EBI Electric Optimized Grading & Scanning Systems

The grading problem, in the industry's own numbers

Manual grading is a judgment call made at line speed, and the published record on how well it goes is uncomfortable. Reviewing the studies, researchers at Purdue University report that the average accuracy of a human lumber inspector runs as low as 48 percent and as high as 75 percent, with production line speed, the mental arithmetic of calculating cuttings, working conditions and fatigue from repetitive work all pulling it down. Set that against the NHLA Sales Code, which calls for a minimum of 80 percent accuracy, and the gap is the reason scanning exists at all.

Worth being precise about what this does and does not say. It is not a claim that your graders are poor. It is a claim that a person doing subjective classification for eight hours performs like a person, and that a mill's tally reflects that. An earlier Forest Service evaluation found company line graders overestimating lumber value by close to 20 percent against an NHLA-certified inspection, while the automated system came within 5 percent. Overestimating is the expensive direction, because you sell against a tally that was never there.

How a multi-sensor scanner actually reads a board

Scanning gets described as cameras, which undersells it and hides why the good systems cost what they do. Four sensor types do different jobs, and the value is in the overlap.

Color cameras are the primary sensor. Thresholds separate clear wood from knots, stains and other defects by contrast. They also faithfully report a bootprint, a dirt smudge or a conveyor mark, which is the whole problem with color alone.

Lasers with scatter cameras solve that. Laser light refracts against the wood cells, and the shape of the scatter shows which way the grain runs, so the system can map fibre deviation around a knot. Grain deviates around a knot. It does not deviate around a bootprint. Combining the two sensors is what lets the machine throw the bootprint away, and it is the single clearest reason multi-sensor beats vision alone.

A 3D profile camera reads the shape of the board, locates wane, and confirms cracks and holes that a straight-down camera view can misread.

X-ray maps density. A knot is denser than clear wood; a hole has almost none. Thresholds are set per defect on each sensor and the sensor data is overlapped, so false defects filter out and only the genuine ones get classified.

What the independent record shows, and whose numbers those are

The strongest public validation of hardwood scanning is a 2018 Purdue study published in Computers and Electronics in Agriculture. It ran 9,454 boards, 39,894 board feet, across nine commercially important hardwood species through a Microtec Goldeneye 300 multi-sensor scanner at 980 lineal feet per minute, with results verified by an NHLA-trained inspector. Across the whole volume the system came out 99.50 percent on-value and 92.22 percent on-grade.

The two figures are different measurements and the difference is the interesting part. On-grade asks whether the board landed in the correct NHLA bucket. On-value asks whether the package came out worth what it should. A scanner can put a board one grade off and still price the package correctly, and value is what you are paid on. The on-grade figure also cleared the NHLA Sales Code's 80 percent minimum, which the published human-grader range does not.

That study tested Microtec equipment, not EBI's. We are citing it because it is the only independent, peer-reviewed validation of this technology at production scale that anyone can go and read, and because it describes what the sensor class does rather than what any one vendor says about itself. EBI does not publish comparable third-party validation. Very few scanner builders do.

The study is equally useful for its limits. It could not calibrate satisfactorily for black walnut, because in walnut the clear wood and the large knots are close in both color and density, which takes away the two strongest signals at the same time. Severe iron stain in white oak needed the grading solution adjusted, and the authors note that some mills see iron stain, some never do, and some see it seasonally. If you run walnut, or you have an iron stain problem, that belongs in the scoping conversation.

One sizing note from the same work, because it answers the question buyers ask first: the authors calculated 980 lineal feet per minute as sufficient for a sawmill producing roughly 20 to 25 million board feet a year, and found that where throughput or material flow calls for a transverse-feed machine instead, the transverse equivalent produced the same results. Feed direction is a plant layout decision, not an accuracy one.

The Inspector family and where each one sits

EBI codes the range by position in the line. Inspector B is the linear machine ahead of the side matcher, running rip, cross-cutting and NHLA grade optimization, scanning 1, 2 or 4 faces with color or 3D cameras. Inspector A takes the position after the side matcher, and Inspector T is the transversal machine. Inspector E handles end-board defect detection with high-frequency inspection, and Inspector F is the finishing scanner, reading stain, varnish and color on a finished surface. Inspector P covers profile and board precision tracking, Inspector R reads I.D. for product conformity, and Inspector S detects species, which is how you keep one species out of another's production lot.

Across the family the optimization modes are cross-cutting, rip, NHLA grade and sorting, and EBI lists its current development work as rot detection, heat traces, color variation control and continuous training of the classification models.

EBI Electric is a third-generation family business, in industrial electrical products since 1946 and in wood-industry scanning since 2005. They also build kiln duty motors and thousands of control panels a year, which is worth knowing if the project needs more than a scanner. Their optimizers already run inline with Lico equipment in plants we serve, so a scanning retrofit or a fully scanned line is quoted as a single integrated project.