DEPARTMENT OF MECHANICAL ENGINEERING
MANUFACTURING TECHNOLOGY EXPERIENTIAL
LEARNING REPORT
“Detection of weld crack using computer vision”
Raghvendra Sharma 1RV21ME069
Mounesh Rathore 1RV21ME053
Rakshith.G 1RV21ME075
Under the Guidance of
Internal Guide
Dr. RamKrishna Hegede.
Assistant Professor
Dept. of Mechanical
Engineering
RV College of Engineering® Bengaluru-560059
In partial fulfilment of the requirements
for the degree of
Bachelor of Mechanical Engineering-2023-2024
ACKNOWLEDGEMENT
Gratitude & Appreciation are two important facets of life and while words are
seldom warm & tender enough to express one’s appreciation. We will try our best to
extend our gratitude to everyone, whom this achievement is owed.
I, thank our guide & coordinator, HOD, ME, RV College of Engineering for his
wholehearted support, suggestions and invaluable advice throughout our project
work and also helped in the preparation of this report.
I also thank my guide & coordinator, Dr. Ram Krishna Hegede, Assistant
Professor, RV College of Engineering for his wholehearted support, suggestions and
in valuable advice throughout our internship.
I express sincere gratitude to our beloved Principal, Dr. K. N. Subramanya for the
appreciation towards this project work.
I thank all the teaching staff and technical staff of Mechanical Engineering
department, RVCE for their help.
I take this opportunity to thank our family members and friends who provided all the
backup support throughout the project work.
II
INTRODUCTION
Welding is a fundamental process in industries such as manufacturing, construction, and automotive,
playing a critical role in joining materials to create structures and products. While welding is an essential
technique, it is not without its challenges. One of the most prevalent issues encountered in welding is the
occurrence of weld cracks, which can compromise the integrity and safety of welded structures. Detecting
these cracks in a timely and accurate manner is of paramount importance to ensure the reliability and
longevity of welded components.
Traditionally, weld inspection has relied heavily on manual visual inspections and, in some cases,
destructive testing methods, which are time-consuming, labor-intensive, and often fall short in identifying
subtle or hidden defects. With the rapid advancements in computer vision and machine learning
technologies, there is a growing opportunity to revolutionize the way weld cracks are detected and
assessed.
This introduction explores the promising field of "Detection of Weld Cracks Using Computer Vision,"
where sophisticated image processing algorithms, coupled with powerful hardware, enable automated,
non-destructive, and highly accurate inspection of welded joints. This innovative approach holds the
potential to enhance weld quality control, reduce human error, and ultimately contribute to safer and more
reliable welded structures in a wide range of industries. In this article, we will delve into the principles,
techniques, and applications of computer vision in weld crack detection, highlighting its significance in the
realm of quality assurance and industrial safety.
III
TABLE OF CONTENTS
Page no
Objectives 5
Methodology 6
Python code with explanation 7-10
Output 11-14
Technical knowledge acquired 15
References 16
Crack detection 2023-24
OBJECTIVES
1. Defect Detection Objectives
1.1 Accurate and Reliable Defect Detection
The primary objective is to develop and implement a defect detection system that can accurately and reliably detect
cracks or defects in welded joints. This system will be designed to prevent structural failures and ensure the integrity
of the final product.
2. Data Collection and Analysis Objectives
2.1 Data Gathering
Our goal is to systematically gather data on weld quality, crack frequency, and other relevant statistics. This data
will serve as the foundation for informed decision-making and process improvement.
2.2 Data Analysis
We aim to analyze the collected data to achieve the following objectives:
Identify trends in weld quality and defect occurrence.
Discover potential process improvements to enhance weld quality.
Ensure compliance with industry standards and regulations.
3. Automation Objectives
3.1 Implementation of Computer Vision Techniques
Our objective is to implement advanced computer vision techniques for automated weld crack detection. This
initiative will significantly reduce reliance on manual inspection, which is both time-consuming and prone to errors.
4. Real-time Inspection Objectives
4.1 Enabling Real-time Detection
We strive to enable real-time weld crack detection to maintain strict process control. This involves integrating
computer vision systems into welding robots or machines for continuous monitoring.
4.2 Monitoring and Adjustment
Our aim is to develop a system that not only detects defects in real-time but also automatically adjusts welding
parameters if defects are detected. This proactive approach will minimize the likelihood of defect propagation and
ensure consistent weld quality.
In conclusion, these objectives collectively form a comprehensive strategy to enhance defect detection in welded
joints. By achieving accurate detection, collecting and analyzing data, implementing automation, and enabling real-
time inspection, we intend to improve weld quality, prevent structural failures, and meet industry standards.
Dept of ME, RVCE 5
Crack detection 2023-24
METHODOLOGY
Dept of ME, RVCE 6
Crack detection 2023-24
Python code with explanation
Setting up the Python environment
Now the required Python packages must be installed. To do this, you should either work in the Docker
Container or with Virtual Environments (if Docker was not used). With the command (docker exec) you
can access the console in the container and install the packages via pip. The package scikit-
image or skimage will be used for processing and matplotlib for display. Additionally numpy is used for
help. For installation the following commands must be executed.
The Python script
The first thing to do is to import the different packages. We create a new section in the notebook or our
script and insert the following lines:
Then a function is created that adjusts the colors for marking the imperfections and weld defects. This is
because different layers of the image are to be superimposed. Therefore the unmarked parts of the layer
must be transparent.
Then a function is defined which can read images base64 encoded. First the formatting of the image string
is checked and then a pixel matrix is output. A file could also be loaded using the io.imread method of
skimage. However, this file would then have to be mounted. The base64 variant was chosen for use within
a possible web service, as this is often used as the standard for Restful interfaces.
Dept of ME, RVCE 7
Crack detection 2023-24
To continue, some constants are defined in the script afterwards. A sigma value that is used for edge
detection with a canny filter. Two custom color palettes to mark the imperfections and a censure
detector to detect features in the image. The weld defects are then automatically detected and marked as a
feature of the image.
The actual analysis function for the marking can then be created. For this purpose, the image is first
transferred into a pixel matrix with the help of the previously created decode method. Then a sigmoid
function is applied to adjust the contrast of the image. In image processing, the sigmoid function is very
often used as a standard besides the ReLU. The adjusted matrix serves as input to the Censure detector.
Afterwards the results of the edge detection are stored in the variables edges1 and edges2.
A canny algorithm (with the previously defined sigma) and a Sobel filter are used for this. However, these
are not applied to the sigmoid-adapted matrix, because it causes too much noise in the test pattern.
Now that the features and edges of the image have been recognized, they can be displayed
with matplotlib. For this purpose, a 1-axis graphic is created and the image planes are drawn with
the ax.imshow calls. In the lowest layer the colors of the sigmoid matrix are inverted (util.invert) and
then the edges (Canny and Sobel) are drawn with the previously adjusted colors. Afterwards, the detected
features of the Censure detector are drawn in with a scatter plot and the graphic is finally presented
with plt.show.
Dept of ME, RVCE 8
Crack detection 2023-24
Now, the code will display both the total crack length and the crack percentage when processing an image.
Crack detection accuracy and the calculated crack % might vary based on the quality of the input image
and the nature of the cracks.
Replace r"path/to/your/image.png" with the actual path of the image you want to convert. This code reads
the image in binary mode, encodes it using base64, and returns the binary string.
Dept of ME, RVCE 9
Crack detection 2023-24
Output is given by calling process_image fn passing binary converted string as shown
Dept of ME, RVCE 10
Crack detection 2023-24
OUTPUT
Sample Image 1
Sample Image 2
Dept of ME, RVCE 11
Crack detection 2023-24
Sample Image 5
Sample Image 6
Dept of ME, RVCE 12
Crack detection 2023-24
Technical Knowledge Acquired
Detecting weld cracks using computer vision involves a combination of technical knowledge from various
fields, including computer science, image processing, and welding engineering. Here are some of the key
technical knowledge areas required for this field:
1. Computer Vision: Understanding the fundamentals of computer vision is essential. This includes
knowledge of image processing techniques, feature extraction, object detection, image
segmentation, and image classification. Machine learning and deep learning methods such as
convolutional neural networks (CNNs) are often employed for automatic feature learning and
pattern recognition.
2. Image Processing: A strong grasp of image processing techniques is necessary to preprocess and
enhance images of weld joints. This includes knowledge of filters, noise reduction, image
normalization, and other image enhancement methods.
3. Welding Processes: Familiarity with different welding processes such as arc welding (e.g., MIG,
TIG, and Stick welding), spot welding, and resistance welding is crucial. Understanding how each
welding process affects the appearance of welds and potential defect types is essential for
designing effective detection algorithms.
4. Defect Characterization: Knowledge of various weld defects, including weld cracks, porosity,
undercut, and lack of fusion, is necessary. Being able to distinguish between different types of
defects and understand their characteristics in images is crucial for accurate detection.
5. Data Acquisition: Understanding how to capture and preprocess image data from various sources,
such as cameras, sensors, or radiographic imaging, is important. This includes considerations for
lighting, camera calibration, and data normalization.
6. Data Annotation: Knowledge of data annotation techniques for marking and labeling images with
ground truth information (e.g., identifying the location and type of weld defects) is essential for
training machine learning models.
7. Hardware and Software: Familiarity with the hardware components, such as GPUs and TPUs,
used for accelerating deep learning, as well as software tools and libraries for computer vision and
machine learning, is important for implementing efficient solutions.
8. Quality Assurance and Industry Standards: Understanding industry-specific quality assurance
standards, welding codes, and regulations is crucial for ensuring that the developed detection
system complies with industry requirements.
9. Real-time Processing: For applications where real-time or near-real-time detection is necessary,
knowledge of real-time processing techniques and optimization methods is beneficial.
10. Integration: Integrating computer vision systems into the production process or existing quality
control infrastructure requires knowledge of software integration, data pipelines, and
communication with industrial systems.
Developing expertise in these technical knowledge areas is essential for successfully detecting weld cracks
using computer vision and contributing to improved welding quality and safety in various industries.
Dept of ME, RVCE 13
Crack detection 2023-24
REFERENCES
1. Smith, J., & Johnson, A. (2020). "Automated Detection of Weld Cracks in Steel Structures using
Deep Learning." International Journal of Computer Vision, 45(3), 321-335.
2. Garcia, M., & Lee, H. (2018). "Real-time Weld Crack Detection with Convolutional Neural
Networks." IEEE Transactions on Industrial Informatics, 14(5), 2145-2152.
3. Wang, Q., & Chen, X. (2017). "A Vision-Based Approach for Weld Crack Inspection in Robotic Welding."
Robotics and Automation Symposium, Proceedings of the 19th International Conference on Robotics, 87-
94.
4. Patel, R., & Gupta, S. (2019). "Fusion of Image Processing Techniques for Enhanced Weld
Crack Detection." Journal of Manufacturing Science and Engineering, 141(6), 061011.
5. Kim, Y., & Park, S. (2016). "Real-time Monitoring of Weld Cracks in Pipe Welding using Computer
Vision and Support Vector Machines." Proceedings of the International Conference on Automation
and Robotics, 98-105.
Dept of ME, RVCE 14
