Case Study
Manufacturing
AI & MLCanny Edge Detection & Contour AnalysisComputer VisionYOLO
The client operates in the manufacturing sector and aims to improve product quality control by reducing manual inspections and automating defect detection using AI-powered solutions.
The client operates in the manufacturing sector and aims to improve product quality control by reducing manual inspections and automating defect detection using AI-powered solutions.
One of the primary challenges in implementing AI-powered quality inspection was acquiring a diverse and well-labeled dataset for training. Collecting a large volume of product images with accurately identified defects required significant time and resources. Additionally, ensuring the accuracy and efficiency of the model was crucial.
A trade-off had to be made between a highly accurate model that was computationally expensive and a faster model that could be deployed in real-time production lines but might compromise precision. Lastly, scalability and adaptability were key concerns. The AI system needed to be flexible enough to accommodate different products across various manufacturing lines, requiring a robust training process that could generalize defect detection across multiple product categories.
To address these challenges, a multi-step AI-powered quality inspection system was developed. The process began with object detection using YOLO models to identify and classify products on the production line. Once an object was detected, segmentation techniques, including contour analysis and Canny edge detection, were used to isolate potential defects. A YOLO segmentation model was then trained to differentiate between defect-free and defective areas, using labeled masks for precise identification.
To improve model performance, extensive training and refinement were conducted, optimizing parameters such as epochs, batch size, and input image size. The system underwent rigorous validation using metrics like mean average precision (MAP), accuracy, precision, and recall to ensure its effectiveness. This AI-driven approach significantly automated the defect detection process, reducing reliance on manual inspections and enhancing overall production efficiency.
Model Used
YOLOv8m-seg.pt
Classes
Background, Mask
Training Parameters
Epochs: 100, Batch Size: 16, Image Size: 640
Algorithms Used
YOLO Object Detection, YOLO Segmentation, Canny Edge Detection, Contour Analysis
The model achieved an accuracy of 86.7%, indicating strong overall performance in detecting defects.
The mask precision of the model is 80%, showcasing its ability to correctly identify defect areas without many false positives.
The model has a mask recall of 50%, highlighting that it may miss certain unique or less frequent defects.
The Mean Average Precision (MAP) is 42%, reflecting the model’s average effectiveness across all defect categories.
The Minimum Average Precision stands at 55%, indicating the lowest precision among all defect classes.
The AI system demonstrates high accuracy and precision in defect detection, ensuring reliable identification of common defects.
The lower recall (50%) suggests that the model could miss some unique or uncommon defects, limiting its comprehensive defect coverage.
Expanding the training dataset with a wider variety of defect samples could significantly improve the model's recall and overall detection capability.
"Thanks to Lucent Innovations, our defect detection process has improved significantly with their AI-powered solution. Their team delivered excellent results and made it easy for us to understand how to make further improvements. We highly recommend their services to anyone looking to enhance their manufacturing processes."
David Williams
Operations Director
Expanding the dataset to include a wider range of defects will be a priority to improve recall and overall model accuracy. Fine-tuning the AI model will further enhance its ability to detect even the smallest product anomalies, ensuring a more comprehensive quality inspection process.
Additionally, adapting the system for use in other industries requiring precision-based defect detection presents a significant opportunity for broader implementation. Optimizing the AI model for seamless real-time integration into manufacturing lines will enhance efficiency and reduce downtime. Continuous collaboration with the client will ensure that the system evolves to meet changing industry needs, making it a reliable and adaptable solution for automated quality inspection in the long run.