Product-adapted grading of Scots pine sawn timber by an industrial CT-scanner using a visually-trained machine-learning method

Figures & data

Figure 1. Material flow chart. The procedure was as follows, following the nomenclature in the flow-chart above 1, data collection of the entire dataset through the entire process. 2, simulated sorting comparing the sorting outcome of the rule-based approach with the PLS-based approaches. A, the 156 logs selected for the study. B, CT-scanning of all the logs by the CT Log. C, the rule-based approach optimising and deciding on the sawing. $\left({\mathrm{D}}_1\right)$ and $\left({\mathrm{D}}_2\right)$, two PLS models correlating variables $\left(\mathbf{X}\right)$ from the CT-scanning and the product-grade ${\mathbf{Y}}_{\mathbf{p}}$, or image-grade $\left({\mathbf{Y}}_{\mathbf{i}}\right)$, quality references. C and D can classify the virtual sawn timber as the main product intended for the customer or, if not suitable for this customer, as a main product for a different customer. E, additional rules applied to C and D to remove outliers only during the simulated sorting. F, sawing the log. G, kiln drying the sawn timber. H, the Boardmaster at the dry-sorting station capturing images of the sawn material whose image-grade is used as a quality reference for $\left({\mathrm{D}}_2\right)$. I, additional rules applied to remove outliers. J, a piece of sawn timber delivered to the customer receiving a quality product-grade $\left({\mathbf{Y}}_{\mathbf{p}}\right)$ based on K. K, split, milled, and quality graded wall panels by the customer.

Figure 2. Example of images used for image-based grading. All four faces of the sawn timber are shown.

Table 1. Material quality grade distribution.

	Product-grade	Image-grade
Grade A	189	114
Grade B	114	189
Total	303	303
Grade A (%)	62	38

Table 2. Sorting outcome using the customer-adapted rule-based grading in the CT Log and using the complementary rules in the CT Log and the Boardmaster.

		Graded as A by
	All sawn timber	CT Log	Boardmaster
Product-grade A	189	159	153
Product-grade B	114	69	47
Total	303	228	200
Delivered (%)	100	75	88
Delivered grade A (%)	62	70	77

Table 3. Sorting outcome using the product-trained PLS model in the CT Log and using the complementary rules in the CT Log and the Boardmaster.

		Graded as A by
	All sawn timber	CT Log	Boardmaster
Product-grade A	189	102	101
Product-grade B	114	21	13
Total	303	123	114
Delivered (%)	100	41	93
Delivered grade A (%)	62	83	89

Table 4. Sorting outcome using the image-trained PLS model in the CT Log and using the complementary rules in the CT Log and the Boardmaster.

		Graded as A by
	All sawn timber	CT Log	Boardmaster
Product-grade A	189	105	104
Product-grade B	114	19	11
Total	303	124	115
Delivered (%)	100	41	93
Delivered grade A (%)	62	85	90

Table 5. Sorting outcome specific to the complementary rules (E) in the CT Log for the rule-based grading method and the two PLS-based methods.

	Rejected by complementary rules in
	Rule-based	Product-trained	Image-trained
Product-grade A	6	3	5
Product-grade B	22	10	16
Total	28	13	21