Accelerated statistical prediction of lumber grades and part yields

摘要

Assignment of lumber grades by lumber scanning followed by computer grading requires faster hardware and/or software. This study explored the potential for determining lumber grades by a linear discriminant analysis model. Designating lumber grades byapplying a discriminant analysis model will dramatically increase software execution speed. Development of special lumber grades would be much easier if they were described by a discriminant analysis model. Quantitative variables expected to discriminate between lumber grades were tested for significance. Eight of 15 variables tested were found to be significant. The best four-variable model was chosen to reduce the amount of data collection and computation. The model successfully classified 92.8,75.0,66.3, and 74.1 percent of sample lumber in the lumber grades FAS, 1C, 2AC, and 3AC, respectively. Overall accuracy for all four lumber grades was 74.3 percent. Parts yields from the lumber graded by the discriminant analysis model were nearly identical to the lumber graded by National Hardwood Lumber Association grading rules for the three cutting orders involved in this study. The discriminant analysis software assigned lumber grades 77,000 percent faster than the lumber grading software tested in this study.