| dc.description.abstract | Pipelife Norge AS is working on automating quality control for their pipes, specifically monitoring the exterior wall colour. Currently, operators manually inspect the pipes and rely on their experience to judge the colour.
When a colour monitoring system take a measurement of the colour they are usually measuring and comparing the colour against a standard, but at the factory a mathematical standard does not exist.
To modernize and streamline this process, a colour quality camera station has been constructed in this study. This station consists of a lightproof metal frame through which newly produced pipes can travel. The station is equipped with LED’s and web cameras controlled by a Raspberry Pi 4 B, designed to capture the colour defects of each pipe in a more objective and consistent manner than human operators.
Approximate 1100 images of brown and red PVC 110ø pipes were collected to the dataset. The histogram used for the evaluation is based on the L*, C*_ab and h_ab from the CIELAB colour space.
A set of machine learning models are developed using a voting ensemble method to evaluate the pipes. The model’s performance was largely satisfactory, producing reliable results with most images. However, some challenges were observed with images that even a human eye would struggle to discern accurately. This indicates that while the system has made substantial strides in automating colour quality control, there are still areas for improvement and fine-tuning. | |
