Image Recognition Solution Identifying Polyurethane Film Defects in Real Time
About Our Client
Headquartered in the Gulf region with offices across the globe, the Client is a petrochemical company with a full-scale manufacturing facility for the production of polyether polyols and polyurethane products, including polyurethane film.
Challenge
Having experienced difficulties in guaranteeing the quality of the produced polyurethane (PU) film, the Client equipped their manufacturing lines with visual inspection cameras. However, to fully modernize the quality control process, the Client needed to develop custom software to analyze camera footage in real time, detect defects in the film, and provide reports on its quality.
Solution
ScienceSoft delivered an application for image recognition running on Windows. The application uses machine learning and computer vision algorithms to detect defects in film and report the results to application users in real time.
Once launched, the application preprocesses an individual image or a batch of images and analyzes them. ScienceSoft’s data scientists used OpenCV library as a foundation for the analysis and employed the cv.findContours() method to single out such areas as a background, damaged areas, and blank area. To compute the defect area ratio, the data scientists used the OpenCV contourArea() method.
The application also clusters ‘damaged’ pixels into cracks, calculates the length and width of each defect, as well as counts the number of defects in each image. The application allows generating statistical reports reflecting the defect dynamics over time, which can be exported to Excel.
Results
The Client has obtained a desktop application that enabled them to detect defects in the PU film being produced in real time, take informed decisions about the production process, and improve production quality. In the future, the application can be integrated with the Client’s MES system allowing for the identification of the root causes of defects and providing recommendations on increasing product quality.
Technologies and Tools
PyQt, OpenCV, Matplotlib, Scikit-Learn, TensorFlow.
