Forge Axiom PLM · Simulation Integration & Generative Design

Engine Technical
Design Document

Architecture, pipeline design, model specification, and performance validation across eight AI engines for simulation data governance, multi-solver orchestration, generative design, reduced-order modeling, and simulation-driven digital twin deployment. Built in Rust. Every model versioned. Every result traceable. Every insight preserved.

Simulation data is the most poorly managed artifact in all of engineering. 80% of organizations have no dedicated system. That ends here.

8
Intelligence Engines
80%
Lack Dedicated SDM
1,000×
ROM Speedup
$163B
Digital Twin Market 2029
engine_index
Eight engines for the data no one is managing
01
Data Governance
Auto-extraction from 5 solver formats into governed PLM
02
Multi-Solver
Ansys, Abaqus, COMSOL, OpenFOAM, Simcenter
03
DOE Automation
1,000+ parametric runs with surrogate generation
04
Verification Matrix
Auto pass/fail against acceptance criteria
05
Generative Design
Topology optimization within governed PLM boundaries
06
ROM Factory
1,000× speedup for digital twin deployment
07
Digital Twin
Physics-data hybrid with IoT sensor fusion
08
Results Intelligence
NLP cross-program pattern mining and knowledge capture
executive_summary
An eight-engine architecture for the simulation data crisis hiding on every engineer’s desktop

Simulation has become central to product development — but simulation data remains the most poorly managed artifact in most engineering organizations. Around 80% of companies have no dedicated simulation data management system. Simulation teams work on desktops, hard disks, shared drives, and unstructured server folders. The result: over the years, companies lose valuable simulation knowledge, cannot reuse results properly, repeat simulations, waste compute resources, and development becomes slower. The connection between a simulation result and the design iteration it validated is documented in someone’s notebook — or worse, in someone’s memory.

SPDM has been recognized as strategic infrastructure for digital engineering. NAFEMS held a dedicated International SPDM + AI/ML Conference in 2026, reflecting the convergence of simulation governance with artificial intelligence. The digital twin market is projected to expand by approximately $163 billion by 2029 at compound annual growth rates near 65% — but digital twins require governed simulation data as their foundation: the high-fidelity models that capture the physics, the reduced-order models that enable real-time execution, and the training data that validates prediction accuracy. Without simulation data governance, digital twin programs are built on sand.

Axiom Nexus makes simulation a first-class PLM artifact. Models, mesh configurations, boundary conditions, solver settings, results, and post-processed outputs are all versioned, linked to the design geometry they analyze, and connected to the requirements they verify. Integration with Ansys, Abaqus, COMSOL, OpenFOAM, and Simcenter enables automated multi-solver workflows. Reduced-order models deliver 1,000× speedup over high-fidelity FEA for digital twin deployment. Generative design explores thousands of candidates within governed PLM boundaries. Every model versioned. Every result traceable. Every insight preserved.

80%
Organizations Lack Dedicated SDM
$163B
Digital Twin Market by 2029
1,000×
ROM Speedup vs. High-Fidelity FEA
5
Solver Platforms Integrated
60%
Physical Prototypes Eliminated
65%
Digital Twin Market CAGR
ENG 01
Simulation Data Governance
Automatic metadata extraction from five solver formats into a governed PLM knowledge graph — because simulation data on local drives is simulation knowledge lost.
5
Solver Formats
Architecture
Parser Pipeline + Knowledge Graph
Format-specific parsers extract metadata (mesh stats, solver settings, boundary conditions, convergence data, result summaries) from Ansys (.rst/.dat), Abaqus (.odb/.inp), COMSOL (.mph), OpenFOAM (case dir), and Simcenter (.sim)
Governance
Versioned + Design-Linked
Every simulation artifact versioned in the Axiom knowledge graph; linked to the specific design revision analyzed; connected to the requirements being verified; searchable by any metadata field
Performance
Auto-Extraction <30 Seconds
Metadata extraction from a complete FEA model (100K+ elements) in under 30 seconds; background processing does not interrupt engineer workflow
Toolchain
Rust / Python / HDF5
Rust-native file parsers for binary solver formats; Python scripting API for custom extraction rules; HDF5 for large result dataset storage; full-text search across all simulation metadata

The fundamental problem is architectural: PLM systems were designed for CAD files and BOMs, not for the terabytes of simulation data that modern engineering produces. Result files live on local drives. The connection between a CFD result and the structural analysis it informed exists only in a manually maintained spreadsheet. When an engineer leaves, their simulation knowledge leaves with them. Engine 01 addresses this by treating simulation artifacts as first-class PLM objects — versioned, searchable, linked to design geometry, and connected to the requirements they verify. Format-specific parsers automatically extract metadata from the five major solver platforms, enabling cross-solver search, comparison, and reuse without requiring engineers to change their simulation workflow.

performance_validation
Metadata Extraction (100K elem)
<30sec
Solver Format Coverage
5/5
Cross-Solver Search Latency
<2sec
Knowledge Retention (vs. attrition)
100%
ENG 02
Multi-Solver Orchestration
Automated cross-solver data mapping between Ansys, Abaqus, COMSOL, OpenFOAM, and Simcenter — chaining thermal-structural-fatigue analyses with verified data handoffs instead of manual file translations.
5
Solvers Unified
Architecture
Workflow DAG + Data Mapping
Multi-physics workflows defined as directed acyclic graphs; automated mesh transfer, boundary condition mapping, and result interpolation between solver formats; versioned data handoff verification at each stage
Solvers
Ansys / Abaqus / COMSOL / OpenFOAM / Simcenter
Structural, thermal, CFD, electromagnetic, and multi-physics; solver-agnostic workflow definition with format-specific adapters; HPC job submission and monitoring
Performance
Zero Manual File Translation
Automated mesh-to-mesh interpolation for cross-solver workflows; verified data handoffs ensure thermal loads transfer accurately from CFD to structural analysis without engineer intervention
Impact
4→1 Solver Platforms Unified
Turbine manufacturer unified 4 separate solver platforms into a single governed workspace; eliminated 60% of physical prototypes; $4.2M annual test cost reduction
ENG 03
DOE & Parametric Automation
Design-of-experiments management for 1,000+ parametric runs with automatic surrogate model generation — exploring the design space systematically rather than one iteration at a time.
1,000+
Runs Managed
Architecture
Latin Hypercube + Surrogate
Latin Hypercube sampling for efficient design space exploration; automatic surrogate model training (Gaussian Process, Neural Network, Polynomial) from DOE results; Pareto frontier extraction for multi-objective optimization
Performance
1,000+ Run Management
Full DOE lifecycle: parameter definition, sampling, HPC job distribution, result collection, surrogate training, and sensitivity analysis — all within governed PLM
Features
Sensitivity & Response Surface
Automatic sensitivity analysis identifies which parameters have the greatest impact on each performance metric; response surfaces enable rapid what-if exploration without additional solver runs
Impact
10× Design Space Coverage
Engineers explore 10× more design alternatives in the same time by automating the run-collect-analyze cycle; every configuration and result governed in the knowledge graph
ENG 04
Requirements Verification Matrix
Automatic pass/fail evaluation of simulation results against acceptance criteria from the Axiom requirements engine — closing the loop between what you designed and what you proved.
Auto
Pass/Fail
Architecture
Requirement ↔ Result Linkage
Requirements from Axiom Meridian (requirements engine) linked to simulation results via typed relationships; automatic extraction of critical values from result datasets; pass/fail evaluation against quantitative acceptance criteria
Performance
Instant Verification Status
Requirements verification matrix updates automatically when new simulation results are available; gap analysis identifies requirements with no verification evidence
Features
Gap Detection
Identifies requirements that lack any simulation or test verification; flags requirements where the latest simulation used an outdated design revision; supports both simulation and physical test evidence
Impact
Zero Missing Verification
Eliminates the manual process of assembling verification matrices for design reviews and regulatory submissions; every requirement traceable to specific simulation evidence
ENG 05
Generative Design & Topology Optimization
Thousands of governed design candidates generated by AI within PLM-managed constraints — because generative design without lifecycle governance produces designs that cannot be manufactured, certified, or maintained.
34%
Weight Reduction
Architecture
Topology Opt + Multi-Objective
Density-based topology optimization with manufacturing constraints (stamping, casting, extrusion, additive); multi-objective Pareto optimization across weight, stress, frequency, and thermal performance
Performance
2,400 Candidates Explored
EV battery enclosure: 2,400 candidates across wall thickness, rib patterns, material grade, and manufacturing process — 34% weight reduction with full crash certification
Features
Manufacturing Feasibility
Every generated design checked against manufacturing process constraints; no topology-optimized geometry that requires 5-axis machining when only 3-axis is available at the production facility
Impact
Design Differentiation
Generative design has moved from experimentation to industrial practice; in competitive markets, it is increasingly a differentiating capability rather than a novelty
ENG 06
Reduced-Order Model Factory
1,000× speedup over high-fidelity FEA through physics-neutral surrogate models that capture system behavior for real-time digital twin deployment — with validity envelope tracking and automatic regeneration.
1,000×
Speedup
Architecture
POD + ML Surrogate + FMU Export
Proper Orthogonal Decomposition for physics-based reduction; ML surrogates (MLP, Gaussian Process) for parametric response; export to MATLAB, Python, FMU/FMI, and Ansys Twin Builder formats
Performance
<5% Error vs. Full-Physics
Reduced-order models achieve under 5% error compared to full high-fidelity solutions while executing 1,000× faster; validity envelope tracking flags extrapolation beyond training domain
Features
Validity Envelope Tracking
Every ROM tagged with parameter ranges for which it has been validated; queries outside the envelope return uncertainty estimates and flag the need for additional high-fidelity training runs
Impact
Digital Twin Foundation
Validated ROMs packaged and deployed as real-time digital twin engines; connected to IoT sensor data, the ROM continuously predicts system behavior and detects anomalies
ENG 07
Simulation-Driven Digital Twin
Physics-data hybrid twins grounded in first principles and calibrated by reality — because the most mature digital twin implementations combine simulation models with real-time operational data.
45%
Downtime Reduction
Architecture
ROM + IoT Fusion + Anomaly Detection
Reduced-order models from Engine 06 deployed as real-time prediction engines; IoT sensor data calibrates model parameters; Kalman filter fuses physics prediction with measurement for optimal state estimation
Performance
Real-Time Prediction + Anomaly
Continuous prediction of system behavior at 1Hz or faster; automatic anomaly detection when measured behavior departs from physics-predicted behavior — indicating developing fault or changed operating conditions
Features
Governed Simulation Foundation
Every model, training dataset, validation result, and deployment configuration governed in the Axiom knowledge graph; full traceability from field prediction back to the high-fidelity simulation that generated the ROM
Impact
1,200 Field Assets Twinned
Compressor OEM deployed simulation-driven digital twins across 1,200 field assets; 41% reduction in unplanned downtime; predictive maintenance accuracy exceeding traditional vibration analysis alone
ENG 08
Results Intelligence & Knowledge Capture
NLP-powered cross-program pattern mining that discovers insights buried across thousands of historical simulations — because the answer to your current design question may already exist in a simulation someone ran three years ago.
NLP
Cross-Program
Architecture
Vector Search + Pattern Mining
Semantic vector search across simulation metadata, results, and engineering notes; pattern mining identifies recurring failure modes, design sensitivities, and optimization opportunities across programs
Performance
Cross-Program Insight Discovery
Natural language queries: “Show me all simulations where thermal stress exceeded yield in titanium alloy joints under cyclic loading” — returns results from any program, any year, any solver
Features
GenAI Integration
2026 frontier: generative AI capabilities integrated into CAE workflows for finding data faster, reusing knowledge, and improving productivity — engineers expect AI to help them extract value from historical simulation data
Impact
Zero Knowledge Lost
Structured simulation data forms the basis for automation, creating a multiplier effect: fewer errors, shorter development cycles, and a growing institutional knowledge base that survives personnel changes
simulation_impact
80%
Organizations lacking dedicated SDM
60%
Physical prototypes eliminated (case study)
34%
Weight reduction via generative design
1,200
Field assets with simulation-driven digital twins