FALCON

Falcon covers slope stability (shear-strength reduction), Biot consolidation and settlement, seismic site response and earthquake slope stability, cyclic liquefaction assessment, soil-structure interaction for piles and retaining walls, staged deep excavation and tunnel support, unsaturated wetting-front slope failure, large-deformation pile driving and cone penetration, and dam and embankment analysis. All analysis types share one material library and one project file.

Falcon ships the model families that actually govern the hardest geotechnical problems — not padded with elasticity variants. For cyclic sand and liquefaction: SANISAND (four family variants for different calibration datasets), Elgamal (seismic sand), MIT-S1 (stress-induced anisotropy), and Norsand (state-parameter). For critical-state clay: CASM, Modified Cam-Clay, SANICLAY, Subloading surface. For unsaturated soils: a thermodynamically consistent effective-stress framework with a single failure criterion from dry to saturated — not a bolt-on. For rock and classical plasticity: Hoek-Brown with shear-strength mobilization, Mohr-Coulomb with smooth corners, Mohr-Hardening, Orthotropic elasticity. For custom models: a published C++ UMAT interface with state-variable machinery and a Minis verification harness.

Yes. Falcon supports three time-integration schemes — explicit, IMEX, and implicit — that can switch within a single staged run. An earthquake shaking stage transitions directly into pore-pressure dissipation consolidation on the same mesh with the same material state.

Falcon implements Perfectly Matched Layers (PML) for in-plane and out-of-plane wave absorption in uncoupled, coupled, and fully coupled formulations; mapped-infinite elements for unbounded static and consolidation domains; and sponge layers for targeted energy absorption. All three methods are available in one solver.

Falcon uses updated-Lagrangian kinematics with Arbitrary Lagrangian-Eulerian (ALE) remeshing. When elements distort under pile driving, footing penetration, or progressive failure, ALE fires, maps state variables to the new mesh, and continues. Adaptive time stepping handles the convergence. This is an Enterprise-tier capability.

Falcon Orbit is a purpose-built AI co-pilot compiled into the same C++20 binary as the solver. It is a three-layer system: (1) a command planner that reads live model state and converts natural language to typed Falcon commands you approve step by step; (2) a four-stage hybrid RAG pipeline — SQLite FTS5 lexical search, local Hugging Face bi-encoder embeddings, cross-encoder reranking, and reciprocal-rank fusion — indexed over the manual, project history, run artifacts, and the GUI command reference; (3) a deterministic mechanics validation layer that checks every result against Terzaghi consolidation, Boussinesq stress distribution, Bishop simplified FoS, and Cam-Clay critical-state endpoints before the AI returns a verdict, followed by a second-opinion LLM reconciliation pass. Local-first, fully inspectable, available in Pro and Enterprise builds.

Orbit's retrieval pipeline has four stages. Stage 1: SQLite FTS5 lexical search for exact and near-exact matches across the indexed corpus — Falcon manual, solver source, validation reports, project revision history, run artifacts, Minis catalog, and the GUI command reference. Stage 2: local Hugging Face bi-encoder embedding models retrieve semantically similar passages from the same corpus, running entirely offline with no cloud dependency. Stage 3: a cross-encoder reranker rescores the combined candidate set for precision. Stage 4: reciprocal-rank fusion merges the lexical and semantic ranked lists into a single grounded context. Every stage runs on the local machine. Project data and model parameters never leave the device without an explicit opt-in to cloud inference keys.

LLMs can produce numerically plausible but physically incorrect answers. Orbit's deterministic validation layer runs closed-form geotechnical baselines — Terzaghi primary consolidation, Boussinesq stress distribution, Bishop simplified factor of safety, Modified Cam-Clay critical-state endpoints, and Terzaghi bearing capacity — on every result before the AI returns a verdict. An independent second-opinion LLM pass then reconciles the mechanics critique with the structured solver-run interpretation. The combined verdict is surfaced alongside every result, making the validation process explicit and auditable.