How Lumber Scanners and Automated Grading Systems Work

A lumber scanner turns each board into measurements, its outline, thickness, grain, and defects, and a grading or optimizing system turns those measurements into a grade, a cut, or a sort decision. UTEC Industrial designs, engineers, machines, fabricates, and installs custom material handling systems for aerospace and heavy industry from its Spokane Valley, WA facility, integrating Allen-Bradley PLC and motion control with in-house CNC machining, heat treating, and stress relief. This article explains the sensing methods, laser profile, tracheid-effect, color, and X-ray, the software that builds a defect map, how hardwood grading rules and softwood machine-grading policy apply, and what published research prototypes measured, along with the handling and controls the scanner depends on. In the build chain, design → engineering → parts machining → fabrication → assembly → weld fatigue → stress relief → drives → controls → tuning → monitoring, the scanner sits in controls and monitoring, but its results depend on the conveyors, drives, and frames upstream.

What does a lumber scanner measure, and what does a grading system do with it?

Scanning and grading are two steps. In the Forest Service's nondestructive-evaluation report, Thomas writes that "the key components of a laser-scanning system are a laser line generator and a camera," and that laser scanning has become an accepted and economical means of determining the size, shape, and features of logs and lumber. Grading is the second step. In the multiple-sensor prototype that Kline, Surak, and Araman tested at Virginia Tech, the final output of the machine vision system is a "defect map" that includes the size, location, and type of every defect, and a separate hardwood grading program, UGRS, then graded each board from that digital map.

Two cautions frame the rest of this article. The performance figures below come from published research on hardwood, including research prototypes, not from specifications of commercial scanners. Grading rules also differ by product: hardwood appearance grades follow the National Hardwood Lumber Association (NHLA) rules, while softwood grading rules are certified under the American Softwood Lumber Standard, PS 20-25, which provides for the grading of structural lumber by both visual and mechanical means. The sorting step that follows is covered in the sawmill material flow article (Ross 2015, FPL-GTR-238, Ch. 9, p. 103; Kline, Surak, Araman 2003, Computers and Electronics in Agriculture 41, §2.2 and §2.4, pp. 143 and 148–149; NIST PS 20-25, Preface).

How does laser profile scanning measure thickness, wane, and voids?

Laser scanning measures distance by triangulation. Thomas describes a camera and laser separated by a measured distance, with the camera aimed at the laser line at a specific angle; from the camera angle, the camera-to-laser distance, and triangulation, the distance of points along the line is found. He adds that, in many instances, scanning systems are calibrated using an object of known shape and size. Some scanners have multiple laser beams and one camera, others two cameras and one laser, and "using two cameras avoids areas of missing data" where a protrusion hides the laser from a camera. Visible-spectrum scanners must operate out of direct sunlight.

On lumber, Thomas says profile scanning yields accurate measurements of wane and board width, and calls the measurement and automated detection of wane and void on lumber edges "perhaps the largest application of laser scanning." The chapter's diagram of a lumber-mill scanner shows a typical configuration with two lasers: a shallow-angle laser for profile, detecting void and wane on the edges as well as splits, cracks, and holes, and a laser perpendicular to the board for the tracheid effect.

Two research prototypes give figures:

• Lee and co-authors (2003) angled the profile laser's plane of light at about 45 degrees to the board surface and measured thickness at 1/16 in spacing along both width and length. They report wane boundaries detected with 3/16 in (5 mm) error on average, and name residual bark, debris, and laser intensity variation as error sources on unplaned boards.
• Kline and co-authors (2003) state that their laser-based ranging system can measure board thickness variations to within 0.4 mm (1/64 in).

The triangulation geometry and its occlusion limits are covered in the 2D vs. 3D vision article (Ross 2015, FPL-GTR-238, Ch. 9, pp. 103 and 106; Lee, Abbott, Araman, Schmoldt 2003, Proc. ScanTech 2003, pp. 49–50 and 54; Kline, Surak, Araman 2003, §2.1.2, pp. 141–142).

What is the tracheid effect, and what does it reveal about grain?

The tracheid effect is light traveling inside the wood. Thomas writes that when a laser beam strikes the wood surface, the beam is propagated along the fibers, and "the angle of glow of the laser beam of the wood surface shows the angle of the wood grain." The grain angle around knots and pitch pockets can then be used as an indicator of defects, a finding the chapter attributes to Jolma and Mäkyen (2007), as it spells the citation. The chapter adds that the tracheid effect can be used to detect sloped grain and potential weakness, citing Olsson and co-authors (2013), who correlated tracheid imagery with stiffness.

Hardwoods lack tracheids. Lee and co-authors note that while hardwoods do not contain tracheids, they do contain vessels that are smaller and fewer in number but produce a similar effect. Their prototype used two side-mounted lasers for this reflectance image, and one approach they describe is to sum pixel intensities perpendicular to the laser line, excluding the central line itself, so that increased scattering marks a possible defect. As engineering reasoning, the measurement depends on the laser striking clean wood at a known angle, so surface water, ice, or debris on the board is part of the measurement problem, not just the handling problem (Ross 2015, FPL-GTR-238, Ch. 9, p. 106; Lee, Abbott, Araman, Schmoldt 2003, pp. 50–51).

How do color cameras and X-ray find defects the laser misses?

In the Kline prototype, each sensor fed a different class of defect. The prototype combined a color camera system, a laser-based ranging system, and an X-ray scanner:

• Color. An RGB line-scan camera with 864 pixels, mounted perpendicular to the wood surface, with four fiber-optic light lines fed by a DC-regulated tungsten-halogen source. Light balance was calibrated on a 75% reflectance target, and each color channel was shade-corrected with a linear function.
• X-ray. A source set to 100 keV and 0.6 mA, with a 256-pixel line-array sensor; image contrast was calibrated with a uniform-density polyethylene target.
• Registration. The color, range, and X-ray images were calibrated to identical spatial resolution, so a pixel on any image refers to the same location on the board, at 30 pixels per inch across the board and 16 pixels per inch along it at a conveyor speed of 2 ft/s (0.6 m/s).

The output of the laser-range operations is geometry-related defect classes, the X-ray output density-related classes, and the color output surface defect classes. Regalado, Kline, and Araman (1992) give the matching limits: laser systems that give only wane or geometry information had been used industrially for many years, and a camera-based computer vision system "can locate defect areas of a board but has difficulty in identifying the type of defect." In their grouping, knot and decay defects are detectable by scanners based on wood density differentiation, such as X-ray systems. Lighting and camera choices for line-scan stations in general are covered in the companion lighting and camera selection article in this category (Kline, Surak, Araman 2003, §2.1, §2.1.1, §2.1.3 and Fig. 2, pp. 141–143; Regalado, Kline, Araman 1992, Forest Products Journal 42-3, p. 30).

How much is each kind of defect information worth?

Regalado, Kline, and Araman tested this directly. On 120 unedged, untrimmed red oak boards from three hardwood mills, they ran edging and trimming optimization with only some defect types as input and compared the value with optimization on complete defect data. Their conclusions:

• Wane information alone gave an average value recovery of approximately 81% of the optimum.
• Adding splits, shake, checks, and holes to the wane data did not significantly increase value recovery.
• Wane, knots, and decay gave approximately 88% of the optimum.
• Wane plus the location and size of all other defects, without the defect type, gave an average of 95% of the optimum.
• Value recovery increased with the number of defect types included, and optimization could exceed actual sawmill output even without all defects.

The same paper summarizes the authors' earlier study, its reference 5 in Forest Products Journal 42(2), in which the actual lumber value from manually operated edging and trimming in the three mills ranged from 62 to 78 percent of optimum. Lee and co-authors' 2003 introduction cites that same earlier paper for lumber values "only 65% of optimum." Each figure is given here as the citing paper states it. The unscrambler and lug loader article covers why that per-board optimization needs singulated boards (Regalado, Kline, Araman 1992, pp. 29 and 34; Lee, Abbott, Araman, Schmoldt 2003, p. 49).

How does software turn scanner images into a defect map?

The two prototypes used different software, and both are research systems. The Kline system worked in stages:

1. Preprocessing: background extraction to find the board edge, histogram extraction, and image registration.
2. Segmentation: multiple thresholds per image, chosen at the inflection points between histogram peaks, then connected-component labeling into candidate regions.
3. Classification: fuzzy-logic rules applied to each region's properties. A region takes the defect class with the largest similarity measure if it exceeds a threshold, for example 0.5; otherwise it is clear wood.
4. Training: parameters came from 300 dry, surfaced red oak samples covering ten defect types.

The Lee system first found the background, wane, and voids from the profile image, then applied a modular artificial neural network to the reflectance image, examining only "suspicious" regions darker than clear wood, in 7 × 7 pixel windows. One network identified clear wood and a second identified knots and decay. In their 10-fold cross-validation, the modular network classified clear wood, knots, and decay with 96.7% accuracy, against 92.7% and 91.2% for the two single networks. A post-processing step corrected manufacturing marks, which the authors describe as discolorations caused by mineral oxidation or by metal deposition from conveyors or side chains. How rule-based and trained models compare in general inspection is covered in Rule-Based vs. Deep-Learning Inspection (Kline, Surak, Araman 2003, §2.2–§2.2.4, pp. 143–148; Lee, Abbott, Araman, Schmoldt 2003, pp. 51–53 and Table 2, p. 55).

How are hardwood grades assigned from a defect map?

Hardwood appearance grades measure how much clear material a board yields. The Forest Products Laboratory's Wood Handbook states that the NHLA rules "are considered standard in grading hardwood lumber intended for cutting into smaller pieces," and that the grade is determined by the proportion of a piece that can be cut into a certain number of smaller pieces, "commonly called cuttings." The grades run FAS, F1F, Selects, No. 1 Common, No. 2A Common, No. 2B Common, No. 3A Common, No. 3B Common, and Sound Wormy, and except for F1F and Selects, the poorer side of a piece is inspected. The handbook's Table 6-1 lists minimum clear-face cutting percentages that vary with surface measure and the number of cuttings allowed; the lowest figures it lists are:

Grade	Lowest figure in Table 6-1's clear-face cuttings column
FAS	83-1/3%
No. 1 Common	66-2/3%
No. 2 Common	50%
No. 3A Common	33-1/3%; a footnote also admits pieces not below No. 2 Common on the good face with sound cuttings on the reverse
No. 3B Common	25%; a footnote states the cuttings must be sound and clear face is not required

The handbook warns that its summary "should not be regarded as a complete set of grading rules." NHLA describes itself as the official governing body for developing, maintaining, and interpreting the hardwood lumber grading rules in North America, and says it reviews the rules every four years. The grading software in the Kline study, UGRS, is described by its authors as "an advanced computer program for grading and remanufacturing lumber" that grades lumber according to NHLA rules and displays boards "digitally described in a computer file," with a grading module based on the 1998 NHLA rules. Kline and co-authors note that UGRS employs "a strict and literal interpretation" of the rules (USDA Forest Products Laboratory 2021, FPL-GTR-282, Ch. 6, pp. 6-2 to 6-3 and Table 6-1; NHLA, Rules for the Measurement & Inspection of Hardwood & Cypress, 2023; Moody, Gatchell, Walker, Klinkhachorn 1998, Forest Products Journal 48-9, pp. 45–46; Kline, Surak, Araman 2003, §3.3, p. 151).

How accurate were automated graders in research trials?

The Kline study measured accuracy against a ground-truth grade. Eighty-nine boards of kiln-dried, re-surfaced 4/4 red oak were graded four ways: by the automated system, by manual digitizing of every defect graded with UGRS (taken as ground truth), by an NHLA-employed certified grader, and by the original mill line graders. The results:

• Board by board: the automated grade was correct for 56 boards (63%), against 43 boards (48%) for the line graders. The abstract's "31% more accurate" is the relative comparison of those two rates, not 31 percentage points.
• Value of the 89 boards: line grade $310, NHLA certified grade $259, digitized grade $247, automated grade $244. The automated value was within 6% of the NHLA value, which the authors note is greater than the 4% money value allowance required by the NHLA grading specification; the digitized grade differed from NHLA by 4.6%.
• Ground truth itself: eight NHLA-graded boards were later reassigned a different grade by the same inspector, chiefly because of more accurate consideration of crook.

Lee and co-authors report a scanning speed of approximately 2 ft per second for their prototype, with classification and grading taking up to 20 seconds per board depending on the number of defects, and note that varying the scan speed requires reconfiguring the camera, which affects image quality. Their statement that each board will have a consistent computer grade with less variation than human graders is the authors' expectation, not a measured result (Kline, Surak, Araman 2003, §2.3–§2.4, §3.2–§3.4 and Table 2, pp. 139 and 148–153; Lee, Abbott, Araman, Schmoldt 2003, pp. 54 and 56).

What causes automated grading errors?

Kline and co-authors report that most automated grading discrepancies resulted from board geometry issues, and that the most significant error observed was in defect recognition:

• Crook (sidebend). Ignoring sidebend tends to increase calculated cutting units, so an automated system that calculates geometry precisely tends to downgrade such boards compared with human graders.
• Surface measure rounding. Surface measure is rounded to the nearest whole square foot, and a small difference in measured width can move a board across a grade line.
• Borderline cutting percentages. Their example is a board with 65% clear cutting units, which grades No. 2 Common although, in their words, No. 1 Common requires 67%; the Wood Handbook's Table 6-1 above lists 66-2/3%.
• Discoloration. The most significant error was misclassifying certain stain and mineral features as knots. In their example, a planer burn mark was falsely detected as a set of knots, which the authors say was not unexpected because the system was not trained to classify burn marks or other innocuous surface discolorations. The authors state that proper training "will require not only examples of all possible grading defects," but also examples of wood features that are not grading defects.

They add that innocuous surface discolorations "can arise from material handling or from the natural variation in wood," and state that such misclassification errors will be an even greater problem for rough-green lumber, where black sawmarks and partially dried surfaces pose a significant challenge. Lee and co-authors likewise had to filter marks from conveyors and side chains. As engineering reasoning, handling design is part of scanner accuracy: chain, roll, and guide contact that marks the board surface ahead of the scanner adds false defects for the software to reject (Kline, Surak, Araman 2003, §3.3.1–§3.3.4, §3.5 and §4, pp. 151–154; Lee, Abbott, Araman, Schmoldt 2003, p. 53).

How does softwood machine grading differ from appearance scanning?

Machine grading of softwood structural lumber sorts by mechanical properties; the Wood Handbook states that "the most common method of sorting machine-graded lumber is modulus of elasticity E." NIST's PS 20-25, the American Softwood Lumber Standard, provides for grading structural lumber "by both visual and mechanical means," and states that grading "cannot be considered an exact science," with grading rule provisions explicit enough "to establish 5 percent below grade as an allowable variation between qualified graders." Under 6.1.7, grading rules may include machine-graded lumber "as an adjunct to visual grading," in compliance with the applicable standard approved under the American Lumber Standard Committee (ALSC) Machine Graded Lumber Policy, and when graded by mechanical means, all grading equipment and methods are subject to approval by the Board of Review. The standard's text contains no provision on optical scanning.

The Wood Handbook defines machine-graded lumber as lumber evaluated by a machine using a nondestructive test followed by visual grading of characteristics the machine cannot or may not properly evaluate; machine-stress-rated (MSR) and machine-evaluated lumber (MEL) are the two North American types, MSR grades are assigned a COV of 11% on modulus of elasticity (E), and MEL grades a COV of 15% or less. The ALSC policy requires, among other things:

• Machine approval: evidence that the machine can measure the property it uses, including measurement accuracy relative to an accepted consensus standard and the maker's operational limits for temperature, speed, humidity, and lumber condition.
• Agency inspection: at least 12 inspections a year, at approximately monthly intervals, of the visual grading accuracy of machine-graded lumber at each mill.
• Calibration: in-plant test equipment calibrated at least monthly by mill personnel, with NIST-traceable third-party calibration at least annually.
• Qualification: for MSR, 95% of pieces with edge modulus of elasticity greater than 82% of the assigned average E (75% for MEL).

The Western Wood Products Association states that it is approved to supervise MSR lumber. As engineering reasoning, the ALSC operational-limit requirement means conveyor speed and board condition at the grading machine are part of its approved operating envelope (NIST PS 20-25, Preface, 6.1.1 and 6.1.7; USDA Forest Products Laboratory 2021, FPL-GTR-282, Ch. 7, p. 7-7; ALSC Machine Graded Lumber Policy 2019, sections B-2, C-3, C-7 and D-2; WWPA, Western Lumber Grading Rules, 2025).

Where does a scanner sit in the sawmill handling line and the build chain?

As engineering reasoning, a scanner's results depend on how each board is presented to it. The research prototypes show what their presentation delivered: the Lee system moved boards with pinch rollers under a downward-looking camera with a 16 in wide field of view, at 1/16 in per pixel, and the Kline system ran at a linear conveyor speed of 2 ft/s with all sensor channels referenced to the same board location. Downstream, Oklahoma State University's extension fact sheet on softwood sawmilling states that, on drop sorters, a photocell system determines which lumber is dropped in which bin.

Mapped onto the build chain, the requirements below are engineering reasoning:

• Design and engineering: a constant, known board speed through the scan zone, a defined board datum, and space for sensors on both faces and both edges.
• Parts machining: roll, guide, and sensor-mount accuracy that keeps each board at the scanner's calibrated distance.
• Fabrication, weld fatigue, and stress relief: a scan frame that stays straight and does not transmit chain or drive vibration to the sensors.
• Drives and controls: a speed-controlled drive and an encoder that tie each scan line to board position.
• Tuning and monitoring: recalibration against a known reference and trending of reject rates by defect class.

UTEC Industrial fabricates, stress-relieves with automated vibratory stress relief, and machines the frames and conveyors of the handling systems it builds. Mill-level flow and throughput are covered in the sawmill material flow and lugs per minute articles (Lee, Abbott, Araman, Schmoldt 2003, p. 50; Kline, Surak, Araman 2003, §2.1, p. 141; Hiziroglu 2017, FAPC-148, Sorting and Planing).

What sensing, PLC control, and interlocks run a scanner-to-sorter line?

The scanner's decision has to reach the right board at the next machine. The pieces are:

• Synchronized acquisition. In the Kline prototype, image collection from each scanner was synchronized using the same clock, and the processing hardware transferred six channels of spatially registered image data. As engineering reasoning, on a production line the equivalent is an encoder on the conveyor that clocks every sensor.
• Board tracking. As engineering reasoning, the PLC assigns each scanned board an identity and follows it, by encoder count or lug number, to the edger, trimmer, or sorter that executes the decision; a lost or doubled board shifts every decision behind it.
• Sorter sensing. On drop sorters, the extension fact sheet states that a photocell system determines which bin each piece drops into and that each bin's capacity is also monitored by a photocell.
• Laser safety. IEC 60825-1:2014 is applicable to the safety of laser products emitting laser radiation in the wavelength range 180 nm to 1 mm, and, as engineering reasoning, the laser sources in these scanners fall within that range; the Kline prototype's ranging laser was a 632.8 nm helium-neon laser.
• Task timing. Logix 5000 controller tasks can be configured as continuous, periodic, or event, and a periodic task performs a function at a specific time interval.

Machine stops, interlocks, and restart protection on the trim line are covered in the unscrambler and lug loader article, and mill-wide controls in Material Handling in Lumber and Wood-Products Mills. UTEC Industrial builds UL 508A control panels and programs Allen-Bradley ControlLogix and CompactLogix controllers for the handling systems it builds (Kline, Surak, Araman 2003, §2.1.2 and §2.1.4, pp. 141 and 143; Hiziroglu 2017, FAPC-148, Sorting and Planing; IEC 60825-1:2014; Rockwell Automation 1756-RM094N-EN-P-2025, Ch. 5 pp. 39 and 41).

What should a mill define before specifying a scanner or grading system?

The research results point to the items a specification has to settle. As engineering practice, drawing on the findings above:

• Rule set and product: hardwood appearance grades under NHLA rules, or softwood structural grades under PS 20-25 and an ALSC-accredited agency, with any machine grading approved under the ALSC policy.
• Board condition at the scanner: rough-green or dry-surfaced. Kline and co-authors expect misclassification to be an even greater problem on rough-green lumber, and Lee and co-authors designed their prototype specifically for unplaned boards in the green state.
• Defect information needed: the Regalado results show value recovery rising from about 81% of optimum with wane alone to about 95% with the location and size of all defects.
• Training and acceptance set: examples of every grading defect and of the non-defect marks the mill produces, including handling marks, with a ground-truth method agreed in advance.
• Speed and presentation: board speed, spacing, and orientation through the scanner, and the encoder and tracking that carry each decision downstream.

UTEC Industrial performs factory acceptance testing and on-site commissioning; as engineering practice, the handling and tracking functions around a scanner can be proved at the factory acceptance test, before the line runs production (Kline, Surak, Araman 2003, §3.3.4, §3.5 and §4; Lee, Abbott, Araman, Schmoldt 2003, p. 50; Regalado, Kline, Araman 1992, p. 34; NIST PS 20-25, 6.1.7; ALSC Machine Graded Lumber Policy 2019, section B).

Related Articles

• Material Handling in Lumber and Wood-Products Mills — lumber mill operations where grading happens
• Rule-Based vs. Deep-Learning Inspection: When AI Wins — deep learning versus rules in defect grading
• Unscramblers, Even-Enders, and Lug Loaders Explained — singulating boards for the scanner and the trim-line controls around it
• Machine Vision Lighting and Camera Selection Fundamentals — lighting and camera choices behind scanner accuracy
• Lugs per Minute and Pieces per Minute: Measuring Sawmill Throughput — the throughput measures scanners feed

References

• Ross, R. J., ed. Nondestructive Evaluation of Wood, 2nd ed. General Technical Report FPL-GTR-238. USDA Forest Service, Forest Products Laboratory, 2015.
• Kline, D. E., C. Surak, and P. A. Araman (2003). "Automated hardwood lumber grading utilizing a multiple sensor machine vision technology." Computers and Electronics in Agriculture, 41, 139-155.
• Lee, S.-M., A. L. Abbott, P. A. Araman, and D. L. Schmoldt (2003). "A Prototype Scanning System for Optimal Edging and Trimming of Rough Hardwood Lumber." Proceedings of ScanTech 2003, Wood Machining Institute, pp. 49-58.
• Regalado C, Kline DE, Araman PA (1992). "Value of defect information in automated hardwood edger and trimmer systems." Forest Products Journal, 42(3), 29-34.
• USDA Forest Products Laboratory. Wood Handbook: Wood as an Engineering Material, FPL-GTR-282. USDA Forest Service, 2021.
• NHLA: Rules for the Measurement & Inspection of Hardwood & Cypress. National Hardwood Lumber Association, 2023.
• Moody, J., C. J. Gatchell, E. S. Walker, and P. Klinkhachorn (1998). "An introduction to UGRS: the ultimate grading and remanufacturing system." Forest Products Journal, 48(9), 45-50.
• NIST PS 20-25: American Softwood Lumber Standard. National Institute of Standards and Technology, U.S. Department of Commerce, 2025.
• ALSC: Machine Graded Lumber Policy. American Lumber Standard Committee, 2019.
• WWPA: Western Lumber Grading Rules. Western Wood Products Association, 2025.
• Hiziroglu, S. Basics of Softwood Sawmilling, Fact Sheet FAPC-148. Oklahoma State University Extension, 2017.
• IEC 60825-1:2014: Safety of laser products — Part 1: Equipment classification and requirements. International Electrotechnical Commission, 2014.
• Rockwell Automation 1756-RM094N-EN-P-2025: Logix 5000 Controllers Design Considerations. Rockwell Automation, 2025.

---

Ready to Discuss a Material Handling System?

UTEC Industrial designs, engineers, machines, fabricates, and installs custom material handling systems for heavy industry, from the stress-relieved structure and drives to the Allen-Bradley PLC controls, tuning, and monitoring that run them, at its Spokane Valley, WA facility. Send UTEC the application, loads, and duty cycle to start a system review.

Request a Quote →

Questions? Call (509) 922-1832 or email sales@utec.co