Advanced search
    Zhu Li, Ma Jingyao, Meng Zhaoxin, Shi Jinsong, Xing Xin, Jiang Zhongjin. Compensation control of woodworking feeding platform based on self-adaptive genetic optimization recurrent neural network[J]. Journal of Beijing Forestry University, 2020, 42(12): 125-134. DOI: 10.12171/j.1000-1522.20200248
    Citation: Zhu Li, Ma Jingyao, Meng Zhaoxin, Shi Jinsong, Xing Xin, Jiang Zhongjin. Compensation control of woodworking feeding platform based on self-adaptive genetic optimization recurrent neural network[J]. Journal of Beijing Forestry University, 2020, 42(12): 125-134. DOI: 10.12171/j.1000-1522.20200248

    Compensation control of woodworking feeding platform based on self-adaptive genetic optimization recurrent neural network

    •   Objective  In view of the complex structure of new woodworking band saw feeding platform with the parallel multi axis linkage, the processing accuracy is low, the control parameters can not be optimized, and there are many uncertain factors affecting the accuracy, et al. Combined with the advantages of self-adaptive genetic algorithm (GA) and recurrent neural network (RNN), a global optimization control strategy was designed, which combined recurrent neural network and self-adaptive genetic algorithm.
        Method  The corresponding error source model was established by analyzing the feeding platform structure and error source. The RNN parameters were optimized in combination with self-adaptive genetic algorithms, and then the PID parameters were optimized. The compensation control strategy was verified by Matlab and Adams joint simulation method, and compared with the traditional PID, genetic algorithm optimization PID parameters and RNN optimization PID parameters. The control parameters, feeding platform displacement and angle change curve were analyzed under different algorithms. Then we set up the actual circuit and controller for experiments.
        Result  According to the simulation results, compared with the other three control strategies, the control strategy had the smallest overshoot, the fastest response, reaching stable at about 0.6 second. And it was quicker to reach stable under external interference at about 0.3 second. After the control strategy compensating, the offset error in the Y direction was reduced from 6 to less than 3 mm, the offset error in the X direction was reduced from 6 to 2 mm, the tilt angle error was reduced from 5.5° to 3°, and most of the platform trajectory curve overlapped the target curve fully. When controlled by traditional PID, the offset error of Y direction was 6 mm, the offset error of X direction was 6 mm, the tilt angle error was 5.5°, and the platform trajectory curve deviated greatly from the target curve. When the genetic algorithm optimized the control of PID parameters, the offset error in the Y direction was reduced from 6 to less than 4.8 mm, the offset error in the X direction was reduced from 6 to 5 mm, the tilt angle error was reduced from 5.5° to 4.5°, and part of the platform trajectory curve overlapped the target curve. When the RNN neural network optimized PID parameter control, the offset error in the Y direction was reduced from 6 to less than 4.5 mm, the offset error in the X direction was reduced from 6 to 4.8 mm, and the tilt angle error was reduced from 5.5° to 4° . Part of the platform trajectory curve overlapped the target curve.
        Conclusion  Compared with the other three methods, this method has fast response speed, small overshoot, good anti-interference performance and strong robustness. Moreover, it can effectively compensate the error, improve its motion accuracy and meet the driving requirements.
    • loading

    Catalog

      Turn off MathJax
      Article Contents

      /

      DownLoad:  Full-Size Img  PowerPoint
      Return
      Return