About the Client
One of the world-leading manufacturers of compressors for refrigeration systems.
The client produces thousands of compressors daily and it’s important to keep the high level of quality. In order to reduce the number of defective items, quality control was implemented throughout the whole production lifecycle using Industry 4.0 technologies.
The production line had cameras installed above assembly lines with the embedded computer vision tool, which turned out to be pretty outdated. This led to the fact that part of compressors was damaged on the painting stage, as the installed tool couldn’t properly identify that some of the compressor’s tubes were not covered with caps. The tool also had issues with identifying caps because of the different location of compressors on the assembly line.
Therefore it was crucial for the client to improve the manufacturing process and cut down costs by reducing the number of manufacturing defects.
Quantum team has come up with a solution that would detect a compressor and identify whether the caps are on the tubes or not, no matter how the compressor is positioned on the assembly line and disregarding other factors, such as lighting conditions.
Video stream from the already-installed cameras above the conveyor belts is used to process incoming video feed in real-time and detect the absence of a cap on any of the compressor tubes and send a warning signal to the assemble line control system. In case of a warning signal, additional checks should be taken by the personnel to ensure that all the tubes have caps on them.
FullHD video camera is attached to NVIDIA Jetson TX2, on which the solution is running. This method of Edge Computing allows processing data closer to location, thus optimizing infrastructure capabilities and saving costs. The developed solution processes the incoming video and detects the absence of a cap on any of the compressor tubes. If the system detects that the cap is missing, it sends an alert and stops the conveyor belt.
The developed solution has offered 99.99% accuracy, which is 10 times more than the previous one.
Let's discuss your idea!
The video stream was processed using OpenCV tools. The input was sent to a trained YOLO object detection model that was chosen because of a higher accuracy achieved during the model training process. The model’s output was processed using historical data to reduce False Positives rate and achieve the target accuracy goal. An alert was sent if objects that didn’t meet the set requirements (each compressor has to have a certain amount of caps attached to it) were detected on the conveyor belt.