Path Planning For Welding Robot Based On Deep Learning Method

Multi-joint tandem industrial robots are widely used in industrial production due to their advantages of easy operation,accurate positioning and flexible execution.The working principle of industrial robot is mainly to remember the running track and use the control through manual instruction to make it reach the specified posture.Generally,industrial robots only follow one or more fixed routes in the actual pipeline work.In the 1990s,the term neural network came into people's vision,and people began to explore the use of neural network to complete control and feedback in the operation of industrial robots.In this paper,based on the BP neural network principle,ABB's IRB6700 6R robot is adopted as the data acquisition source,the robot's transformation matrix and spatial position are used as the input and the joint Angle is used as the output,to complete the mapping relationship between the robot's pose and joint,and replace the complex formula calculation of kinematics equation.Firstly,in order to obtain network training data,it is necessary to realize the programmable robot.The change matrix and position of the research object are analyzed.The spatial coordinate system of IRB6700 six-joint robot is established,and the linkage parameter table of the robot is obtained.The spatial coordinate system follows the d-h rule.Through kinematics analysis,the forward and inverse kinematics equations are written,and the forward and inverse solutions are derived by combining the linkage parameters.Input linkage parameters into the Robotics Toolbox of MATLAB to establish a robot model and conduct simulation on the motion space,so as to verify the correctness of kinematics analysis and lay a theoretical foundation for the acquisition of data in network training.Secondly,the paper analyzes the neural network theory and the process of data transmission and error transmission of BP neural network,and gives the method of adjusting weights and thresholds.Then,through the kinematics analysis of the research object and the linkage parameter table analyzed in the second chapter,the pose and joint Angle data of the robot are obtained in MATLAB as the training data.Different Numbers of hidden layer neurons were set for the experiment,and the network parameters and the number of hidden layer neurons were obtained by comparing the results.Each group of data will stimulate the network and play a training role.Due to the large number of training data,the weights and thresholds of the network are constantly modified under the stimulation of multiple groups of data,and the neural network is adjusted so as to obtain the optimal neural network.In addition,the positions and joint angles of eight groups of IRB6700 robots in the motion space were obtained from ABB's RobotStudio simulation software to verify the network.Finally,the 3d simulation model of automobile side circumference was established,and the 3d model was imported into RobotStudio software to simulate the welding path of robot welding automobile side circumference,and ten sets of path points were obtained.The data of ten groups of path points are input into the trained neural network to obtain the output data.And input ten sets of output data into the RobotStudio software to run the planned path of the network.Ten sets of data at the network output of two groups of data,using five times polynomial interpolation in joint space path planning,six joint position,velocity and acceleration of variation over time and through the results to verify the neural network at the completion of the cartesian space to joint space path planning in the process of mapping the rationality and validity of the path planning.