东北大学学报(自然科学版) ›› 2022, Vol. 43 ›› Issue (12): 1688-1693.DOI: 10.12068/j.issn.1005-3026.2022.12.003

• 信息与控制 • 上一篇    下一篇

基于深度学习图像分析的晶体特性反馈控制

王良勇, 朱耀龙, 淦晨阳   

  1. (东北大学 流程工业综合自动化国家重点实验室, 辽宁 沈阳110819)
  • 发布日期:2022-12-26
  • 通讯作者: 王良勇
  • 作者简介:王良勇(1980-),男,山东聊城人,东北大学教授.
  • 基金资助:
    沈阳市中青年科技创新人才项目(RC200519).

Feedback Control of Crystal Characteristics Based on Deep Learning Image Analysis

WANG Liang-yong, ZHU Yao-long, GAN Chen-yang   

  1. State Key Laboratory of Integrated Automation for Process Industry, Northeastern University, Shenyang 110819, China.
  • Published:2022-12-26
  • Contact: WANG Liang-yong
  • About author:-
  • Supported by:
    -

摘要: 针对晶体尺寸期望以及标准偏差特性的在线反馈控制问题,提出一种基于深度学习图像分析的在线反馈控制方法.首先,通过基于深度学习神经网络的晶体图像分析方法在线分析晶体的形状与尺寸;然后,对图像分析的结果进行数学统计分析,得到某一批次晶体的尺寸期望与标准偏差;最后,针对晶体尺寸期望和标准偏差控制的欠输入特性设计了一种路径跟踪算法与PID相结合的反馈控制器,以获得具有目标尺寸期望与标准偏差的晶体.通过明矾冷却结晶实验验证了所提方法的有效性和可行性.

关键词: 冷却结晶;晶体特性;图像分析;路径跟踪;在线反馈控制

Abstract: Aiming at the online control of crystal size expectation and standard deviation characteristics, an online feedback control method based on deep learning image analysis is proposed. Firstly, the crystal image analysis method using deep learning neural network is introduced to analyze the shape and size of crystals online. Then, mathematical statistical analysis is performed to obtain the size expectation and standard deviation of a certain batch of crystals. Finally, a feedback controller combining path tracking algorithm and PID algorithm is designed to deal with under-input characteristics, so that the target size expectation and standard deviation is obtained. The effectiveness and feasibility of the proposed method is verified by the alum cooling crystallization experiment.

Key words: cooling crystallization; crystal characteristics; image analysis; path tracking; online feedback control

中图分类号: