project, set up the error Back Propagation（BP） network model of the 5-8-3 structure with arc length, welding current, welding speed and wire feed speed and protection gas flow rate as input, weld width and weld height and the dilution rate after surfacing as output. Particle Swarm Optimization（PSO） was used to optimize the BP network to get the optimal weights and threshold to predict the quality of Inconel625 nickel base alloy surfacing X65 steel after welding. The results showed that the PSO＋BP algorithm has higher accuracy compared with the single BP algorithm, and is more efficient than Genetic optimizing BP network（GA ＋BP） algorithm. Compared with the GA ＋BP algorithm, the dilution rate of average error and computing time were 0.30, 1.05 and the 6 726 s and 11 034 s. To contrast the optimal weights are optimized by PSO and Chebyshev direct method, it is concluded that the PSO＋BP algorithm has not trapped in local optimal solution, and has the advantages of accurate, efficient and reliable, is suitable for welding quality prediction.
出处
《焊管》
2015年第2期5-10,15,共7页
Welded Pipe and Tube
基金
天津市科技支撑计划重点项目(11ZCGYSF00100)
关键词
焊接
堆焊
质量预测
粒子群算法
神经网络
welding
surfacing
quality prediction
particle swarm optimization（PSO）
neural network
