欢迎访问《人工晶体学报》官方网站,今天是 分享到:

人工晶体学报 ›› 2022, Vol. 51 ›› Issue (2): 229-241.

• 研究论文 • 上一篇    下一篇

基于贝叶斯参数优化的无模型自适应硅单晶直径控制

林光伟1, 王珊1, 张西亚1, 彭鑫1, 高俊伟2, 高德东1   

  1. 1.青海大学机械工程学院,西宁 810016;
    2.阳光能源(青海)有限公司,西宁 810000
  • 收稿日期:2021-09-26 出版日期:2022-02-15 发布日期:2022-03-14
  • 通讯作者: 高德东,博士,教授。E-mail:gaodd@zju.edu.cn
  • 作者简介:林光伟(1997—),男,四川省人,硕士研究生。E-mail:1026468163@qq.com
  • 基金资助:
    高海拔环境下光伏电站组件质量评估及衰减机制研究(2019-ZJ-7002);工业和信息化部2018年绿色制造系统集成项目(130)

Model-Free Adaptive Diameter Control of Monocrystalline Silicon Based on Bayesian Parameter Optimization

LIN Guangwei1, WANG Shan1, ZHANG Xiya1, PENG Xin1, GAO Junwei2, GAO Dedong1   

  1. 1. School of Mechanical Engineering, Qinghai University, Xining 810016, China;
    2. Solargiga Energy (Qinghai) Co., Ltd., Xining 810000, China
  • Received:2021-09-26 Online:2022-02-15 Published:2022-03-14

摘要: 直拉硅单晶的生长过程涉及多场多相耦合与复杂的物理化学变化,其中工艺参数的波动是导致晶体直径不均匀的重要原因,如何实现工艺参数的控制以获得理想的、均匀的晶体直径具有重要的研究意义。本文分析现有控制方法存在不稳定以及控制效果不佳的问题后,提出基于贝叶斯参数优化的无模型自适应控制模型来控制硅单晶生长过程中的晶体直径。首先以坩埚上升速度与加热器的功率作为控制输入参数,晶体直径作为输出,搭建无模型自适应控制模型,并分析算法的稳定性。其次将控制模型进行仿真实验,发现硅单晶直径控制模型中不同的超参数设定会影响控制过程的迭代次数以及控制效果。最后,利用贝叶斯优化超参数的取值范围,并进行最终的仿真实验,结果表明,经贝叶斯参数优化后的控制模型计算快、迭代次数少,输出的晶体直径稳定,同时将生长工艺参数控制在实际生产要求范围内。因此,基于贝叶斯参数优化的无模型自适应控制实现了硅单晶直径均匀稳定的有效控制,具有结合工程背景的实际应用前景。

关键词: 硅单晶, 直径控制, 无模型自适应控制, 超参数, 贝叶斯参数优化

Abstract: The growth process of Czochralski silicon is related to multi-field and multi-phase coupling and complex physical and chemical changes. The fluctuations in process parameters are an important cause of uneven crystal diameter. This is significant research on how to achieve the control of process parameters to obtain an ideal and uniform crystal diameter. In this paper, after analyzing the instability and control effect degradation of the existing control method, a model-free adaptive control model based on Bayesian parameter optimization is presented to control the crystal diameter of the monocrystalline silicon growth process. Firstly, a model-free adaptive control model was established with crucible rising speed and heater power as control input parameters and crystal diameter as output parameters, and the stability of the algorithm was analyzed. Secondly, the control model was simulated and it is found that different super-parameter settings in the Czochralski silicon diameter control model will affect the number of iterations and the control effect. Finally, the value range of the hyper-parameter was optimized by Bayesian and the final simulation experiment was carried out. The results show that the model-free adaptive control optimized by the hyper-parameters is faster in calculation, less in iteration times, and the growth process parameters are controlled within the actual production requirements. Therefore, the model-free adaptive control based on Bayesian parameter optimization can effectively control the diameter of monocrystalline silicon, which has a practical application prospect combined with engineering background.

Key words: monocrystalline silicon, diameter control, model-free adaptive control, hyper-parameter, Bayesian parameter optimization

中图分类号: