Research on Intelligent Evaluation Technology of Weld Quality based on Machine Vision and Deep Learning Neural Networks

WANG Ren; CUI Yunlong; HE Jianying; LI Gangqing; HAN Xiaohui

doi:10.7512/j.issn.1001-2303.2025.10.09

“Huaguang”Cup Excellent Thesis Recommendation | Views : 0 下载量: 875 CSCD: 0

PDF
Export
Share
Collection
Album

Research on Intelligent Evaluation Technology of Weld Quality based on Machine Vision and Deep Learning Neural Networks
Vol. 55, Issue 10, Pages: 77-85(2025)
Received：28 August 2025，

Published：20 October 2025
DOI： 10.7512/j.issn.1001-2303.2025.10.09
稿件说明：

移动端阅览

汪认，崔云龙，何建英，等.基于机器视觉和深度学习神经网络的焊缝质量智能评价技术研究［J］.电焊机，2025，55（10）：77-85. DOI： 10.7512/j.issn.1001-2303.2025.10.09.

WANG Ren, CUI Yunlong, HE Jianying, et al.Research on Intelligent Evaluation Technology of Weld Quality based on Machine Vision and Deep Learning Neural Networks[J].Electric Welding Machine, 2025, 55(10): 77-85. DOI： 10.7512/j.issn.1001-2303.2025.10.09.

摘要

针对传统宏观金相分析依赖于人工观察和经验判断，存在主观性强、效率低等问题，提出了一种融合机器视觉和深度学习神经网络的焊缝质量智能评价技术，通过构建深度学习目标检测模型和深度学习语义分割模型实现了对金相试样的自动检测和金相识别。实验结果表明，基于Faster RCNN网络的目标检测模型识别精度最高可达98.53%，基于HRNetV2网络的语义分割模型的平均交并比最高可达96.72%。深度学习模型能够从大量数据中学习复杂的特征，实现对焊缝宏观形貌和缺陷的智能评价。相比人工分析，该技术单图分析时间从30 s缩至1 s，检测效率提升5倍，检测成本降低60%，可满足批量检测需求，为宏观金相技术的自动化发展提供了新的思路。

Abstract

Addressing the issues of strong subjectivity and low efficiency in traditional macro-level metallographic analysis

which relies on manual observation and experience-based judgment

a technology for intelligent evaluation of weld quality combining machine vision and deep learning neural networks is proposed. Through constructing a deep learning object detection model and a deep learning semantic segmentation model

automatic detection and identification of metallographic samples are realized. Experimental results show that the object detection model based on the Faster RCNN network has a recognition accuracy of up to 98.53%

and the semantic segmentation model based on the HRNetV2 network has an average intersection over union (IoU) of up to 96.72%. Deep learning models can learn complex features from a large amount of data

achieving intelligent evaluation of the macroscopic appearance and defects of welds. Compared with manual analysis

this technology reduces the analysis time per image from 30 seconds to 1 second

increasing detection efficiency by 5 times and reducing costs by 60%

meeting the requirements for batch detection

providing a new direction for the automated development of macro-level metallographic technology.

关键词

Keywords

references

ZHU B ， CHEN Z ， HU F . et al . Feature Extraction and Microstructural Classification of Hot Stamping Ultra-High Strength Steel by Machine Learning ［J］. JOM ， 2022 ， 74 ： 3466 - 3477 .

LIU Y ， ZHANG J ， WANG L ， et al . High-performance ceramic coatings for aerospace applications：A review ［J］. Measurement ， 2023 ， 206 ： 112345 .

ZHANG P ， YIN Z Y . A novel deep learning-based modelling strategy from image of particles to mechanical properties for granular materials with CNN and BiLSTM ［J］. Computer Methods in Applied Mechanics and Engineering ， 2021 ， 382 ： 113858 .

陈嘉林，张玉琪，徐伟 . 机器学习在金属微观组织图像分割中的应用［J］. 鞍钢技术， 2024 （ 06 ）： 61 - 69 .

CHEN J L ， ZHANG Y Q ， XU W . Application of Machine Learning in Image Segmentation of Metal Microstructures ［J］. Angang Technology ， 2024 （ 06 ）： 61 - 69 .

王炫东 . 基于卷积神经网络与先验知识的钢铁材料金相图谱智能识别［D］. 北京：北京科技大学， 2024 .

WANG X D . Intelligent Recognition of Steel MaterialMetallography Based on Convolutional NeuralNetworks and Prior Knowledge ［D］. Beijing ： University of Science and Technology Beijing ， 2024 .

陈哲兴 . 基于机器学习的高强钢显微组织图像分割方法研究［D］. 湖北：华中科技大学， 2022 .

CHEN Z X . Research on lmage Segmentation Method ofMicrostructure of High Strength Steel Based on Machine Learning ［D］. Hubei ： Huazhong University of Science and Technology ， 2022 .

赵隆，韩小稚，王景人，等 . 基于卷积神经网络的12Cr1MoV钢金相组织球化级别智能化分析［J］. 中国特种设备安全， 2024 ， 40 （ 3 ）： 80 - 83 .

ZHAO L ， HAN X Z ， WANG J R ， et al . Intelligent Analysis of 12Cr1MoV Steel Metallographic Structure Sphericity Level Based on Convolutional Neural Network ［J］. China Special Equipment Safety ， 2024 ， 40 （ 3 ）： 80 - 83 .

王森 . 基于计算机视觉的钢材金相图像晶粒度评价方法研究［D］. 陕西：西安工业大学， 2023 .

WANG S . Research on the Evaluation Method of GrainSize of Steel Metallographic lmages based onComputer Vision ［D］. Shanxi ： Xi'an Technological University ， 2023 .

朱家乐 . 基于卷积神经网络的20钢劣化分级与金相组织检测方法研究［D］. 辽宁：大连理工大学， 2023 .

ZHU J L . Research on 20 Steel Deterioration Grading and Metallographic DetectionMethod Based on Convolutional Neural Network ［D］. Liaoning ： Dalian University of Technology

王佳锐，刘能锋，曲鹏 . 卷积神经网络金相组织自动识别［J］. 智能系统学报， 2022 ， 17 （ 4 ）： 698 - 706 .

WANG J R ， LIU N F ， QU P . Automatic identification of metallographic structure based on convolutional neural network ［J］. CAAI Transactions on Intelligent Systems ， 2022 ， 17 （ 4 ）： 698 - 706 .

杨俊鸿 . 基于深度学习的金相组织智能评级方法研究［D］. 江苏：江苏大学， 2018 .

YANG J H . Research on Intelligent Rating Method of Metallurgical Organization Based on Deep Learning ［D］. Jiangsu ： Jiangsu University ， 2018 .

何惠珍 . 基于深度学习的钢材金相组织自动化识别［D］. 湖北：武汉工程大学， 2022 .

HE H Z . Automatic ldentification of Steel Metallographic Structure Based on Deep Learning ［D］. Hubei ： Wuhan Institute of Technology ， 2022 .

田中华 . 面向铝合金金相图像的晶界检测方法研究［D］. 辽宁：东北大学， 2021 .

TIAN Z H . Research on Grain Boundary Detection Methodfor Metallographic Image of Aluminum Alloy ［D］. Liaoning ： Northeastern University ， 2021 .

Alert me when the article has been cited

提交

Research Progress on the Application of Machine Vision in the Whole Process of Robot Welding

Research on Quality Control Strategy of Automatic Assembly Welding of Guard Board

Typical applications of Artificial Intelligence Surrogate Models in Laser Metal Additive Manufacturing

Gas Shielded Metal Arc Welding with Misalignment-Gap Mode of Penetration Prediction and Feature Analysis

Pipeline Girth Weld Defect Intelligent Identification of Digital Radiograph based on an Improved YOLOv5 Network