Machine grading: indicating properties, settings and yields

The idea

A grading machine measures indicating properties on every piece at line speed and predicts strength. Automated lines can run at up to 200 m/min (Polish paper). Machine grading has a visual override in EN 14081-2.

Dynamic MOE

The classical machine measures dynamic modulus of elasticity, computed from the longitudinal resonance frequency and the density of the piece. Stress waves obey v = sqrt(E / rho), so E = rho x v^2. In the resonance method, v = 2 x f x L, where f is the fundamental frequency and L the length. The 2023 review notes that resonance is more repeatable than time-of-flight. The Brookhuis MTG used in the Polish studies uses the natural frequency after a short impact.

Worked example: dynamic MOE of one board

A board is 4.0 m long, its fundamental frequency is 600 Hz and its density, from weight and dimensions, is 480 kg/m3 (illustrative values).

1. Velocity: v = 2 x 600 x 4.0 = 4800 m/s.
2. Dynamic MOE: E = 480 x 4800^2 = 1.106e10 Pa, about 11.1 GPa.

The machine compares this value against thresholds in the approved setting.

How good is the prediction?

Open data show the ceiling of simple resonance grading:

• Gradewood (Europe): the best machine model had R2 of 0.63 with bending strength for spruce and 0.67 for pine; laboratory frequency-based MOE alone gave 0.51 for spruce.
• Olsson and Oscarsson (2017, 936 Norway spruce pieces from Sweden, Norway and Finland): R2 of 0.53 for dynamic MOE.

Roughly half the variation in strength is explained; settings must absorb the rest.

Multi-sensor machines

A knot can sharply reduce local strength while hardly changing a board's average stiffness, which is why MOE alone is imperfect. Machines add knot and fibre information, for example laser-based grain-angle measurement (tracheid effect) or X-ray density. In Olsson and Oscarsson, laser scanning plus dynamic excitation raised R2 from 0.53 to 0.69 to 0.70. The yield gain appeared in higher classes: for a C40/C24 double grade the C40 share was 14.5% with dynamic MOE, 30.3% with the new method and 47.7% for a hypothetical perfect machine. The single-class C24 yield was already 99.5% for all indicators.

Worked example: what the extra sensors are worth

Per 1000 boards: dynamic MOE gives 1000 x 0.145 = 145 boards in C40; the multi-sensor method 1000 x 0.303 = 303. That is 158 more boards, a factor of 0.303 / 0.145 = 2.09. Sensor investment pays mainly if you want to sell high classes.

Settings: machine control and output control

A setting combines a machine, a species group, a size range and a set of classes with threshold values for the indicating property. The European system recognises two approaches:

• Machine control: extensive initial testing, fixed settings. Common in Europe, with settings approved by a CEN task group.
• Output control: less initial testing, but stricter factory production control. Less common.

Settings were derived using the cost matrix method. Not every permitted setting has a viable yield, and yields are confidential.

Visual versus machine: the numbers

Stapel's thesis (Norway spruce, Central and Eastern European origins) compiles class shares:

Share at C24 or better: DIN 4074-1 50 + 12 = 62%; machine dynamic MOE 76 + 16 = 92%. The machine puts 30 percentage points more material in C24 or higher, 300 boards per 1000.

Yields are sample-specific, and a higher yield from a weak predictor may be less safe. Adding knot information moved material to C35 (28% instead of 16%) but also raised rejects (23% instead of 8%).

Polish Scots pine (Greater Poland-Pomerania, n = 150)

Visual PN-D-94021 rejected 39.4%, machine grading 1.3%. That is about 150 x 0.394 = 59 versus 150 x 0.013 = 2 boards rejected. Source area matters too: Gradewood spruce mean bending strength ranged from 34.8 MPa (Slovakia) to 43.7 MPa (Slovenia), so a setting cannot be moved between sources unchecked.

Key points

• Machine grading uses indicating properties (dynamic MOE, plus density, knots, grain angle); resonance gives E = rho x (2fL)^2.
• Simple dynamic MOE explains roughly half of strength variation (R2 about 0.5 to 0.6); multi-sensor methods reach about 0.7.
• Extra sensors matter mostly for high classes.
• Machine control is the common European route; settings are tied to machine, species, size and classes.
• Yields and class shares are sample-specific, and a high yield is not by itself a sign of good grading.