Computer Vision for Manufacturing
Enhancing Manufacturing Processes with AI-Powered Defect Detection
problem
The client, a leading manufacturer of compressors for refrigeration systems, faced a significant challenge in its production process. Despite implementing Industry 4.0 technologies for quality control, the company struggled with identifying defective items, particularly compressors lacking caps on their tubes. The existing computer vision tool, mounted above the assembly lines, proved inadequate in detecting these missing caps due to variations in compressor arrangement and lighting conditions. Consequently, some compressors were damaged during the painting stage, leading to increased manufacturing defects and costs.
problem
The client, a leading manufacturer of compressors for refrigeration systems, faced a significant challenge in its production process. Despite implementing Industry 4.0 technologies for quality control, the company struggled with identifying defective items, particularly compressors lacking caps on their tubes. The existing computer vision tool, mounted above the assembly lines, proved inadequate in detecting these missing caps due to variations in compressor arrangement and lighting conditions. Consequently, some compressors were damaged during the painting stage, leading to increased manufacturing defects and costs.
solution
Our team devised a solution to address the client s manufacturing challenge by enhancing the detection of compressors and identifying missing caps on tubes, irrespective of assembly line configurations and lighting variations. Leveraging video streams from cameras positioned above the conveyor belts, we implemented real-time processing capabilities to detect absent caps and notify the assembly line control system promptly. This allowed personnel to conduct additional checks and ensure that all compressor tubes were properly equipped with caps. The solution utilized OpenCV tools for video stream processing and employed a trained YOLO object detection model renowned for its accuracy during the training phase. Additionally, historical data was utilized to refine the model s output and minimize false positives, thus achieving the desired accuracy goal
solution
Our team devised a solution to address the client s manufacturing challenge by enhancing the detection of compressors and identifying missing caps on tubes, irrespective of assembly line configurations and lighting variations. Leveraging video streams from cameras positioned above the conveyor belts, we implemented real-time processing capabilities to detect absent caps and notify the assembly line control system promptly. This allowed personnel to conduct additional checks and ensure that all compressor tubes were properly equipped with caps. The solution utilized OpenCV tools for video stream processing and employed a trained YOLO object detection model renowned for its accuracy during the training phase. Additionally, historical data was utilized to refine the model s output and minimize false positives, thus achieving the desired accuracy goal
results
The implemented solution incorporated a FullHD video camera connected to an NVIDIA Jetson TX2, enabling Edge Computing for efficient data processing closer to the source. This approach optimized infrastructure capabilities and reduced costs. With the system achieving a remarkable 99.99% accuracy in detecting missing caps, the client experienced a significant improvement, surpassing previous detection rates tenfold. Furthermore, the system s capability to immediately halt the conveyor belt and alert staff upon detecting a missing cap enhanced production efficiency and reduced the risk of damaged compressors, ultimately leading to substantial cost savings and improved product quality.
results
The implemented solution incorporated a FullHD video camera connected to an NVIDIA Jetson TX2, enabling Edge Computing for efficient data processing closer to the source. This approach optimized infrastructure capabilities and reduced costs. With the system achieving a remarkable 99.99% accuracy in detecting missing caps, the client experienced a significant improvement, surpassing previous detection rates tenfold. Furthermore, the system s capability to immediately halt the conveyor belt and alert staff upon detecting a missing cap enhanced production efficiency and reduced the risk of damaged compressors, ultimately leading to substantial cost savings and improved product quality.
tech stack
tech stack
