introduction
Trajectory tracking of welding robots is a major challenge in welding robot control. The robot is a typical nonlinear dynamic system with large inertia and delay. Currently, robot control mainly employs traditional PID control. Due to the high system complexity, designers make various assumptions and simplifications to build the system model, which significantly affects the control accuracy of the mathematical model. This paper proposes an iterative learning control strategy based on neural network identification, combining neural network identification with iterative control. This strategy can improve system control accuracy and enable the system to achieve its control functions and desired performance over a wider range of operating conditions, thus improving system control performance.
1 Neural Network Model Identification
The generalization and rapid learning capabilities of neural networks provide an effective approach for identifying nonlinear systems. In the research of robot systems, neural networks fully utilize their mapping capabilities to solve nonlinear problems such as dead zones and friction in robot control systems. The application of neural network model identification in robot control is shown in Figure 1.
Figure 1. Structure diagram of the rigid manipulator based on neural network control.
The key characteristic of neural network system identification is that it does not require building a system identification model. The process of system identification is essentially a process of learning the system's input and output data. The goal of learning is to minimize the error function value, reflecting the mapping relationship between the input and output data. If the output of the neural network can approximate the output of the system under the same input signal, then the neural network can be considered to have achieved identification of the original system, and the output of the neural network can be used as an estimate of the output of the actual system.
2 Design of Iterative Learning Controller
2.1 Iterative Learning Control
Iterative learning control is suitable for controlled objects with repetitive motion properties. It does not require identification of system parameters. It improves the system control target by iterative correction. The iterative control method does not rely on the precise mathematical model of the system. It can solve complex control problems with extremely simple algorithms within a given time interval [5].
For an nth-order linear time-varying discrete system, the iterative control model can be described as:
Figure 2 Iterative Learning Controller
2.2 Design of Robot Iterative Learning Controller
The system control block diagram is shown in Figure 3.
Figure 3 System control block diagram
3. Simulation of Welding Robot Control
A robotic arm with flexible connections was used as the simulation object, as shown in Figure 4.
Figure 4 Schematic diagram of the robotic arm
The following dynamic equations can be derived using the Lagrange method:
Figure 5. Neural network inverse identification structure diagram
Given a sinusoidal excitation signal, the control system is simulated using Matlab code. The iterative learning control curve and the response curve of the traditional PID control are shown in Figure 6.
Figure 6. Response curves of iterative learning control based on neural network model recognition and conventional PID control.
4. Conclusion
Simulation results show that the proposed iterative control scheme based on a neural network identification model significantly outperforms traditional PID control. This control strategy exhibits high tracking accuracy and low overshoot, effectively improving the system's control precision. Furthermore, optimizing this control strategy allows it to be applied to other robot tracking control systems, enabling the system to be deployed in a wider range of robot control applications.
