# Measuring aero on track: coast-down, constant-velocity & logged-data estimates

> Drag from coast-downs, downforce and balance from constant-speed runs with suspension loads, and both from normal laps with the right corrections. Includes a fully worked GT3 straight-line test.

Area: Aerodynamics · Starting level: L2 Practitioner · Updated 2026-11-16
Source page: https://race-engineer-academy.vercel.app/lesson/aero-track-testing-coastdown

## L1 · Foundation

Wind tunnels and CFD give the aero map, but the engineer needs to know what the car actually does **on track**, and to check whether a new part works. Two classic straight-line tests do this with the normal data system (Segers §13.6–13.8):

**Coast-down test → drag.** Accelerate to high speed on a straight, select neutral, coast with the steering still. With no engine force, Newton's second law says the deceleration is caused by the resistive forces:

$$m_{eff}\,a = D + F_{roll} \;\Rightarrow\; D = m_{eff}\,a - F_{roll}$$

$F_{roll}$ (rolling resistance + driveline friction) is measured separately by a low-speed coast-down where drag is small (but not zero: correct for it, or fit $A + BV^2$ over the whole speed curve). Then $C_DA = D/q$.

**Constant-velocity test → downforce.** Hold a steady speed so there is no longitudinal acceleration. The suspension compresses under the downforce; from the spring deflections (or pushrod load cells) and the known spring rates you get the extra load on each axle. A low-speed run (50–60 km/h) gives the reference. It is a low-$q$ reference, not a zero-downforce one: it still carries $(V_{low}/V)^2$ of the downforce (9 % for 60 vs 200 km/h), so divide the measured load increment by $\Delta q = q - q_{low}$. Then $C_LA = (L_F + L_R)/\Delta q$ and the aero balance is $L_F/(L_F + L_R)$.

**Rules for usable results:**
- Run in **both directions** and average — this cancels wind and track gradient to first order.
- Log the weather (density!) and keep tyre pressures, fuel and temperatures consistent.
- Start coast-downs as fast as possible: drag grows with $V^2$ while rolling resistance is nearly constant, so the signal-to-noise ratio is best at high speed.
- Keep the car straight: steering and yaw add tyre drag.

**Typical result quality:** good teams get $C_DA$ repeatability of ~1–2 % and $C_LA$ of ~2–4 % — enough to see a two-click wing change, not enough to see a small bodywork detail. That is why these tests are usually done on airstrips or long straights with many repeats.

**Misconception:** "a faster speed trap proves less drag." Trap speed depends on engine power, gear ratios, exit speed of the previous corner, wind and density. Use a coast-down or a calculated drag channel instead.

## L2 · Practitioner

**Worked example — GT3 straight-line test.** Weather: 18 °C, 1009 hPa, 60 % RH → $p_{sat} \approx 2060$ Pa, $p_v \approx 1240$ Pa, $\rho \approx 1.192 + 0.009 = 1.20$ kg/m³. Car mass with driver and fuel 1310 kg; rotating inertia (wheels, brakes, gearbox output) adds ~3 %: $m_{eff} \approx 1349$ kg.

*Step 1 – rolling resistance.* Coasting in neutral from 60 to 20 km/h, mean deceleration 0.028 g: total resistance $= m_{eff}\,a = 1349\times0.028\times9.81 \approx 370$ N. Drag is not negligible here: with a roughly linear speed decay the mean $V^2$ is $(16.7^2 + 16.7\times5.6 + 5.6^2)/3 \approx 134$ m²/s², so $q \approx 80$ Pa and drag ≈ 80 N for $C_DA \approx 1$. Hence $F_{roll} \approx 290$ N. At high speed the tyres carry ~3500 N of extra aero load; with a rolling-resistance coefficient ~0.015 add ~50 N. (Better still: fit $m_{eff}\dot V = -(A + BV^2)$ to both coasts together, see L4.)

*Step 2 – coast-down.* Speed falls from 211.8 to 199.2 km/h in 2.0 s: $a = (3.50\text{ m/s})/2.0\text{ s} = 1.75$ m/s². Total resistance $= 1349\times1.75 \approx 2360$ N. Drag $D = 2360 - 290 - 50 \approx 2020$ N. Mean speed 205.5 km/h = 57.1 m/s → $q = 0.5\times1.20\times57.1^2 \approx 1956$ Pa. **$C_DA \approx 1.03$ m².**

Error check: using the static mass in both steps gives ~1.00 (−3 %); treating the 60→20 km/h coast as drag-free overstates $F_{roll}$ by 80 N and costs another ~4 %; a 0.3 % downhill grade (0.029 m/s²) would bias $C_DA$ by ~2 % — hence the two directions.

*Step 3 – constant velocity at 200 km/h* (zeroed at 60 km/h). Front springs 300 N/mm, motion ratio (spring/wheel) 0.60; rear springs 220 N/mm, MR 0.80. Average spring compression: front 3.5 mm, rear 5.0 mm. Wheel force per corner $F = MR\cdot k\cdot x$:
- Front: $0.60\times300\times3.5 = 630$ N per wheel → $\Delta L_F = 1260$ N
- Rear: $0.80\times220\times5.0 = 880$ N per wheel → $\Delta L_R = 1760$ N
- Zeroed at 60 km/h, these deflections measure $L(200) - L(60)$, so divide by $\Delta q = q_{200} - q_{60} = 1852 - 167 = 1685$ Pa → **$C_LA \approx 3020/1685 = 1.79$ m², AB ≈ 41.7 %.** Dividing by $q_{200}$ alone would understate $C_LA$ by $(60/200)^2 = 9\,\%$ (1.63 m²); the balance is unaffected to first order because both axles scale alike.

Run the opposite direction and average. If run 1 gives $C_LA = 1.93$ and run 2 gives 1.65, the car had a headwind on run 1; the average 1.79 is the estimate, and the wind can be back-calculated from the difference between the runs (Segers §13.8).

**Setup for the tests (constant velocity):** soft springs (more deflection = better resolution), no packers or bump rubbers in contact, ARBs disconnected or at least not loaded, static heights chosen so that the *target dynamic* heights are reached at the test speed. Record dynamic ride heights with every point — the result is a point on the aero map, not "the" $C_LA$.

**Calibrations to check before the day:** potentiometer zeros (a 0.3 mm zero error at the front is ~110 N per axle in this example — about 9 % of the measured front load increment), spring rates measured on a spring tester, load-cell calibration, ride-height laser offsets.

## L3 · Advanced

**Drag from normal laps (full-throttle acceleration).** Without a dedicated test, use every full-throttle straight:

$$D = \frac{T_e(n)\, i_{tot}\, \eta}{r_{dyn}} - m_{eff}\,a_x - F_{roll} - mg\sin\gamma$$

(Segers Eq. 13.17 form). Error budget: at 250 km/h a GT3 drive force is ~5000 N while drag is ~3000 N. A 4 % engine-torque error (torque map vs. actual, BoP restrictor, ambient) is 200 N → ~7 % drag error. A 1 % error in $r_{dyn}$ or gear ratio does the same. So logged-lap drag estimates are good for **trends within a session** (same engine, same day) but poor for absolute values. The coast-down avoids the torque term entirely, which is why it's preferred.

**Downforce from damper/pushrod data on laps.** For straight-line points with $|a_y| < 0.1$ g:

$$L_F = \sum_{F} MR_F k_F x_F + (1 - A_F)\frac{m\,a_x\,h_{CG}}{WB}, \qquad L_R = \sum_{R} MR_R k_R x_R - (1 - A_R)\frac{m\,a_x\,h_{CG}}{WB}$$

with $a_x > 0$ under acceleration. Acceleration transfers load rearward and *unloads* the front springs, so the transfer is added back at the front and removed at the rear (Segers Eq. 13.18–13.19). $A_F$, $A_R$ are the fractions of the transfer carried by the suspension links through anti-geometry (anti-squat at the driven rear on power, anti-dive/anti-lift under braking). That part never shows in the springs. Use $A = 0$ if unknown (it is zero at an undriven front on power), and check the signs on a braking zone where aero is small. Example: GT3 at $a_x = 0.077$ g, 1380 kg, $h_{CG} = 0.36$ m, WB 2.7 m → 139 N per axle; with the wrong sign the front error is ~280 N, several points of aero balance. Include heave/third springs and bump rubbers in the sum if engaged; they often carry most of the aero load. Pushrod load cells remove the spring-rate and friction uncertainty.

**Drag-moment convention.** At constant speed the drag acts at a height of roughly 0.4–0.6 m while the tractive force acts at the ground. This couple transfers ~$D\,h_D/WB$ from front to rear — about 300–400 N for the GT3 above. It is part of what the tyres "see", and tunnels quoting loads at the contact patch include it; tunnels restraining at the wheel centres may not. Make sure your track and tunnel balances use the same convention before claiming a correlation error of several percent.

**g–g envelope method (whole lap).** For a constant-μ tyre the maximum combined acceleration rises linearly with dynamic pressure: $a_{max} = \mu g + \mu\,q\,C_LA/m$. Plot the outer envelope of combined g vs $q$ from many laps; slope/intercept gives $C_LA/(m g)$ roughly: $C_LA \approx m g \cdot \text{slope}/\text{intercept}$. This is crude (load sensitivity bends the line, driver may not use the limit everywhere) but needs no suspension instrumentation and is excellent for comparing aero configurations over a test day.

**Wind correction properly.** With headwind $w$, run speeds $V\pm w$ give $q \propto V^2 \pm 2Vw + w^2$. Averaging the two *forces* leaves a bias of $w^2/V^2$ — for 10 km/h at 200 km/h that's 0.25 %, negligible. Averaging the two *coefficients* computed with ground speed has the same small bias. Gusts within a run, however, are not cancelled; a pitot (or multi-hole probe) measuring real dynamic pressure (Segers §13.3) is the cure.

## L4 · Expert

**Identifying an aero map from track data.** Treat the problem as regression: unknowns are map coefficients $\boldsymbol{\beta}$ (e.g. a quadratic surface in $h_F$, $h_R$ for front and rear load), measurements are axle loads from load cells at many time samples. Model per sample:

$$L_{F,i} = q_i\,\big(\beta_0 + \beta_1 h_{F,i} + \beta_2 h_{R,i} + \beta_3 h_{F,i}^2 + \dots\big) + \varepsilon_i$$

If you regress the loads $L_i$ directly with roughly constant load noise (load cells), ordinary least squares is already correct: the high-$q$ points dominate naturally. If you regress the coefficients $C_LA_{,i} = L_i/q_i$, whose noise variance is $\sigma_L^2/q_i^2$, weight each sample by $q_i^2$. Either way, exclude transients (high $|\dot h|$, kerbs, $|a_y| > 0.2$ g unless you include roll as a regressor). Check identifiability: if $h_F$ and $h_R$ always move together (no pitch variation), the separate gradients are unobservable — you need runs at different static rakes, which is exactly what Segers' advice to run at several static ride heights to populate the aero maps achieves (§13.7–13.8).

**Uncertainty propagation.** For $C_LA = L/q$: $\sigma_{C}/C = \sqrt{(\sigma_L/L)^2 + (2\sigma_V/V)^2 + (\sigma_\rho/\rho)^2}$. A 1 % speed error is a 2 % $C_LA$ error — GPS speed is better than wheel speed for this because wheel speed is corrupted by tyre growth and load-dependent rolling radius (several mm at 250 km/h on a slick). For coast-downs, differentiate speed carefully: fit a model $m_{eff}\dot V = -(A + B V^2)$ to the whole speed-time curve (closed-form solution $V(t) = \sqrt{A/B}\tan(\dots)$) instead of differentiating noisy data. $B = \tfrac12\rho C_DA$ comes out with confidence bounds.

**Systematic traps.** Tyre temperature and pressure change rolling resistance during a run; downforce raises rolling resistance; driveline drag in neutral differs from that in gear (oil churning depends on speed). Gearbox oil temperature can change "rolling resistance" by 50–100 N. Ride height varies with speed, so $C_DA$ from a coast-down is an average over a range of platforms — report it with the ride-height window.

**Correlation philosophy.** The goal is rarely the absolute number; it's the *delta* between configurations measured under the same conditions, plus a check that tunnel/CFD gradients match reality. Plan the test as an A-B-A-B sequence to remove drift (fuel burn, wind changes, track temperature), and decide beforehand what effect size you need to detect and how many repeats that requires given your measured repeatability.

## Key formulas

**Coast-down drag**

$$D = m_{eff}\,a - F_{roll} - m g \sin\gamma, \qquad C_DA = \frac{D}{\tfrac12\rho V^2}$$

- m_{eff}: mass incl. rotating inertia (kg)
- a: measured deceleration (m/s²)
- F_{roll}: rolling + driveline resistance (N)
- \gamma: track gradient (rad)

**Wheel load from spring deflection**

$$F_{wheel} = MR\cdot k_s\cdot x_s$$

- MR: motion ratio = spring travel / wheel travel
- k_s: spring rate (N/mm)
- x_s: spring (damper) compression (mm)

**Downforce coefficient from axle loads**

$$C_LA = \frac{L_F + L_R}{\tfrac12\rho V^2}, \qquad AB = \frac{L_F}{L_F + L_R}$$

- L_F: absolute front aero load (N); if loads are zeroed at a low-speed run, use ΔL/(q − q_low) instead of L/q
- L_R: absolute rear aero load (N)
- AB: aero balance (front share)

**Drag from a full-throttle run**

$$D = \frac{T_e\,i_{tot}\,\eta}{r_{dyn}} - m_{eff}a_x - F_{roll}$$

- T_e: engine torque (Nm)
- i_{tot}: overall gear ratio
- \eta: driveline efficiency
- r_{dyn}: dynamic rolling radius (m)

## Trackside takeaways

- Coast-down for drag, constant velocity for downforce and balance; always both directions.
- Measure rolling resistance with a low-speed coast-down (correct for the drag still present, or fit A + BV²); include rotating inertia (~3 %).
- A 60 km/h reference still carries ~9 % of the 200 km/h downforce: divide load increments by Δq, not q.
- F_wheel = MR·k·x with MR = spring/wheel travel; heave springs and bump rubbers must be included.
- A 0.3 mm potentiometer zero error can be ~10 % of front downforce: zero at low constant speed every run.
- Logged-lap drag estimates depend on engine torque accuracy: fine for trends, poor for absolutes.
- Every result is a point on the aero map: report it with the dynamic ride heights.

## By car category

### GT3

GT3 teams rarely get airstrip time; most aero checks are done from logged laps and occasional straight-line runs at tests. Damper-pot-based downforce estimates are workable because GT3 springs are relatively soft (deflections of several mm).

### FIA F3

Stiff springs + large tyre deflection make damper-pot downforce estimates poor; pushrod strain gauges or a tyre-deflection model are needed. Coast-downs are still valid for drag.

### LMP2

Spec chassis with a Le Mans kit and a standard kit: coast-downs at the pre-season test help confirm which wing settings give the expected drag, important for fuel and top speed planning.

### Hypercar (LMH/LMDh)

Aero performance is homologated and verified by the regulator; teams still run straight-line checks to confirm the car sits in its declared window and to correlate their own models.

### IndyCar

Speedway configurations are trimmed to very low drag; coast-down-type analysis and drag estimates from logged laps are standard at oval tests, with careful attention to wind.

### One-make Cup

In Cup cars aero adjustability is limited; logged-lap comparisons (trap speeds at the same density, g–g envelope vs speed) are usually enough to confirm a wing change works.

## References

- Segers, Analysis Techniques for Racecar Data Acquisition, 2nd ed. (SAE 2014): §13.5 Estimating Drag and Downforce from Logged Data
- Segers, Analysis Techniques for Racecar Data Acquisition, 2nd ed. (SAE 2014): §13.6 The Coast-down Test
- Segers, Analysis Techniques for Racecar Data Acquisition, 2nd ed. (SAE 2014): §13.7 The Constant Velocity Test
- Segers, Analysis Techniques for Racecar Data Acquisition, 2nd ed. (SAE 2014): §13.8 A Worked out Example of a Straight-Line Test
- Milliken & Milliken, Race Car Vehicle Dynamics (SAE 1995): Ch. 15 Applied Aerodynamics
- Other: SAE J1263 Road Load Measurement and Dynamometer Simulation Using Coastdown Techniques
