Key highlights
Over 85 percent
defect detection accuracy
Up to 70 percent
less manual effort
Real-time prediction in 0.5 seconds
Challenges
1.
Noise, poor contrast, and inconsistent sizes reduced image quality.
2.
Extracting relevant features from weld images posed challenges.
3.
Detecting welding defects manually led to errors and delays.
4.
Classifying weld quality remained essential
Solution
1.
Users uploaded weld images, and the system preprocessed them for noise reduction, contrast enhancement, and resizing.
2.
The system extracted weld features such as edges, textures, and patterns using image processing.
3.
It detected defects such as cracks, porosity, undercuts, and incomplete fusion to classify quality.
4.
The system generated a report on welding quality with detected defects, locations, and severity.
Impact
Enhanced quality inspection Identified defects with precision and improved localization.
Structured defect classification Categorized defects systematically for better analysis.
Automated reporting Generated reports instantly with minimal manual effort.
Real-time edge predictions Delivered instant insights directly on edge devices.