Manufacturing Process Simulation

That's a really natural question, and the answer is: because the manufacturing process determines the part's properties — sometimes more than the design geometry does. When you stamp a steel door panel, the pressing operation induces residual stresses and local thinning. If you don't account for those in your crash simulation, your safety predictions can be off by 20% or more. The part that goes into service isn't the idealized CAD model — it's the deformed, pre-stressed, work-hardened result of the manufacturing process.

Because a lot can go wrong between liquid metal and solid part. The molten alloy has to fill the mold completely before it freezes — if it solidifies prematurely in a thin section before the far end fills, you get a cold shut defect. As the alloy solidifies, it contracts; if that contraction can't be fed by molten metal from a riser, you get shrinkage porosity inside the casting. And the non-uniform temperature during cooling causes residual stresses and dimensional distortion. Casting simulation predicts all of these before you cut a single mold — saving weeks of trial-and-error tooling work.

Manufacturing simulation is genuinely one of the most challenging areas in all of CAE — you often need large plastic deformation, contact with rigid tooling, heat generation from plastic work, phase transformation, and all of this happening simultaneously. For sheet metal forming: large strain plasticity up to 50% local strain, contact between sheet and die at high pressure, springback after tool release. For welding: a moving heat source, melting and re-solidification, solid-state phase transformations in the heat-affected zone, and residual stresses that can cause the part to distort out of tolerance the moment you remove the fixtures.

Explicit time integration — where you step forward without solving a system of equations — handles contact discontinuities and large deformations with far fewer convergence issues than implicit. When you're stamping a sheet at 0.5 m/s with hundreds of contact nodes changing state every microsecond, the implicit Newton-Raphson scheme can fail to converge. The explicit scheme marches through all of that robustly. The price is a very small stable time step — governed by the Courant condition, $\Delta t \leq L/c$ where $c$ is the acoustic wave speed — which means you often need millions of time steps. For welding residual stress and springback, however, you switch back to implicit for the quasi-static springback release, because explicit would require impossibly many steps for slow processes.

Additive manufacturing simulation is at the frontier of the field right now. The challenge is scale: a laser powder bed fusion (LPBF) machine deposits layers 20–50 μm thick, and a finished part might have thousands of layers. If you simulate every layer with a fine mesh resolving the melt pool — which is only about 100 μm wide — the computation would take years. So the industry has developed "inherent strain" methods that skip the detailed melt pool physics and instead map pre-computed layer-by-layer distortion data onto a coarser model, giving good deformation and residual stress predictions in hours instead of years. It's a major approximation, but it works well enough for support optimization and build orientation selection.

Each has carved out a niche based on where they excel. AutoForm is the dominant tool for sheet metal forming in the automotive industry — purpose-built, fast, with excellent springback prediction and die compensation workflows. LS-DYNA is the most general explicit solver and handles forming, crash, and impact; it's widely used for everything from car body stamping to bird strike on aircraft. DEFORM specializes in bulk forming processes — forging, extrusion, rolling — where very large strains and high temperatures are involved. MAGMASOFT is the industry leader for casting simulation. For welding and additive manufacturing, Simufact Welding, Sysweld, and Amphyon (AM) are purpose-built tools, though Abaqus with user subroutines is used for research-level accuracy.

Using the wrong material model — or more precisely, using a material model that isn't calibrated for the actual strain rate and temperature range of the process. In sheet metal forming, using isotropic hardening instead of kinematic hardening means you'll miss the Bauschinger effect, and your springback prediction will be wrong by several degrees. In casting, using a simple elastic model for solidification stresses instead of a viscoplastic model means your residual stress predictions are unreliable. Material data calibration — tensile testing, biaxial testing, high-strain-rate Split Hopkinson bar tests — is often more important than mesh refinement or solver choice. Garbage in, garbage out, as always in CAE.