# Lap time simulation: steady-state, quasi-steady-state, transient, and correlation with data

> Simulation lets you test without the track, but only a model correlated against logged data is worth trusting. Learn the simulation levels, the vehicle–tyre–track model structure, how to correlate, and how to use sims before, during and after an event.

Area: Data Acquisition & Analysis · Starting level: L2 Practitioner · Updated 2026-11-16
Source page: https://race-engineer-academy.vercel.app/lesson/daq-simulation-correlation

## L1 · Foundation

**Why simulate?** Track time is scarce and expensive. A lap time simulation (LTS) predicts speed, accelerations, gear use and lap time for a car model on a track model. It lets you:

- arrive with a near-optimal baseline setup and a prioritised test plan,
- isolate one parameter (real changes often have side-effects – ride height changes aero *and* roll centres),
- estimate channels you can't measure (tyre loads, slip angles),
- explore beyond physical adjustment ranges,
- feed strategy (lap-time sensitivity to fuel mass).

Series organisers use LTS too: Segers describes the FIA's GT Balance of Performance using the D.A.T.A.S. RaceSim package, with models validated on car data. Current BoP processes (SRO, IMSA, WEC) combine organiser tests, sealed-logger data and simulation.

**Three ingredients:** a **vehicle model** (mass, aero, power, gearing, suspension), a **tyre model** (forces vs load and slip), and a **track model** (curvature vs distance, elevation, banking, grip factors). The track model is usually created from logged data; the vehicle+tyre model is validated against logged data.

**Golden rule:** a simulation is only as good as its inputs. Garbage in, garbage out.

## L2 · Practitioner

**Simulation levels.**

| Type | Idea | Good for | Misses |
|---|---|---|---|
| Steady-state | car is braking, accelerating or cornering at constant radius | first estimates | transitions |
| Quasi-steady-state (QSS) | track split into many small constant-radius segments; car in steady state in each | gearing, power, mass, drag/downforce trade-offs, fuel effect | yaw inertia, dampers, transient balance |
| Transient (dynamic) | multi-DOF model with inertia, dampers, tyre dynamics | damper/inertia effects, transient balance | needs many accurate parameters |

**A point-mass model is surprisingly useful.** Segers' Auto GP example: 940 kg car, 1.80 g at 75 km/h (negligible aero) gives $\mu \approx 1.8$. At 215 km/h it pulls 2.38 g, so the normal force must be $2.38/1.8$ of weight; the excess is downforce:

$$L = m g\left(\frac{G_{lat}}{\mu} - 1\right) = 940\cdot 9.81\,(1.322-1) \approx 2.97\ \text{kN}$$

**Correlation workflow.** Overlay simulated and measured speed, lateral/longitudinal g and steering (if modelled) against distance. Targets: lap time within ~0.5 %, minimum speeds within 2–3 km/h, top speeds within 1–2 km/h, braking points within a few metres. Tune physical parameters first (drag, downforce, power, tyre grip), local grip factors last.

**Parameter studies.** With a correlated model, change one parameter at a time by a fixed amount and record Δ lap time. Segers' touring-car study at Interlagos (illustrative) worsened every parameter by the same 1 %, which is what makes the ranking meaningful: grip 0.38 s; mass (13 kg) 0.17 s; power (5 hp) 0.06 s; CoG height (3.5 mm) 0.055 s; front weight, air density and $C_D A$ about 0.035 s each; $C_L A$ 0.017 s; aero balance only 0.005 s. On a GT3 Nogaro model, −2 % lateral grip cost 0.59 s while −2 % braking grip cost only 0.08 s. Such tables tell you what to prioritise at *that* track.

**Use around an event:**
- *Before:* virtual test day – baseline setup, gear ratios, test plan; discard changes the sim says are pointless.
- *During:* first check after outing 1 – does the real car match the sim? Large discrepancies point to a technical issue (e.g., roll gradient differs → bar not connected). Use the sim as a calculator for what-ifs.
- *After:* re-correlate, improve aero maps, track bump profile and tyre model; log everything.

## L3 · Advanced

**How QSS works.** From the track curvature $\kappa(s)$ compute the maximum cornering speed at each point from the g-g-V envelope:

$$V_{max}(s) = \sqrt{\frac{a_{y,max}(V)}{\kappa(s)}}$$

(iterated because grip depends on speed through downforce). Then integrate **forward** from each apex with traction-limited acceleration (power, combined grip, drag) and **backward** with braking-limited deceleration; the speed profile is the minimum of the three envelopes. Lap time is $\int ds/V$. Combined grip is usually an ellipse: $(a_x/a_{x,max})^2+(a_y/a_{y,max})^2 \le 1$.

**Track model from data.** Curvature from logged data: $\kappa = G_{lat}\,g/V^2$ (the reciprocal of radius), filtered. Add elevation (GPS or survey) and banking (from lateral g vs yaw-rate × speed mismatch). Because the curvature comes from a real driver's line, it carries that driver's line choice.

**Tyre model.** Pacejka "Magic Formula" curves are the standard representation; fitted coefficients need rig data that suppliers rarely release. A first-cut model can be derived from logged data: assume a normalised force–slip shape (peak around 5–7° for race tyres) and fit a parabolic load-sensitive peak force $F_{y,max} = k_a F_z - k_b F_z^2$ to measured cornering using estimated wheel loads; adjust $k_a$ front/rear to match measured balance.

**Pitfalls:**
- *Over-fitting with grip factors*: local grip multipliers can force any model to match, but then the model predicts nothing. Keep them few, physically motivated (surface change, dirt), and the same across setups.
- *Driver model*: QSS assumes the limit everywhere; real drivers aren't, especially in low-confidence corners. Discrepancies concentrated in one corner may be driver, not model.
- *Large predicted gains* (> a few tenths from a small change) usually indicate a model problem, not a discovery.
- *Secondary effects*: a sim change of ride height without the associated camber/aero map change is not the real change.

## L4 · Expert

**Top-level practice.** Leading teams use optimal-control lap simulation (minimum-time OCP with a multi-DOF vehicle and transient tyre model) rather than QSS: the optimiser finds the line and inputs that minimise time subject to vehicle dynamics, so line choice and transient balance are accounted for. Combined with driver-in-the-loop simulators, the toolchain becomes: offline OCP for setup space exploration → DIL for driveability and driver feedback → track for validation.

**Model validation as a process.** Treat correlation as statistical: compare distributions (speed histograms, g-g envelopes per speed band) not just one lap; run sensitivity analysis to find which uncertain inputs dominate output uncertainty (often tyre peak μ and its load/temperature sensitivity, then aero balance vs ride height). Identifiability matters: if lowering downforce and lowering tyre μ give the same lap time and speed trace, your data cannot separate them – add a straight-line aero test or ride-height-based downforce measurement.

**BoP and regulation simulation.** Organisers simulate each make with validated models to set weight, power and aero adjustments. Teams simulate the same to predict how a BoP change affects them – e.g., the lap-time effect of +15 kg ballast versus −1 % power at a given track – and to decide whether qualifying or race pace suffers more.

## Key formulas

**Downforce estimate from point-mass model**

$$L = m g\left(\frac{G_{lat}}{\mu} - 1\right)$$

- m: vehicle mass (kg)
- G_{lat}: lateral acceleration at speed (g)
- \mu: friction coefficient from a low-speed corner

**Track curvature from logged data**

$$\kappa = \frac{1}{R} = \frac{G_{lat}\, g}{V^2}$$

- \kappa: curvature (1/m)
- V: speed (m/s)

**QSS cornering speed limit**

$$V_{max}(s) = \sqrt{\frac{a_{y,max}(V)}{\kappa(s)}}$$


**Friction ellipse**

$$\left(\frac{a_x}{a_{x,max}}\right)^2 + \left(\frac{a_y}{a_{y,max}}\right)^2 \le 1$$


## Trackside takeaways

- QSS sims are excellent for gearing, mass, power, drag and fuel effects; transient sims are needed for dampers and inertia.
- Correlate on speed, g and steering vs distance before using a model; tune physical parameters before local grip factors.
- Parameter studies show what matters at a specific track – grip almost always dominates.
- The first outing of an event is a check that the real car matches the sim; big mismatches can reveal technical faults.
- Large predicted gains from small changes usually indicate a model error.

## By car category

### GT3

BoP is set with simulation; team QSS models help predict the lap-time effect of BoP weight/power changes and choose wing/ride height within BoP limits.

### LMP2

Spec car with limited adjustments makes models easier to correlate; sims guide gear choice (where permitted) and aero level per track.

### Hypercar (LMH/LMDh)

Optimal-control lap sims and DIL simulators are core tools; energy-limited stints (BoP energy per stint) make energy deployment optimisation part of the simulation.

### FIA F2

Teams use sims for setup baselines within the spec car limits; DIL time is used for new tracks and driver preparation.

### IndyCar

Oval simulations focus on drag/downforce trims and tow effects; road-course sims on mechanical setup and hybrid deployment.

### Formula 4

Even a simple point-mass model correlated to logger data teaches gearing and the value of minimum-speed vs exit-speed trade-offs.

## References

- Segers, Analysis Techniques for Racecar Data Acquisition, 2nd ed. (SAE 2014): Ch. 15 Vehicle Simulation (§15.1, §15.3 Lap Time Simulation, §15.4 Worked Example, §15.5 Integration in Daily Data)
- Milliken & Milliken, Race Car Vehicle Dynamics (SAE 1995): Ch. 9 "g-g" Diagram
