What real tests has this calculator been validated against?

Against 312 instrumented rear-end tests from the AGU/AZT database of the Dynamic Test Center in Zurich, 184 car-to-car tests with one vehicle stationary, the 12 RICSAC oblique collision tests and two published rollover tests.

What error does the Delta-V calculation carry?

It depends on the regime. In straight rear-end impacts, 9.6% median error with 96% of cases within ±30%, and a 7.2% tendency to over-predict that should be declared. With partial overlap or an angle, the error doubles, and there the eccentricity correction must be ticked: with it the median drops from 22.0% to 16.2%. Across impacts of every severity, the median error is 15.4%.

Is it better than a commercial reconstruction package?

That cannot be claimed from four tests. On the four RICSAC tests Virtual CRASH publishes, and with the same approach —momentum from trajectories— module 03 gives 8.6% against their 10.9%. Their full simulation, tuning four parameters per test, gives 5.4%.

Do these data apply to European vehicles?

The physics is the same, but the stiffness and inertia values come from tests on the US fleet, because no equivalent published data exist for the European fleet. That is a limitation worth declaring in the report.

How this calculator is validated

Updated 2026 · 5 min read

Anyone can publish a calculator. What distinguishes one from another is whether somebody has measured it against accidents where the answer was known, and whether it also publishes where it fails. That last part is what is almost never found, and it is what this page is about.

What it has been measured against

Reference	Module	Result
312 straight rear-end tests from the AGU/AZT database (Zurich), ΔV measured with accelerometers	01	9.6% median error · 96% within ±30% · over-predicts by 7.2%
· separately, 102 tests with partial overlap or an angle	01	error doubles: 22.0% median · with the eccentricity correction, 16.2% and 78% within ±30%
184 car-to-car tests with one vehicle stationary	01 and 04	median error 15.4% · 77% within ±30%
12 RICSAC tests, instrumented oblique collisions	03	8.4% mean error
2 published rollover tests	08	reproduces the speeds within 0.5%

The straight rear-end row and the 184-test row say different things and should not be mixed: at low speed the method is accurate, which is precisely the regime in which neck injuries are argued; across the whole sample, with impacts of every severity, the median error doubles. A report quoting 10% for a 60 km/h impact would be quoting the wrong number.

Against a commercial package

Virtual CRASH publishes its own results on four of the RICSAC tests. Comparing on those same four:

Method	Mean error in speed
Virtual CRASH · simple momentum	10.9%
DELTAVCAL · module 03	8.6%
Virtual CRASH · simulation with four tuned parameters	5.4%

Their full simulation is the best of the three, and it ought to be: it tunes four parameters per test until it reproduces rest positions it already knows. What is interesting is what happens to the ΔV, which is what matters in an injury report: their error in ΔV is 10.0%, nearly double their error in speed. That is no accident — they optimise against the rest positions, not against the ΔV, and what does not enter the fit does not get fitted.

What cannot be claimed

Four tests are not a sample. Nothing follows from them about one method being better than another, only that on those cases and with the same approach it does not fall behind.
The test bench has its own noise. The measured RICSAC data are internally inconsistent by a median of 13.8%: the module lands below the error of its own reference, and that puts a floor on what can be claimed.
The vehicle data are from the US fleet: 370 vehicles with measured inertias and 266 records with wheel weights and centre of gravity. There are no published measured inertias for the European fleet. The physics transfers well, but the starting data are what they are and that should be said.
An electric vehicle does not fit the splits of a combustion car: centred battery, motors on one or both axles and a lower centre of gravity. It is offered as its own option, but with fewer tests behind it.

Open the calculator →

Frequently asked questions

What real tests has this calculator been validated against?: Against 312 instrumented rear-end tests from the AGU/AZT database of the Dynamic Test Center in Zurich, 184 car-to-car tests with one vehicle stationary, the 12 RICSAC oblique collision tests and two published rollover tests.
What error does the Delta-V calculation carry?: It depends on the regime. In straight rear-end impacts, 9.6% median error with 96% of cases within ±30%, and a 7.2% tendency to over-predict that should be declared. With partial overlap or an angle, the error doubles, and there the eccentricity correction must be ticked: with it the median drops from 22.0% to 16.2%. Across impacts of every severity, the median error is 15.4%.
Is it better than a commercial reconstruction package?: That cannot be claimed from four tests. On the four RICSAC tests Virtual CRASH publishes, and with the same approach —momentum from trajectories— module 03 gives 8.6% against their 10.9%. Their full simulation, tuning four parameters per test, gives 5.4%.
Do these data apply to European vehicles?: The physics is the same, but the stiffness and inertia values come from tests on the US fleet, because no equivalent published data exist for the European fleet. That is a limitation worth declaring in the report.

Sources

The sets of instrumented tests the calculator has been measured against:

Neades, J. (AiTS) & Smith, R. (De Montfort University). The Determination of Vehicle Speeds from Delta-V in Two Vehicle Planar Collisions, appendices B and E. The data for the twelve full-scale RICSAC tests (NHTSA, 1978), with measured impact speeds, are taken from there.
AGU/AZT database, Zurich (Dynamic Test Center): real instrumented rear-end impacts, used to check module 01.
NHTSA rigid-barrier frontal tests, the source of the stiffness coefficients and the benchmark for the damage module’s accuracy.
McHenry, R. R. (1976). User’s Manual for the CRASH Computer Program. Calspan ZQ-5708-V-3. Its Exhibit 9-5 lists fourteen surfaces with four pairs of friction values each.