Quality improvements in forest products industry : classification of biological materials with inherent variations

Abstract: Classifying logs and boards has an essential economic impact on the forest products industry. In contrast to other process industries the sawmills do not try to homogenize the raw material, but instead try to utilize the inherent properties of every single log and board in the best way. The aim of this thesis is to improve the quality control and decrease the costs for poor quality by introducing new methods to measure and describe the raw material. The work is focused on the grading procedures, and thereby 1) analyzing the relevance of the current classification systems and investigating the characteristics that are the most important ones in today´s grading procedures; 2) evaluating models developed to detect interior defects in logs automatically, giving suggestions for improvements; 3) suggesting ideas and methods for a classification system for tomorrow. The investigation comprises partly about 1100 Scots pine (Pinus silvetris) logs harvested on 16 randomly selected stands in Sweden and graded both by graders and by an automatic equipment; partly around 600 pine logs from permanent sample plots in Sweden and scanned by a CT-scanner (Computed tomography). The predictability between the grade of a log and its boards is very low, around 20%. Repeated investigations show that two graders assess the same grade on about 50% of the logs and boards, i.e., they judge the properties equally. The results also emphasize that the current manual classification systems are not in concordance with the customer´s demand and the natural variations in the material cannot be handled in an efficient way by current grading rules and transformed by human beings. The CT-scanner is used to measure the interior properties (defects) of a log. The validation of the methods and models shows that when the accuracy is improved for small knots, the models identify a knot with an accuray of +/- 5 mm. An embryo to a property index, PI, that gives an individual description of the inherent properties of logs and boards, is given. Based on customer-orientation and various strategies for describing log properties, a clustering procedure can be evaluated in order to form and describe appropriate classes (clusters, grades). Then, by an allocation rule with the function for assessing a grade, every single log or board can be automatically classified into one group.