Skip to content
ES

Materials simulation

The AI Scientist that computes, verifies and ships on its own.

ACHT architecture: plans the simulation, extracts data, computes and traces every figure to its run with SHA-256. VAP peer review. The verified dossier comes out of the pipeline, not a consultant.

  • Peer-reviewedACHT and VAP architectures
  • Counter-signedLicensed mechanical engineer
  • In useCeramics laboratory
Speed of 6 750 particles · one revolution, 12 sGeneric rotary dryer · 2.0 m · 5 rpm

Verified science dossier

Open a real Verified science dossier

Demo A: shell-to-bottom joint of a generic atmospheric tank, computed with finite elements. Every figure links to its results file, with its SHA-256, and to its logged run. It is published open, as UNSIGNED, for your review.

Also open: Calculators · Publications

01Methodology

How the AI Scientist works

Five stages, always in the same order. No figure reaches the dossier without the run that produced it.

  1. 01 Plan

    Converts the question into a calculation plan: applicable standards, declared assumptions and acceptance criterion.

  2. 02 Extract

    Gathers geometry, materials and published tests, each datum with its source.

  3. 03 Compute

    Runs the finite elements and DEM, with mesh convergence study.

  4. 04 Verify

    Cross-checks every figure against the record of its run and reference solutions.

  5. 05 Report

    Drafts the dossier; the engineer reviews and counter-signs it.

Eight in-house modules, each with its evidence

All of them run today; the right-hand column says how each one is checked.

ModuleWhat it doesStepVerification
DEM engineGeometries and particle simulations with implementación de referencia DEM on remote serverComputeCalibrated against the published cone test of Li et al. (2021)
FE engineS8R mesh, calculation with the finite-element solver, results in JSONCompute< 0.3 % vs exact solution and mesh convergence (Richardson)
Wear-life modelRelative steel plate life via two-segment power law calibrated with Monte CarloComputeMonte Carlo of 200,000 samples, declared falsifiers
Web calculatorsThree dependency-free calculators: silo Janssen, weld fatigue S-N, membrane stressReport20 JS cases = Python at 1e-9 tolerance
Dossier builderGenerates self-contained scientific dossier: JSON, HTML in ES/EN, ZIP and SHA-256ReportHash per figure and execution log
Paper pipeline (pf)Orchestrates the full article lifecycle: ideation, experiments, submissionPlan / Verify105 accepted works, register as of 17-10-2026
Anti-fabrication auditExtracts numerical claims and verifies each value against attached artefactsVerifyClaims checked against their artefacts
Proof ledgerBuilds the unified submission ledger reproducibly and deterministicallyVerifyDeterministic, verifiable with --check
DEM engine · Compute

Geometries and particle simulations with implementación de referencia DEM on remote server

Calibrated against the published cone test of Li et al. (2021)

FE engine · Compute

S8R mesh, calculation with the finite-element solver, results in JSON

< 0.3 % vs exact solution and mesh convergence (Richardson)

Wear-life model · Compute

Relative steel plate life via two-segment power law calibrated with Monte Carlo

Monte Carlo of 200,000 samples, declared falsifiers

Web calculators · Report

Three dependency-free calculators: silo Janssen, weld fatigue S-N, membrane stress

20 JS cases = Python at 1e-9 tolerance

Dossier builder · Report

Generates self-contained scientific dossier: JSON, HTML in ES/EN, ZIP and SHA-256

Hash per figure and execution log

Paper pipeline (pf) · Plan / Verify

Orchestrates the full article lifecycle: ideation, experiments, submission

105 accepted works, register as of 17-10-2026

Anti-fabrication audit · Verify

Extracts numerical claims and verifies each value against attached artefacts

Claims checked against their artefacts

Proof ledger · Verify

Builds the unified submission ledger reproducibly and deterministically

Deterministic, verifiable with --check

Dossier service and orchestrator: roadmap.

Two architectures and the peer-reviewed papers behind them

The five-step workflow rests on two published architectures of the AIXC Research platform.

ACHT (Active Causal Hypothesis Testing)

An engine that generates and prioritises mechanistic hypotheses using causal graph learning and Bayesian optimal experimental design, maintaining an explicit belief state and replayable decision traces. [1, 2]

VAP (Verified Agentic Pipeline)

A verification architecture, first published for formal validation in time-series systems and extended to science agents and materials simulation, that gates every agent step against physics-informed constraints and formal checks. [3, 4]

VAP has been applied in three domains (financial time series, robotics for autonomous science and molecular simulation). ACHT and VAP are platform architectures: they were not used in the steel cases on this page.

Nine base papers

  1. The Autonomous Experimenter (ACHT)IEEE IROS 2025 Workshop AIR4S · workshop · 2025
  2. Provably Safe Robotics for Autonomous Science (VAP)IEEE IROS 2025 Workshop AIR4S · workshop · 2025
  3. The Architectural Immune SystemAgents4Science 2025 · symposium · 2025
  4. Escaping the Plausibility Trap: A Verified and Abstract Memory FrameworkACM CIKM 2025 Workshop TrustKDD · workshop · 2025
  5. LabMemo: Concept-Level Memory for Autonomous Scientific DiscoveryIEEE IROS 2025 Workshop AIR4S · workshop · 2025

What the alternatives do, and what AIXC adds

Four categories, without brand names.

Fatigue post-processors on finite elements (desktop licence)
What they do wellMaturity, material libraries, integration with major solvers.
What AIXC addsOne dossier per case, traceable and countersigned; the client needs neither licences nor an in-house FEA team.
Commercial DEM
What they do wellGPU, graphical interfaces, CFD coupling.
What AIXC addsMaterial calibration against the published cone test of Li et al. (2021), declared in the dossier.
AI surrogates trained on simulations (SaaS)
What they do wellDesign evaluation speed.
What AIXC addsAIXC does not replace the solver: it runs it, verifies it and logs the execution in the dossier.
Traditional calculation consultancy
What they do wellEngineer's judgement and sector knowledge.
What AIXC addsThe same judgement, with full traceability and an AI Scientist that plans, searches the literature and audits.

Who signs, who leads

Enrique Zueco

Cofounder and Chief Technology Officer (CTO), COIIAR no. 2653

Thirty years in finite elements. Designs the DEM and finite-element models, reviews the results and attests every dossier; none is released without his signature, except samples, which are published as UNSIGNED. Co-author of AIXC Research papers (see Publications).

David Scott Lewis

Cofounder and Chief Executive Officer (CEO)

Nearly five decades in technology and innovation; AI researcher since the 1990s. Former EVP at Meta, with senior leadership roles at Microsoft, Samsung and Oracle.

The chartered industrial engineer's signature is a professional attestation, not a qualified electronic signature.

Frequently asked questions

  • Who signs the dossier?

    A person, not the AI. The AI Scientist plans, extracts, computes, verifies and drafts; the signatory checks the results against their files and attests the dossier. Until then the dossier is UNSIGNED; afterwards it becomes SIGNED or REJECTED. The signature is a professional attestation, not a qualified electronic signature.

  • What is verified?

    Every claim carries its evidence: the results file that produced it, with its hash, and the check applied. The granular material is checked against a published slump test (+7 % / −2 %), the finite elements against an exact solution (< 0.3 %) and the wear model against a steel that was not in the fit (4.8 %).

  • What is not claimed?

    The notes are calculated on generic geometries and do not replace the standard for each part. Today each engine is checked on its own; feeding the DEM load directly into fatigue is under development. ACHT and VAP are architectures of the AIXC Research platform: we do not claim they were used in these cases. Product certification is issued by your certification body.

  • What about data confidentiality?

    The drawings and data of a case are used only for that case and are not shared with third parties. We can sign a non-disclosure agreement before you send anything. A dossier is published only with the written permission of the data owner; the sample dossier uses a generic geometry.

Propose a case

Describe your materials question in an email: the load of a bulk material, the fatigue of a weld, the wear of a plate or a clay blend. We will tell you whether it can become a case study and with which data.

Write to info@aixcresearch.com