How a grade is decided
from raw x/y/z vibration to a Good / Fair / Poor / Severe gradeThe grade isn’t a guess — it’s a short chain of arithmetic on the vibration signal. Pick a segment and follow it through: the three-axis signal, the RMS energy, normalisation for speed, and the threshold that assigns the grade. Grounded in the response-based road-roughness literature (mean-square acceleration ÷ speed on the unsprung mass; ISO 8608 · IRI). The signal is illustrative until client NVH data arrives.
The signal
Three acceleration axes, one shared scale
Same amplitude scale on all three. On a rough segment the vertical (Z) trace fills the frame while X and Y stay small — that is why roughness is read from Z, normalised for speed below.
From signal to grade
Vibration energy → speed-normalised → grade
How hard the truck is shaking (vertical axis).
Divide out speed, so a rough grade means the road — not a fast driver.
Which band the normalised value lands in.
The decision · thresholds
Every segment's speed-normalised RMS against the grade thresholds — click a bar
Feature space · why normalise
Raw RMS vs speed — why normalisation is needed; the diagonals are the grade boundaries; click a point
Raw RMS alone can’t grade a road — the same shaking means different things at 90 km/h vs 45 km/h. The diagonal lines are constant speed-normalised RMS; they rotate the confusion into clean grade regions. Marker shape + colour + region all carry the grade; the selected segment is ringed.
What kind of road? · surface type
Frequency content of the selected segment — the route to classifying surface type
Severity (Good→Severe) is how much the truck shakes. Surface type (asphalt / concrete / gravel / muddy) is what shape the shaking has — a concrete-joint spike, gravel’s broadband hiss, a smooth-but-fast road. That needs frequency-domain features and labelled examples per surface — an open item for the workshop.