Industries / Food, farming and forestry / Forestry and logging / Timber and sawmills
Grading
Sorting each piece of lumber into a grade that tells a buyer what it can be used for, judged by its appearance, its strength or both.
7
Milestones achieved, from 1898 to 2018.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1898
Hardwood gets common rules
Thirty-seven lumbermen founded the National Hardwood Lumber Association (NHLA) in Chicago in 1898 "to establish uniform standards for grading and trading hardwood lumber". Its rules grade a hardwood board by the size and number of clear pieces, called cuttings, that it will yield. NHLA held its first inspector training class in Chicago in 1920 and moved the school to Memphis, Tennessee, in 1948.
Credit
National Hardwood Lumber Association
1924
One standard for softwood
Early in 1922, at the lumber industry's request, US Secretary of Commerce Herbert Hoover offered his department's help to simplify and standardize lumber. The result, published in 1924 as Simplified Practice Recommendation R 16, was the American Lumber Standard, now Product Standard PS 20. Since 1953 the American Lumber Standard Committee, appointed by the Secretary of Commerce, and its independent Board of Review have accredited the agencies that grade-mark US lumber.
Credit
Herbert Hoover, the US Department of Commerce and the lumber industry
1960s
Machines measure stiffness
Grading machines came to North America in the 1960s. They bent each piece slightly and measured the load needed to do it, which gives its stiffness, written E for modulus of elasticity. Lumber sorted this way is sold as machine stress rated (MSR), and mills must run daily strength tests to check the machines. X-ray density measurement followed in the 1990s, and sound-wave machines later.
Credit
the Southern Pine Inspection Bureau's account names no first maker (see Checks)
2011
Stains fool the scanner
In a 2001 test on 89 red oak boards, a Virginia Tech and Forest Service scanner with lasers, color cameras and X-ray was 31% more accurate than the mill's line graders, who overvalued the lumber by close to 20%. But in a 2011 test of a laser-and-camera prototype on 176 yellow-poplar and red oak boards, the machine graded 17% of boards higher and 43% lower than a human expert, mainly because it read natural and mechanical stains as defects.
LessonA grading machine must tell harmless marks from real defects, and human graders make errors too.
Credit
Philip Araman (US Forest Service), D. Earl Kline, A. Lynn Abbott and colleagues at Virginia Tech
2012
Europe requires strength grades
Since 1 January 2012, under EU Regulation 305/2011, structural solid timber sold in the European Union must carry the CE mark, which requires strength grading under the standard EN 14081. Grading is done by eye under national rules, such as Italy's UNI 11035 or Germany's DIN 4074-1, matched to common strength classes, or by approved machines that measure stiffness from vibration or sound speed.
LessonThe Italian authors noted that grading machines could be too expensive for their country's mostly small, family-owned carpentry firms.
Credit
the European Parliament and Council, and the European standards body CEN
2013
Southern pine marked down
In January 2012 the American Lumber Standard Committee cut the design values, the figures engineers use for safe loads, of No. 2 southern pine 2x4s by as much as 30%. After the Southern Pine Inspection Bureau and Timber Products Inspection destructively tested more than 7,400 full-size pieces, the committee's Board of Review approved new values for visually graded southern pine on 31 January 2013, effective 1 June 2013.
LessonThe bureau says a change in the timber resource is one of several things that can change lumber's strength, so grade values need retesting as forests change.
Credit
Southern Pine Inspection Bureau, Timber Products Inspection and the American Lumber Standard Committee
2018
Hardwood graded by machine
A 2017 test at a Midwestern hardwood sawmill ran kiln-dried boards of nine species through a Microtec Goldeneye 300 scanner, which combines color, black-and-white and profile cameras, lasers and X-ray, with GradeView grading software. It matched the human-verified grade on 92.2% of boards and the value on 99.5%, above the NHLA sales code minimums of 80% on grade and 96% on value. Earlier studies put human inspectors at 48% to 75% accuracy. The paper says the US Forest Service first introduced automated hardwood grading in 1983.
Credit
Logan Wells, Rado Gazo and co-authors, Purdue University, with Microtec
Watch for
AI strength grading at Värö
Södra's Värö sawmill in Sweden installed MiCROTEC's Goldeneye Transverse SE, described as the world's first strength-grading scanner driven by artificial intelligence. It inspects four sides of up to 240 boards per minute and grades to EN 14081. The report gives no accuracy figures and names no certifying body.
Achieved whena named certification body publishes its approval of the scanner's grading under EN 14081 and the mill reports grade yields from routine production.
Claimed by
MiCROTEC and Södra, reported by Wood Central, September 2025
AI hardwood grading by Comact
Comact, a sawmill equipment maker, says its AI grading tools, including GradExpert, cover more than 35 hardwood species, that it is "the only company" to have trained its AI models specifically on hardwood, and that mills often see "improvements of up to 10% in productivity".
Achieved whena test at a working mill by someone outside Comact reports its accuracy on grade and on value against NHLA inspectors.
Claimed by
Comact, in an article published by NHLA, November 2025
Open
01
Hardwood grades from open data
A public set of hardwood boards photographed on both faces and both edges, each graded by an NHLA-certified inspector, and an open model that grades a board from those photos, tested with low-cost cameras or a phone.
The briefCredit · unclaimed02
Tell stains from defects
A labeled image set of hardwood boards in which marks that do not lower the grade, such as stains, dirt and chain marks, are labeled apart from grade-lowering defects, with labels checked by a certified inspector, and a model that sorts the two.
The briefCredit · unclaimed
Open 01 · the brief
Hardwood grades from open data
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of hardwood boards photographed on both faces and both edges, each graded by an NHLA-certified inspector, and an open model that grades a board from those photos, tested with low-cost cameras or a phone.
Why it is still open
The systems shown to meet NHLA's accuracy thresholds are commercial scanners with closed software. The largest public wood-surface image set, 20,275 images with more than 43,000 labeled defects from VSB-Technical University of Ostrava, labels defects but no lumber grades. A search of GitHub and Hugging Face found no public image set graded to NHLA rules.
Built with what exists
A phone or industrial cameras over a roller table; the Ostrava defect set for pretraining; a detection or segmentation model in PyTorch, or a small open vision model fine-tuned on one graphics card, such as Liquid AI's LFM3 (see the Note "One GPU to fine-tune"); and the NHLA rule book, coded directly or through the Forest Service's UGRS program (1998), which grades a board by NHLA rules from a map of its defects.
First step
Ask NHLA's National Hardwood Academy in Memphis, or a local hardwood mill, to let you photograph a pack of boards right after an inspector grades them.
Done looks like
At least 1,000 boards of at least three species published under an open license, and a model that scores at least 80% on grade and 96% on value on boards it has never seen, the NHLA sales code thresholds.
Open 02 · the brief
Tell stains from defects
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A labeled image set of hardwood boards in which marks that do not lower the grade, such as stains, dirt and chain marks, are labeled apart from grade-lowering defects, with labels checked by a certified inspector, and a model that sorts the two.
Why it is still open
Stains were the main reason a 2011 Forest Service test scanner graded 43% of boards lower than a human expert, and in November 2025 Comact still wrote that "dirt, spots, and chain marks often mimic natural defects". The large public defect set has no classes for dirt or chain marks.
Built with what exists
Cameras at a mill's grading chain, Label Studio for labeling, the Ostrava defect set for pretraining, and PyTorch.
First step
Photograph 100 boards at a hardwood mill's grading chain and have the inspector circle each mark they ignored and each defect they counted.
Done looks like
At least 2,000 labeled marks from at least two mills, and a model that calls fewer than 5% of harmless marks defects while missing fewer than 5% of real defects, on boards from a mill it was not trained on.