Post by Abhinand Nair

I am excited to share my latest portfolio project: an automated AI-Based Weld Defect Detection and Quality Certification System, designed to streamline quality assurance in shipbuilding and structural engineering. Manual weld inspection is slow and subject to human error. In safety-critical sectors, missing a single defect can cause catastrophic failure. To address this, I built an end-to-end computer vision pipeline that detects surface weld defects and instantly generates official compliance certificates. Here is a summary of the technical achievements and results: Dataset Pipeline: Verified and stratified a raw dataset of 2,991 high-resolution images mapping 3,750 annotated defect instances across 8 classes. Iterative Training and Optimization: Trained a baseline YOLOv8 Nano model (30 epochs) to verify the pipeline. Upgraded to a YOLOv8 Small architecture (100 epochs total) using the AdamW optimizer and a 10x reduced learning rate (lr = 0.0001) for the second phase to prevent catastrophic forgetting. Improved overall Precision from 32.2% to 41.0%. Increased Recall from 24.4% to 36.2%. Lifted overall [email protected] from 21.3% to 33.1%. Achieved a significant 26% accuracy gain on critical Crack detection and a 2.5x gain on Slag Inclusion detection. Industrial Quality Certification: Developed a compliance engine (based on AWS D1.1 and ISO 5817 standards) that parses model outputs, applies density checks for porosity, enforces zero-tolerance rules for cracks, and generates structured PDF inspection reports. Deployment: Built a responsive, dark-themed Streamlit application and deployed it to the cloud for real-time inference and report generation. Check out the project: Live Application: https://lnkd.in/gbpE_SgP GitHub Repository: https://lnkd.in/gMWpbZqX I would love to hear your feedback on the architecture and MLOps workflow. #MachineLearning #ComputerVision #ObjectDetection #YOLOv8 #Streamlit #AI #MLOps #SoftwareEngineering