Seam Tracking and Detection of Groove by using Neural Network in Robotic Welding

在这项研究中，在不使用垫板或衬垫材料，单面背面堆焊系统提出了获得良好的焊接效果，没有变化的根部间隙。在该系统中，焊枪不仅可以在槽内摆动，而且可以像开关一样前后移动。在前进过程中，热量被给予并熔化基材及其根部边缘。在反向过程中，液滴沉积在根部边缘，并形成桥。之后，在产生烧穿之前，使焊炬高速向前移动。重复该焊炬运动。为此，1）系统被构造为对焊炬运动（摆动宽度、焊接速度）、送丝速率和电源特性进行协调控制，即，保持稳定的熔池熔深并进行焊缝跟踪。脉冲电流变为峰值， 关于我们 Orch接近根部边缘。液滴沉积在根部边缘上。制作了计算机与焊接机器人、焊接电源和送丝机的接口电路，根据焊枪位置对送丝速度和电流波形进行协调控制。2）提出了一种在瞬态响应中实时检测电弧长度和电极丝伸出长度的技术。利用焊接开始时的延伸长度、每次采样时的焊接电流和送丝速度建立了神经网络模型。其模型输出为电弧长度和延伸长度，并通过基础焊接实验获得了瞬态响应。通过模型输出与实验结果的比较，验证了该神经网络的有效性。结果表明，该神经网络模型可以由电压和电流估计出电弧的延伸长度和弧长，可以作为电弧长度的传感器。通过使用其模型，可以在切换回运动的前进过程期间检测槽形状。少

In this study, where backing plate or backing material are not used, one side back bead welding system is proposed to obtain a good welding result without the variation of the root gap. In the system, the torch is not only oscillated in the groove, but also moved before and behind like the switch back. During the forward process, a heat is given and is melted the base material and its root edge. During backward process, the droplet is deposited at the root edge and the bridge is formed. After that, the torch is moved forward by high-speed before burn through is generated. This torch motion is repeated. Therefore, there is no incomplete fusion and the stable back bead is obtained.For this purpose,1) The system is constructed to do the cooperative control of the torch motion (weaving width, welding speed), the wire feed rate and the power source characteristic, i.e., the stable weld pool penetration is kept and the seam tracking is carried out. The pulsed current becomes peak where the t … More orch approaches the root edge. The droplet is deposited on the root edges. The interface circuit between computer and welding robot, the welding power source and the wire feeder was made, and the cooperative control of the wire feed rate and current waveform was carried out according to the torch position.2) The technique is proposed for enabling the real-time sensing of the arc length and the extension length of the electrode wire in transient response.3) The neural network models are constructed by using the extension length in the beginning of the welding, and the welding current at each sampling, the wire feed rate. Its model output is the arc length and the extension length.The transient response is obtained by carrying out the fundamental welding experiment. The validity of this neural network is confirmed by comparing the output of the model with the experimental result. It is shown that the extension length and the arc length from voltage and current can be estimated and the neural network model is applicable as a sensor of the arc length. By using its model, it is possible to detect the groove shape during the forward process in the switch back motion. Less