Harness Voltage Drop Is a Geometry Problem

The number everybody computes

Voltage drop on a DC feed is not difficult. Resistance per unit length comes from the conductor gauge, you double the length because current has to return, and Ohm's law does the rest:

drop_V = current_A × resistance_per_m × length_m × 2

Every term on the right is easy to look up except one. Gauge is a design decision. Current is a requirement. Length is geometry, and geometry is the term that models routinely get wrong, not by a little, and always in the same direction.

Straight lines are not wire

If a model stores a connection as a pair of endpoints, the only length available is the straight-line distance between them. No harness has ever taken that path. Wire follows structure: around the inside of a shell, along a chassis rail, through a grommet, down the side that has the tie-down points rather than the side that is shorter.

The straight-line figure is therefore not an estimate of the run. It is a strict lower bound on it, and a bound that gets looser exactly when the routing gets constrained, which is to say, on the runs that matter.

The fix is unglamorous: store the route. A connection carries an ordered list of waypoints in world coordinates, and its length is the polyline through its endpoints and those waypoints, plus a stated slack for service loops and strain relief.

length = Σ |pᵢ₊₁ − pᵢ|  over the routed polyline
       + slack_m

The point is not that the polyline is sophisticated. It is that it is recorded, so the length is a measurement of a decision somebody made rather than a consequence of the model's storage format.

A worked case

A roof-mounted sensor pod. The compute module draws 22 A from a 12 V distribution unit, and the interface control document allows a 5 % drop.

Stored as endpoints, the pod's eleven runs measure 7.73 m of harness. Routed the way the loom would actually be built, off each port on a stub, down into the raceways along the baseplate, around the cold plate and the compute module rather than through them, the same eleven runs measure 14.46 m. Not one part moved. The only thing that changed is that the path is now written down.

The interesting part is which limit that breaks. No single run exceeds the 5 % drop allowance either way: the worst is 1.23 % on the straight lines and 2.37 % routed, and both pass. The constraint that fails is the one nobody was watching. Six and a half extra metres of copper dissipate 4.16 W instead of 3.62 W, and that is enough to take the autonomy budget past its allocation.

Length used	Harness loss	Autonomy budget, 320 W allocated
Straight lines, 7.73 m	3.62 W	319.62 W, passes by 0.38 W
Routed polylines, 14.46 m	4.16 W	320.16 W, fails by 0.16 W

The part that surprises people

The drop is only half of what that routing decision costs. The same resistance that produces the voltage drop is also dissipating power, and it has to come from somewhere upstream:

loss_W = current_A² × resistance_per_m × length_m × 2

Across the pod that is 4.16 W, and the compute feed dominates it: loss goes as the square of current, and at 22 A that one run carries an order of magnitude more current than anything else in the shell.

Now the power budget. The pod's components draw 316 W in autonomy mode against a 320 W allocation. That passes with 4 W to spare. Add the routed harness and the source has to supply 320.16 W, which does not.

A power budget that counted only loads and not conductors would have cleared this design by 4 W. Counting the conductors, it fails by 0.16 W. That margin is uncomfortably thin, and it is supposed to be: it is the honest distance between a design that closes and one that does not, and reading it off the straight lines would have hidden the whole of it.

Why this belongs in the geometry model

The reason to compute drop where the parts live, rather than in a separate electrical tool, is that every input is already there and every output feeds something else that is already there.

• Route length falls out of the same coordinates that drive clearance.
• Copper mass falls out of gauge and length, and lands in the mass roll-up and the centre of gravity.
• Dissipated watts land in the power budget and in the thermal screen, as heat at a location.
• Moving a part re-routes its feeds, so the drop updates while you drag it.

Split those across tools and each one is individually correct while the system is wrong. Keep them in one model and moving the power distribution unit 200 mm visibly changes the drop, the loss, the budget and the CG at once, which is the actual trade being made.

What to take from it

If you record only the endpoints of a connection, you have not modelled the harness; you have modelled a wish about the harness. The gap between the two is where the drop lives.

And when a limit is exceeded, the finding is worth nothing unless it says by how much and over what. “Voltage drop exceeded” sends someone hunting. “Lidar power drops 2.37 % of 12 V against a 5 % limit, over 1.94 m of AWG 20 at 2.2 A” tells them the run, the path length, the gauge, the current and the margin, which is enough to decide between a heavier conductor and a shorter route without opening anything else.