基于SAM预训练大模型智能化组合策略的燃料组件水下快速精确定位优化研究

陈相羽; 李豪; 王炳炎; 王坤; 万浩; 温忻

doi:10.13832/j.jnpe.2023.S2.0140

基于SAM预训练大模型智能化组合策略的燃料组件水下快速精确定位优化研究

doi: 10.13832/j.jnpe.2023.S2.0140

中国核动力研究设计院，成都，610213

详细信息

作者简介:
陈相羽（1992—），男，工程师，现主要从事反应堆智能化专用设备的研究，E-mail: xiangyu.c@qq.com

中图分类号: TL3
计量
- 文章访问数: 240
- HTML全文浏览量: 64
- PDF下载量: 34
- 被引次数: 0
出版历程
- 收稿日期: 2023-07-11
- 修回日期: 2023-09-20
- 刊出日期: 2023-12-30

Optimization Research on Rapid and Accurate Underwater Positioning of Fuel Assemblies Based on SAM Pre-Trained Large Model Intelligent Combination Strategy

Nuclear Power Institute of China, Chengdu, 610213, China

摘要

摘要: 反应堆换料过程中燃料组件的精确定位是成功抓取与无损提升的关键，由于对安全性与效率的高要求，面对堆内复杂水下环境，需要在保证高精度的同时实现在多变环境中的适应性与可靠性，从而实现全自动精确定位。本研究针对该情况，建立目标检测深度学习模型（YOLO），并利用基于自然语言处理的“分割一切（SAM）”图像分割预训练大模型进行进一步切分，最后通过先验结构实现目标图形学后处理优化，提升燃料组件的定位精度。该多阶段的组合策略模型实现了全自动智能化定位过程的算法解释性，满足了换料保障现场的调试要求，同时通过目标快速检测-目标分割-形态学后处理计算的优化方案解决水下不确定性环境中的不稳定问题，具有高范用性与鲁棒性。在模拟试验环境中，该模型快速可靠、调试性强，单组件定位时间不超过1 s，误差不超过0.5 mm，换料定位效率提升超过90%，大幅度降低操作人员受辐照剂量。
- 反应堆燃料组件 /
- 机器视觉 /
- 图像分割 /
- 目标检测 /
- 大模型 /
- 反应堆换料
Abstract: Accurate positioning of fuel assembly is the key to successful grasping and nondestructive lifting in the process of reactor refueling. Due to the high requirements for safety and efficiency, it is necessary to realize adaptability and reliability in a changeable environment while ensuring high accuracy in the face of complex underwater environment in the reactor, so as to realize full-automatic accurate positioning. In this study, the object rapid detection model (YOLO) is established, and the natural language processing-based SAM (Segment Anything Model) image segmentation pre-training large model is used for further segmentation. Finally, the post-processing optimization of target graphics is realized through a priori structure, and the positioning accuracy of fuel assembly is improved. This multi-stage combined strategy model achieves algorithmic interpretation of fully automated intelligent positioning process to meet the debugging requirements at the refueling site, while solving the instability problem in the underwater uncertain environment through fast detection-intelligent segmentation-post process to achieve high paradigm and robustness. In the simulated test environment, the model is fast, reliable and highly debuggable. The positioning time of a single assembly is less than 1s, the error is less than 0.5 mm, and the positioning efficiency of refueling is improved by more than 90%, allowing significant reduction of the irradiation dose of operator.
- Reactor fuel assembly /
- Computer vision /
- Image segmentation /
- Object detection /
- Foundation model /
- Reactor refueling process

HTML全文

图 1 GIoU模型计算情况

红框—预测区域；绿框—基准区域；蓝框—预测区域与基准区域的最小外接矩形区域

Figure 1. Different Situations of GIoU Precision Model

下载: 全尺寸图片幻灯片

图 2 目标分割示意图

Figure 2. Schematic Diagram of Object Segmentation

下载: 全尺寸图片幻灯片

图 3 模拟定位试验环境

Figure 3. Simulated Positioning Test Environment

下载: 全尺寸图片幻灯片

图 4 训练过程GIoU_loss变化曲线

Figure 4. Change Curve of GIoU_loss in Training Process

下载: 全尺寸图片幻灯片

图 5 燃料组件初步定位结果

Figure 5. Preliminary Positioning Result of Fuel Assembly

下载: 全尺寸图片幻灯片

图 6 燃料组件图像分割效果

Figure 6. Image Segmentation Effect of Fuel Assembly

下载: 全尺寸图片幻灯片

图 7 燃料组件精确定位结果

Figure 7. Accurate Positioning Result of Fuel Assembly

下载: 全尺寸图片幻灯片

参考文献(13)

[1]	程伟,王磊,胡建华,等. 基于机器视觉技术的乏燃料贮存格架自动定位试验方法研究与应用[J]. 核动力工程,2020, 41(6): 198-201.
[2]	高永明,李声,李丽丹. 燃料组件水下非接触变形测量方法[J]. 核动力工程,2010, 31(4): 87-90.
[3]	刘赫. 核燃料组件变形尺寸测量系统设计[D]. 成都: 四川大学,2005.
[4]	任亮,李国云,江林志,等. 压水堆燃料组件池边检查技术研究进展[J]. 科技导报,2015, 33(18): 91-95. doi: 10.3981/j.issn.1000-7857.2015.18.015
[5]	朱江,杜瑞,李建奇,等. 基于注意力机制的曲轴瓦盖上料机器人视觉定位和检测方法[J]. 仪器仪表学报,2021, 42(5): 140-150. doi: 10.19650/j.cnki.cjsi.J2006768
[6]	MAZUROWSKI M A, DONG H Y, GU H X, et al. Segment anything model for medical image analysis: an experimental study[J]. Medical Image Analysis, 2023, 89: 102918. doi: 10.1016/j.media.2023.102918
[7]	ZHU X K, LYU S, WANG X, et al. TPH-YOLOv5: improved YOLOv5 based on transformer prediction head for object detection on drone-captured scenarios[C]//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision Workshops. Montreal, BC, Canada: IEEE, 2021: 2778-2788.
[8]	ZHANG Y, SHEN Y L, ZHANG J. An improved tiny-yolov3 pedestrian detection algorithm[J]. Optik, 2019, 183: 17-23. doi: 10.1016/j.ijleo.2019.02.038
[9]	HU X L, LIU Y, ZHAO Z X, et al. Real-time detection of uneaten feed pellets in underwater images for aquaculture using an improved YOLO-V4 network[J]. Computers and Electronics in Agriculture, 2021, 185: 106135. doi: 10.1016/j.compag.2021.106135
[10]	WANG C Y, LIAO H Y M, WU Y H, et al. CSPNet: a new backbone that can enhance learning capability of CNN[C]//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Seattle, WA, USA: IEEE, 2020: 1571-1580.
[11]	GHIASI G, LIN T Y, LE Q V. DropBlock: a regularization method for convolutional networks[C]//Proceedings of the 32nd International Conference on Neural Information Processing Systems. Montréal: Curran Associates Inc, 2018: 10750-10760.
[12]	BOUGUEZZI S, BEN FREDJ H, FAIEDH H, et al. Improved architecture for traffic sign recognition using a self-regularized activation function: SigmaH[J]. The Visual Computer, 2022, 38(11): 3747-3764. doi: 10.1007/s00371-021-02211-5
[13]	ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: faster and better learning for bounding box regression[C]//Proceedings of the 34th AAAI Conference on Artificial Intelligence, The Thirty-Second Innovative Applications of Artificial Intelligence Conference, The Tenth AAAI Symposium on Educational Advances in Artificial Intelligence. New York, NY, USA: AAAI, 2020: 12993-13000.