Composite control of the SiC arc welding power source based on the expert system and neuron PID

: A third generation wide-band-gap SiC semiconductor device is used in the SiC arc welding power source, which has a higher inverter frequency and greatly improves the dynamic characteristics of the arc welding power source, providing opportunities for control algorithm optimization. A composite control method of the arc welding power source combining the expert system and single neuron Proportional-Integral-Derivative (PID) is proposed in this paper, aimed at the fact that the proportional cofficient of neuron PID can hardly be adapted to rapid welding current changes. The SiC arc welding power source is taken as the plant of study in this paper. A mathematical model of the arc welding power source-arc system is established, and the controller of the arc welding power source based on the neuron PID and corresponding expert rules are defined to adjust the proportional coefficient of neuron PID Finally, the neuron PID controller (SNC) and the composite controller based on the expert system and neuron PID (ESNC) are simulated and verified. The results show that compared with the neuron PID algorithm, this method can adjust the proportional neuron PID coefficient in real time according to the welding current and has a better adaptive ability and superior tracking performance for arc welding power source control.


Introduction
Theband gap of a SiC power device is much wider than that of a Si-based power device, and the switching speed is higher, which provides better high-frequency characteristics for the arc welding power source. Based on analysis of the switching waveform and loss data in double-pulse tests by Hazra, it can be concluded that SiC power devices have more advantages than Si-based power devices in both the switching speed and loss [1]. The differences between Si-based and SiC power devices under the same conditions in high-frequency switching applications are compared in reference [2], and at present, the switching loss of a SiC power device is smaller than that of the best Si-based power device. A new generation of wide-bandgap welding power sources based on SiC power devices have been developed, as reported in reference [3], whose inverter frequency can be as high as 200 kHz; a full-bridge resonant converter with a new SiC power device, whose main circuit adopts an LLC topology structure and whose resonant converter frequency ranges from 260-to 310 kHz, has been established in reference [4]. The wide-bandgap welding power source has greatly improved the dynamic characteristics of the hardware platform of high-power arc welding power sources [5,6]. To fully utilize the hardware platform performance of the new ultrahigh-frequency SiC arc welding power source, an advanced control algorithm is needed. Currently, the incremental PIDcontrol algorithm is widely used in welding power control because of its simple principle and convenient use [7]. However, once the conventional incremental PID parameters are adjusted, they will remain unchanged in the whole control process, which is not conducive to tracking the set value and adapting to changes in welding process characteristics.
Many scholars have studied a variety of improved PID algorithms. It is pointed out in reference [8] that when the working area is frequently disturbed, artificial intelligence technology such as neural networks, fuzzy control or expert controllers can be used to optimize the adaptive welding parameters in real time and significantly reduce welding defects. Reference [9] adjusts the proportional coefficient in the incremental PID algorithm to achieve a rapid response through fuzzy immune PID control based on the biological immune feedback adjustment strategy and fuzzy logic reasoning. Zhang Liling uses the welding expert system of the Newton interpolation algorithm to realize self-adjustment of the process parameters of the power supply for cold welding of aluminum magnesium alloy sheets [10]. The application of a neuron PID controller with a strong robustness and adaptability is beneficial to further improve the control effect on an arc welding power source [11]. The simulation results in reference [12] reveal that the pulse current of the welding power source operated by a single neuron adaptive PID controller has a low intensity when the peak and base values are superposed. However, there are still shortcomings in the neuron PID control approach, such as a long adjustment time and difficulty in optimizing the proportional coefficient of neurons. There are few neuron PID improvement algorithms that can be applied to ultrahigh inverter frequency arc welding power sources.
2 Problem description 2.1 Principle of the wide-bandgap SiC arc welding power source control system A block diagram of the wide-band-gap SiC arc welding power source control system is shown in Figure 1. The wide-band-gap SiC arc welding power source control system consists of a power module, welding voltage and current sampling and modulation circuits, human-computer interaction interface, ARM Cortex-M4 digital control processor, wide-bandgap power device driver module, wire feed control module, etc. The control principle is as follows: the ARM Cortex-M4 digital control processor samples and filters the electric signal detection circuit of the welding arc through the Analog-to-Digital (ADC) module, and the value is compared with the given value to generate a deviation; the controller uses the deviation to adjust the Pulse Width Modulation (PWM) duty cycle of the timer output, and the driving signal obtained through the wide-band-gap power device driving module isolation and amplification can control the conduction time of the power switch of the inverter power source and accurately adjust the output energy of the power source so that the arc current tends to remain stable near a given value. The arc welding power control system model consists of an arc welding power source-arc system, a digital PID controller and a feedback link. And arc welding power source-arc systems include inverter power sources and arc loads. The hardware platform of the arc welding power source adopts the wide-band-gap SiC arc welding inverter power source independently researched and developed in the laboratory, whose inverter frequency can reach 200 kHz; its dynamic response is rapid, and the energy adjustment period is shorter. The output ( ) u k ranges from 10 to 375, and is related to the digital PWM output of the control chip, which determines the PWM duty cycle, and the range of the voltage for arc welding power source is from 0 to 70 V. The inverter link can be regarded as a purely delayed amplification link, which is caused by the out-of-control time constant of the inverter link. The transfer function is thus approximated.
After Laplace transformation, the equation becomes: Therefore, the transfer function of the arc welding power source-arc system is: 3 Arc welding power controller based on the expert system and neuron PID Figure 2 shows the controller based on the expert system and neuron PID (ESNC), which further integrates the expert system based on the neuron PID control structure. According to the change in the feedback deviation, the proportional coefficient of the neuron PID is adjusted online, and the output is controlled, while the PWM duty ratio is adjusted to achieve fine control of the output current. At the same time, when the arc characteristics change, the proportional coefficient of the neuron PID can be adaptively adjusted so that the system can obtain a fast response with a lower overshoot..

Neuron PID control structure
Neural network intelligent control technology has been applied in high-power power systems [11]. Among the various applications, neuron PID control is a combination of conventional incremental PID control and the neuron structure, and its control equation is: In this paper, the supervised Hebb weight coefficient learning rule is used to update the weight coefficient to improve the robustness and performance of the controller and promote convergence of the learning algorithm. The equation is as follows: where , , and are the learning rates whose value range is [0,1]. The value of the proportional coefficient is closely related to the dynamic response and stability of the system: And the welding process has nonlinear and time-varying characteristics. The system needs to be adaptively adjusted according to the relative error of the welding process to achieve better stability and speed.

Adjustment of the expert rules
To improve the system response speed, the value of K is considered in this paper. According to the absolute value of the standard deviation of the closed-loop response, | ( )| er k , the response is divided into several intervals based on expert knowledge, and the value of K is different in each interval, which is modified according to the variation characteristic of the deviation within a certain range to reduce overshoot and adjustment time. This paper adopts production representation based on rules, which is adjusted by the IF-THEN structure. According to many experiments, the adjustment rules are as follows: As shown in Equation 7, the regulation strategy of K is as follows: at the initial stage of system response, a larger value of K is used to enable the system to quickly reach a given value. Once the proportional coefficient continues to be large, the rise time will decrease, but overshoot will increase. To restrain system overshoot, the value of K is reduced to ensure system stability and reduce overshoot and the adjustment time in the second stage. The third and fourth stages are to optimize the learning function of the neuron PID in the refined adjustment process that tends to be steady state.
The value of K is controlled by expert rules, considering both the fast and stable requirements of the system response. Among them, a is used for approximate tuning, and b is used for fine-tuning.

Simulation and analysis
The . The initial learning rates and weight coefficients are the same as above.
The system has a step response at the current setpoint of 300 A, and at , when wire sticking occurs during welding, the arc load changes, whose transfer function is updated to the following: The simulation curve is shown in Figure 3. Curve comparison shows that the overshoot of ESNC is <1.5%, and its performance is good. The ESNC system is more stable and faster than the SNC system, and the adjustment time of ESNC is 0.64 ms, which is 47% less than that of SNC. The current can rise when wire sticking occurs during welding, and the system oscillates. The output is closer to the original given value of 300 A, which is attained more smoothly and faster by ESNC compared with SNC, and the overshoot is reduced by 6%, and the adjustment time is decreased by 38% in ESNC. Therefore, the composite control of a wide-bandgap SiC arc welding power source based on the expert system and neuron PID is feasible. The robustness and adaptability of the control system and the tracking performance under the condition of changing arc characteristics are greatly improved.

Conclusion
1) At a given current value, the arc welding power source controller based on the expert system and neuron PID can adjust the weight coefficient by the supervised Hebb learning algorithm and can adjust the proportional coefficient in four intervals based on expert rules, which results in the rise time of the welding current and overshoot being reduced.
2) The composite arc welding power controller based on the expert system and neuron PID has the advantages of a strong robustness, good adaptability and easy setting. When wire sticking occurs during welding, the composite controller based on the expert system and neuron PID can modify the proportional coefficient of the neuron PID dynamically based on expert rules, thereby enabling the system to be easily adapted to arc load changes with a shorter adjustment time and less overshoot, which is beneficial to improving the stability and response speed of the system.