Deep Learning Coupled FEM Composite Optimization

A fully parametric CAD and FEM workflow built to explore the composite architecture of launch vehicle structures — fairings, tanks, and payload adapters — far faster than a manual design study allows. Rather than sizing one configuration at a time, the same Python-driven pipeline automatically builds geometry, applies laminate stacking, and submits FEM jobs across a wide sweep of structural architectures.

On a composite fairing structure, this workflow was used to systematically compare monolithic, sandwich, and stiffened-monolithic architectures against a shared mass-efficiency metric, converging on manufacturable configurations that met a demanding buckling safety margin at minimum mass.

The fairing was defined as a fully parametric surface model rather than a fixed geometry: skin thickness and layup, longitudinal and circumferential reinforcement count/width/thickness/layup, interface corner fitting dimensions, and nose material and thickness were all exposed as independent parameters. Separation device count and position, and the overall ogive/cylinder/cone shape, were parametrised alongside them — giving the automation pipeline full control over both the structural architecture and the outer mould line from a single model definition.

Each candidate configuration was evaluated under a representative external pressure field, applied as distinct constant and linearly-varying pressure distributions across the ogive, cylindrical, and conical zones, using orthotropic carbon/epoxy (UD and woven), structural foam core, and aluminium fitting properties as FEM inputs. A mass-efficiency parameter — buckling safety factor divided by mass squared — was computed automatically for every configuration in the sweep, letting the workflow rank dozens of architectures on a single, consistent basis rather than requiring engineering judgement call-by-call.

This automated sweep showed that a classic monolithic skin construction converged on an optimum skin thickness for a target buckling safety factor above 5, landing in a 60–70 kg mass range; thinner-skin monolithic variants pushed toward a lighter ~48 kg but with safety margins the pipeline correctly flagged as insufficient. Stiffened-monolithic ("sandwich raidi") variants — both longitudinal and circumferential omega-stiffened architectures — were also swept automatically, but the pipeline surfaced a key manufacturing trade-off: added stiffeners reduce usable payload envelope and drive up mass elsewhere in the structure, an insight that would be far slower to reach through one-off manual iterations.

The best-performing architectures identified by the sweep were carried through to full-assembly FEM validation — buckling, static, and modal analysis under representative boundary conditions — confirming a first lateral mode at 62 Hz and a first longitudinal mode at 360 Hz for the selected configuration, with buckling and displacement results within target margins.