3. Simulation data sources

Correlation work begins with a deceptively simple question: which simulation data are we actually comparing against the test? In crash simulation, the answer is rarely a single file. An explicit finite element (FE) run produces a directory of outputs, each with different time sampling, spatial resolution, and physical meaning. Choosing the wrong channel, the wrong sampling rate, or the wrong coordinate convention will corrupt the correlation before any metric is computed. This section covers where explicit solver data comes from, how LS-DYNA structures its output, and how to extract result channels that are physically comparable to test measurements.

Why Explicit Solvers Produce the Data They Do

Crash simulation is dominated by transient, high-rate, nonlinear events: contact, buckling, rupture, and large deformation. Explicit time integration advances the solution with a central-difference scheme, where the stable time step is governed by the smallest element and the material wave speed:

dt_stable ≈ L_min / c,   c = sqrt(E / ρ)

For a typical steel shell element with characteristic length L_min = 5 mm, E = 210 GPa, and ρ = 7.85e-9 tonne/mm³, the wave speed is about 5.17 mm/ms, giving a stable step near 0.97 µs. A 120 ms event therefore requires on the order of 10⁵ time steps. This matters for correlation because the solver writes output at user-specified intervals, not at every step. The output interval, not the integration step, determines the effective sampling rate of your simulated signal.

Two consequences follow. First, if the test data are sampled at 10 kHz (0.1 ms) but the simulation writes state every 1 ms, the simulated peak may be missed entirely. Second, filtering must be applied consistently: test accelerometers are typically filtered per SAE J211, and the simulation output must be filtered with the same class and cutoff before comparison.

LS-DYNA Output Families

LS-DYNA organizes results into distinct output families, each controlled by its own keyword and written to a separate file. Understanding the family is the first step in extraction.

Output family	Keyword	Typical file	Content and correlation use
State database	*DATABASE_BINARY_D3PLOT	d3plot	Full-field state (stress, strain, displacement, plastic strain) at discrete times; used for deformation and intrusion correlation.
Time-history, ASCII	*DATABASE_ASCII_*	nodout, elout, secforc, rcforc	Nodal kinematics, element results, section forces, contact forces; the primary source for channel-by-channel correlation.
Time-history, binary	*DATABASE_BINARY_D3THDT	d3thdt	Binary time histories for selected nodes/elements; compact and fast to post-process.
Interface/contact force	*DATABASE_BINARY_RCFORC	rcforc	Resultant contact forces per interface; used for barrier force and occupant load-path correlation.
Energy	*DATABASE_GLSTAT, *DATABASE_MATSUM	glstat, matsum	Global and material energy balance; used to validate the run before correlation.

The d3plot file is the workhorse for spatial correlation, while ASCII time-history files are the workhorse for signal correlation. A common mistake is to extract a nodal acceleration from d3plot by differentiating displacement; this amplifies numerical noise and produces a signal that cannot be compared to a filtered accelerometer. Prefer the solver-computed acceleration from nodout or a dedicated accelerometer element.

Extracting Channels for Correlation

Correlation requires that the simulated channel and the test channel share the same physical quantity, location, direction, and sign convention. The extraction workflow below is written for a frontal impact model with a vehicle body and a rigid barrier.

1. Define the measurement point. Identify the test sensor location, for example a left B-pillar base accelerometer. In the model, locate the nearest node or define a nodal set. Record the node ID and its coordinates.
2. Fix the coordinate convention. LS-DYNA uses a global Cartesian system. Define the vehicle +X as forward, +Y as left, +Z as up, and state this explicitly. Test accelerometers are often mounted with a polarity that must be reconciled; a sign flip is a frequent root cause of poor correlation.
3. Request the correct output. Add the node to a set and request acceleration and velocity:

*DATABASE_ASCII_NODOUT
$#    dt    lcdt    binary    dthdf
     0.1       0         0        0
*SET_NODE_LIST_TITLE
B_PILLAR_BASE_ACCEL
$#     sid       da1       da2       da3       da4
         1
$#   nid1      nid2      nid3      nid4
   10234
*DATABASE_HISTORY_NODE_SET
$#     id1       id2       id3       id4
         1

Here dt = 0.1 ms matches a 10 kHz test sampling rate. The nodout file will then contain time, X/Y/Z displacement, velocity, and acceleration for node 10234 at each interval.

4. Extract and align. Parse the nodout block for the node, build a time vector, and interpolate both test and simulation onto a common time base. A typical alignment uses the time of first barrier contact as t = 0.
5. Filter consistently. Apply the same SAE J211 filter class to both signals. For a B-pillar base channel, CFC 60 is common; for a lower leg channel, CFC 180. Never filter one signal and not the other.
6. Check the run before trusting the channel. Inspect glstat for energy balance and matsum for negative or excessive internal energy. A run with hourglass energy above roughly 5% of internal energy is not a reliable correlation baseline.

From Raw Output to Correlation-Ready Data

The figure below summarizes how a single LS-DYNA run feeds multiple correlation channels, and where each channel is extracted from.

Common Extraction Pitfalls and How to Detect Them

Three failure modes account for most correlation problems at the data-source stage. First, sampling mismatch: the simulated signal is written too coarsely and misses the peak. Symptom: simulated peak is consistently lower than test, and refining dt changes the result. Second, sign and axis mismatch: the test channel is positive in compression while the simulation reports tension-positive. Symptom: the two curves are near mirror images. Third, unverified run: hourglass energy or negative internal energy indicates the model is not physically sound, and no amount of signal processing will fix it.

Correlation Data Extraction Checklist

Item	What to confirm	Typical value / example
Output interval	Matches or exceeds test sampling rate	dt = 0.1 ms for 10 kHz test
Node / element ID	Nearest to test sensor, correct set membership	Node 10234, set ID 1
Coordinate convention	Axis and sign documented and consistent	+X forward, +Y left, +Z up
Filter	Same class and cutoff on both signals	CFC 60 for B-pillar base
Time zero	Aligned to first barrier contact	t = 0 at contact
Run quality	Energy balance acceptable	Hourglass < 5% of internal energy

With these channels extracted and verified, the correlation stage can proceed on a sound footing. The next step is to define the comparison metrics and tolerances that will be applied to these aligned, filtered signals.