Real-time tracking of the characteristics of strands in OSB production lines

Abstract

Strand characteristics, i.e. orientation, length, width, and size, have a substantial effect on the mechanical properties of Oriented Strand Board (OSB). In this study, an automatic method was established to obtain the characteristics of the strands on the surface layer of the OSB mattress in real time by taking images and using neural networks. The Segment Anything Model was used to extract surface layer strands, the YOLOv5 model was used to distinguish and position strands, and the minimum bounding rectangle algorithm was used to measure characteristics of each strand. Based on the results obtained from manual measurement, the performance of the automatic method was evaluated. In laboratory tests, this method presents great performance in extracting and distinguishing characteristics of strands. This method also shows good adaptability for production line application. In the production line, around 80% of strands can be correctly extracted and distinguished, with a strong correlation between manual measurements and automatic method results (R² > 0.97). It takes 37.7ms to process one image containing approximately 500 strands. Strand orientation in the production line nearly concords with normal distribution (N (-1.25, 30.5²)). The size of strands significantly affects the relative intensity of the strand orientation (with P < 0.05). There is a positive and linear relationship between the strand size and the orientation of strands. The outputs of this study contribute to a better understanding and management of OSB manufacture in the production line.