We Rebuilt Four Viral Destruction Videos in LS-DYNA

The internet runs on destruction: bullets fired through stacks of steel, a hydraulic press flattening a soda can, a bowling ball dropped onto a trampoline, an anvil dropped through the roof of a car. These clips rack up hundreds of millions of views — and every one of them is, underneath, an explicit finite-element problem. So we picked four of the most recognizable formats and rebuilt each as an Ansys LS-DYNA solve, aiming for a render you could set next to the original and see the same event. Here is the batch, and the engineering that makes each one behave.

1 · How many steel plates does the bullet go through?

This is the classic ballistics-channel format: fire a round into a row of steel and count how many it punches through. Each bullet is a copper-jacketed lead core meshed as conformal all-hex solids; both metals use a Johnson–Cook strength model with a Grüneisen equation of state (literature constants for lead, brass and steel), and shed material through effective-plastic-strain erosion. The plates are thin mild-steel shells that perforate and petal. A single *CONTACT_ERODING_SINGLE_SURFACE handles the bullet, every plate and all the debris at once, with element erosion switched on so collapsing elements delete cleanly instead of tangling the solve.

The payoff is the spread — and it falls out of the physics, not a script:

The little 9 mm flattens against the first plate; the heavy .50 BMG drives its core through five and shatters the front of the stack. And it’s momentum, not muzzle speed, that sets the depth: the fast, light .223 (950 m/s) cores fewer plates than the slower but nearly three-times-heavier .308 (840 m/s) — exactly what sectional density predicts. That the ordering follows mass and momentum rather than raw speed is the tell that the solve is keeping the books honestly.

2 · Soda can vs hydraulic press

The press is the friendliest of the four to model and the hardest to make look right: a thin-wall aluminum can is nothing but a buckling problem. The can is a 0.15 mm shell of work-hardened 3104 aluminum (*MAT_PLASTIC_KINEMATIC) with a faint twelve-lobe geometric imperfection seeded into the wall so the buckle mode is deterministic rather than a product of mesh noise. A moving rigid wall drives the crush; *CONTACT_AUTOMATIC_SINGLE_SURFACE lets the collapsing folds stack against themselves. It reaches a flat disc with zero spurious added mass — the sign that the buckling energy is real.

3 · Bowling ball vs trampoline

The trampoline is a thin elastic shell (a fully-integrated shell with negligible bending, so it drapes like fabric) with a strain-based tear (*MAT_ADD_EROSION), suspended over a soft dirt bed of solid elements a meter below. The whole trick is the ordering, and it comes from one number — the failure strain. Set it too low and the ball punches a hole almost on contact: a flat bed, no stretch, no rebound. Raise it and the bed draws into a deep cone first, storing the ball’s energy elastically, and only tears at the very bottom of the draw. The instant that central patch lets go, the pre-tensioned ring of surviving fabric loses its load and snaps back up past flat — the rebound — while the ball, still carrying most of its speed, falls the last stretch of open air and craters the dirt. Elastic (not plastic) fabric and almost no damping are what keep that snap energetic; a plastic bed would have swallowed the energy and just sagged.

4 · Anvil through the roof of a car — with sound, generated from the simulation

The car is a procedurally-built shell model — floor, rockers, firewall, hood, doors, a fine roof, A/B/C pillars, and brittle glass — welded into a continuous roof→pillar→rocker→floor load path so the crush travels the way it does in a real body. The steel dishes and tears on a plastic-strain criterion (a thin, weak roof so it perforates; stiff pillars so they hold the roof edge and the anvil punches through rather than just denting it); the glass is elastic with a max-principal-stress failure, so the windows blow out as the frame deforms. The anvil is a rigid forging with a finely-subdivided striking face — the burned-in lesson that one giant rigid segment against a fine mesh explodes the penalty contact.

The sound is the simulation. A car crash makes noise because the panels vibrate and radiate it into the air. So we take the roof’s surface acceleration — sampled from the solve at 48 kHz — and run the acoustic radiation (Rayleigh) integral, p(t)=Σ ρA an(t−r/c)/2πr, summing each panel element’s contribution to the pressure at a listener a few meters away, each delayed by its travel time. Band-limit, normalize, and it is a WAV — the same physics LS-DYNA’s own vibro-acoustic solver uses, done in the time domain so we get a waveform to put on the video. It is an honest sonification of the simulated panel motion (point listener, free-field, one-way coupling), not a psychoacoustic prediction and not a library sound effect.

What actually makes these work

The right failure model for each material. Bullets and plates shed mass through Johnson–Cook plasticity and strain-based erosion; the can never fails at all — it just buckles; the trampoline draws elastically into a deep cone and then tears on a strain criterion so it can rebound; the car’s thin roof tears while its stiff pillars buckle and its glass shatters — three different failure modes in one body. Matching the response mode to the real object is most of the battle.

Contact is where solves live or die. An eroding single-surface for the fragmenting ballistics; a self-contact for the folding can; a thick, segment-based, sort-every-cycle contact for the fast bowling ball (a fast projectile can otherwise skip straight through a thin target between time steps); a finely-subdivided striking face on the anvil so the penalty stiffness stays stable against the car’s roof mesh. Each is chosen for its event.

The timestep has to respect the speed. A fast bowling ball moving through a 2 mm membrane will tunnel straight through it in a single step unless the step is small enough — and the contact zone is thick enough — to catch it; diagnosing that tunnelling was the key to the trampoline.

How it was built

Every geometry (the O-grid bullets, the plate stacks, the can, the trampoline bed, the anvil and panel) is generated in Python straight to LS-DYNA keyword decks. Candidate videos were shortlisted by a fleet of research agents scoring each on recognizability, feasibility and how closely we could match it, then the two clearest picks — the ballistics stack and the press — were built first on recipes we already trusted, and the trampoline and drop added for range. Post-processing reads the d3plot through Ansys PyDPF (using the erosion flag so torn and perforated elements vanish cleanly) and renders in PyVista with physically-based metals and a neutral studio environment. Four viral formats, one long night.

Ansys LS-DYNA

Explicit dynamics

Johnson–Cook

Eroding contact

Membrane + failure

PyDPF

PyVista PBR

What happens to your design in the millisecond after impact — and could you defend the answer? Four Ansys LS-DYNA explicit solves, post-processed through Ansys PyDPF and rendered in PyVista — Johnson–Cook plasticity with eroding contact that ordered penetration by momentum exactly as sectional density predicts, a can crush held to zero spurious added mass, a trampoline tear that only behaved once the timestep stopped letting a fast bowling ball tunnel through the membrane — is how simulation lets you watch a destructive event frame by frame before you sacrifice hardware to film it. That's innovation through insight.