Purpose

The purpose of this paper is to accurately obtain the deformation of a hybrid robot and rapidly enable real-time compensation in friction stir welding (FSW). In this paper, a prediction algorithm based on the back-propagation neural network (BPNN) optimized by the adaptive genetic algorithm (GA) is presented.

Design/methodology/approach

Via the algorithm, the deformations of a five-degree-of-freedom (5-DOF) hybrid robot TriMule800 at a limited number of positions are taken as the training set. The current position of the robot and the axial force it is subjected to are used as the input; the deformation of the robot is taken as the output to construct a BPNN; and an adaptive GA is adopted to optimize the weights and thresholds of the BPNN.

Findings

This algorithm can quickly predict the deformation of a robot at any point in the workspace. In this study, a force-deformation experiment bench is built, and the experiment proves that the correspondence between the simulated and actual deformations is as high as 98%; therefore, the simulation data can be used as the actual deformation. Finally, 40 sets of data are taken as examples for the prediction, the errors of predicted and simulated deformations are calculated and the accuracy of the prediction algorithm is verified.

Practical implications

The entire algorithm is verified by the laboratory-developed 5-DOF hybrid robot, and it can be applied to other hybrid robots as well.

Originality/value

Robots have been widely used in FSW. Traditional series robots cannot bear the large axial force during welding, and the deformation of the robot will affect the machining quality. In some research studies, hybrid robots have been used in FSW. However, the deformation of a hybrid robot in thick-plate welding applications cannot be ignored. Presently, there is no research on the deformation of hybrid robots in FSW, let alone the analysis and prediction of their deformation. This research provides a feasible methodology for analysing the deformation and compensation of hybrid robots in FSW. This makes it possible to calculate the deformation of the hybrid robot in FSW without external sensors.

中文翻译：

---

基于自适应GA-BPNN和5自由度混合机器人在线补偿的变形预测

目的

本文的目的是准确地获得混合机器人的变形并快速启用摩擦搅拌焊（FSW）的实时补偿。提出了一种基于遗传算法（GA）优化的BP神经网络的预测算法。

设计/方法/方法

通过该算法，将五自由度（5-DOF）混合机器人TriMule800在有限位置上的变形作为训练集。机器人的当前位置及其所承受的轴向力用作输入。以机器人的变形为输出，构建BPNN。采用自适应遗传算法优化BPNN的权重和阈值。

发现

该算法可以快速预测机器人在工作空间中任何位置的变形。本研究建立了一个力-变形实验台，实验证明模拟变形与实际变形的对应关系高达98％。因此，仿真数据可以用作实际变形。最后，以40组数据为例进行预测，计算出预测和模拟变形的误差，验证了预测算法的准确性。

实际影响

整个算法已通过实验室开发的5-DOF混合机器人进行了验证，并且也可以应用于其他混合机器人。

创意/价值

机器人已在FSW中广泛使用。传统系列机器人在焊接过程中无法承受较大的轴向力，并且机器人的变形会影响加工质量。在一些研究中，混合动力机器人已用于FSW。但是，在厚板焊接应用中混合机器人的变形不容忽视。目前，尚无关于FSW中混合机器人变形的研究，更不用说对其变形的分析和预测了。该研究为分析FSW中混合机器人的变形和补偿提供了一种可行的方法。这使得无需外部传感器即可在FSW中计算混合机器人的变形。