Automated Metallic Weld Defect Detection
Automated Metallic Weld Defect Detection
Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1806
Automated Metallic Weld Defect Detection
Sonal Chaudhari1; Siddhant Nitin Rege2; Sakshi Ganesh Manjrekar 3;
Aarti Aklu Gupta4; Sahil Pravin Satardekar 5
1
Professor
1,2,3,4,5
Department of Computer
Datta Meghe College of Engineering Airoli, lndia
Publication Date: 2025/04/11
Abstract: In several industries, such as manufacturing, construction, and the automotive sector, welding is an essential
procedure. Safety and structural integrity are directly impacted by weld quality. Conventional welding inspection techniques
result in inconsistencies and inefficiencies because they are labor-intensive, manual, and prone to human error. This study uses
convolutional neural networks (CNN) and machine learning (ML) to offer an automated welding fault diagnosis method. Before
assigning the weld to one of six categories—Good Weld, Burn Through, Contamination, Misalignment, Lack of Penetration,
and Lack of Fusion—the system first confirms whether welding has been done. The model achieves great accuracy in defect
identification after being trained on a variety of datasets. This method is appropriate for industrial applications since it increases
efficiency, decreases reliance on humans, and improves the accuracy of defect identification by utilizing deep learning
techniques.
Keywords: Machine Learning, CNN, Welding Fault Detection, Image Processing, Automated Inspection.
How to Cite: Sonal Chaudhari; Siddhant Nitin Rege; Sakshi Ganesh Manjrekar; Aarti Aklu Gupta; Sahil Pravin Satardekar. (2025).
Automated Metallic Weld Defect Detection. International Journal of Innovative Science and Research Technology,
10(3), 2596-2599. https://doi.org/10.38124/ijisrt/25mar1806.
I. INTRODUCTION Decentralized Automated Inspection Systems
In industrial applications, detecting welding faults is Automated weld inspection with CNNs overcomes these
crucial to preserving quality and safety. The subjective, limitations by: Enabling real-time defect detection: Deep
unpredictable, and time-consuming nature of manual inspection learning models classify weld quality in real time, enabling
might result in possible flaws in welded structures. Deep quicker quality control.
learning-based automated inspection systems present a viable Removing human bias: Machine learning offers consistent,
remedy by increasing precision and effectiveness. This study unbiased defect detection.
introduces a CNN-based model that detects the presence of a
weld and divides it into six groups. II. LITERATURE SURVEY
Centralized Welding Inspection Systems Deep learning application for Metallic weld defect
detection through automation has been the focus of great
Conventional inspection techniques are subject to interest in the past few years owing to the necessity of reliable
discrepancies since they depend on human knowledge. and efficient non-destructive testing (NDT) techniques. The
These systems are: Labor-intensive and slow: manual literature review given below summarizes available research
evaluation raises operating costs and calls for skilled based on welding operations, methods for defect detection, and
personnel. deep learning methodologies like Convolutional Neural
Lack of consistency: Reliability may be diminished by Networks (CNNs) used for this purpose.
inspectors' differing assessments of flaws.
Scalability issue: Manual approaches are unable to deliver The American Welding Society (AWS) provides a
the quick and precise flaw detection needed for large-scale comprehensive overview of welding techniques, welding
manufacturing. metallurgy, and common defects encountered in welded joints.
It serves as a fundamental reference for understanding the
IJISRT25MAR1806 www.ijisrt.com 2596
Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1806
welding process and the types of discontinuities that can occur IV. METHODOLOGY
in metal welds. [1]
A. Dataset Collection
Davis discusses various non-destructive testing (NDT) The dataset is comprised of labeled weld images gathered
methods for weld inspection, such as radiographic testing (RT), from industry sources and public databases. It contains six
ultrasonic testing (UT), and eddy current testing (ECT). The classes: Good Weld, Burn Through, Contamination,
study highlights the strengths and limitations of each method, Misalignment, Lack of Penetration, and Lack of Fusion.
emphasizing the importance of accurate defect detection Rotation, flip, brightness adjustment, and noise addition data
techniques in industrial applications. [2] augmentation techniques are used to enhance model
generalization.
Smith and Johnson examine the causes and prevention of
welding defects, identifying common issues such as porosity, B. Data Preprocessing
cracks, and incomplete fusion. Their work provides insights
into defect formation mechanisms and suggests methods to High to prepare images for training:
minimize welding anomalies through process control and Image normalization: Normalizes pixel values to the same
quality assurance.[3] range.
Grayscale conversion: Minimizes computation complexity
Rao explores the application of machine learning while maintaining the most significant features.
techniques in welding, discussing various predictive models Data augmentation: Enhances dataset diversity, which
and classification algorithms used to detect welding defects. enhances model robustness.
The study highlights the potential of artificial intelligence (AI)
in enhancing the accuracy and efficiency of defect
identification processes. [4]
Brown focuses on advancements in ultrasonic testing
(UT) for weld inspection. The study introduces novel
techniques that leverage signal processing and AI algorithms to
improve defect characterization and localization. [5]
Lee provides a more focused investigation into deep
learning applications for weld defect detection. The study
explores the use of CNNs to analyze weld images and detect
anomalies with high accuracy. The results demonstrate the
effectiveness of deep learning models in automating the defect
detection process and reducing reliance on manual inspection.
[6]
III. PROBLEM STATEMENT
Welding is an elementary process across different
industries wherein faults can reduce structural integrity as well
as compromise safety. Handheld inspection is a time-
consuming process with an element of randomness and hence
lacks consistency in controlling quality. This research aims at Fig 1: Workflow Diagram
the creation of an automatic weld fault detection system
employing Machine Learning (ML) and Convolutional Neural C. CNN Model Structure
Networks (CNNs) to deal with the issue.
The CNN Model Structure Consists of:
The system will initially detect welding presence in an
image such that only the samples of interest are analyzed. It will Convolutional layers: Detect spatial patterns and features.
then categorize weld quality into any one of six classes: Good Pooling layers: Down-sample with minimal loss of
Weld, Burn Through, Contamination, Misalignment, Lack of important information.
Penetration, and Lack of Fusion. Process automation will Fully connected layers: Map features to classification
improve industrial quality control by minimizing inspection output.
time and maximizing accuracy of defect detection. SoftMax activation function: Maps probabilities to classify
weld defects.
IJISRT25MAR1806 www.ijisrt.com 2597
Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1806
Fig 2: Deep Convolutional Neural Network for Weld Defect Classification based on ResNet50[7]
D. Model Training and Evaluation V. CONCLUSION AND FUTURE WORK
The model is trained on the categorical cross- entropy loss Enhancing dataset diversity: Collecting more diversified
function and optimized with Adam optimizer. images to enhance generalization capability.
50 epochs with a batch size of 32 are performed to learn
efficiently. Real-time deployment: Utilizing the model in factory
Performance is examined in the form of accuracy, precision, settings for immediate fault detection This research suggests
recall, F1-score, and a confusion matrix to look for a welding fault detection system based on CNN that can
misclassifications. accurately detect six types of defects. The developed model
enhances factory efficiency, reduces human error, and allows
enhanced quality control in industrial operations. The future
directions are:
Hybrid deep learning techniques: Combining CNNs and
transformers to obtain improved defect classification.
Deep learning-based automatic welding inspection greatly
enhances quality control, lowers operating expenses, and raises
the reliability of welded structures in most industries.
REFERENCES
[1]. American Welding Society (AWS). (2020). Welding
Handbook. Miami, FL: American Welding Society.
[2]. B. A. Davis, "Non-Destructive Testing Methods for
Welds," Journal of Materials Engineering, vol. 45, no.
3, pp. 123-130, 2021.
[3]. J. Smith and R. Johnson, "Welding Defects: Causes
and Prevention," International Journal of Welding
Science, vol. 12, no. 2, pp. 45-60, 2022.
[4]. K. R. K. Rao, "Machine Learning Applications in
Welding," Journal of Manufacturing Processes, vol. 35,
pp. 123-135, 2020.
[5]. M. T. Brown, "Advancements in Ultrasonic Testing for
Fig 3: Sample Output Weld Inspection," NDT & E International, vol. 112, pp.
IJISRT25MAR1806 www.ijisrt.com 2598
Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1806
123-130, 2020.
[6]. S. Lee, "Deep Learning for Weld Defect Detection,"
IEEE Transactions on Industrial Informatics, vol. 16,
no. 4, pp. 2345-2353, 2020.
[7]. Palma-Ramírez, R., Ross-Veitiá, J., & Rodríguez, D.
(2024). Deep convolutional neural network for weld
defect classification in radiographic images. Journal of
Manufacturing Processes, 78, 345-356.
