About Industrial Twin Lab

Status: Research / Experimental

Industrial Twin Lab is a living technical manifesto, research publication, architecture atlas, and interactive concept demonstrator for industrial machine intelligence. It begins with a strict proposition: an AI system should perform its first experiment in a validated digital environment, not on an operational machine. The publication is authored and maintained in the public aserdargun/itl-aserdargun-com source repository (opens in a new tab).

It is not a commercial SaaS workflow, a control system, a safety system, a production-data service, or a claim that the P-101 fixture describes a real plant. Phase 1 has no operational-technology connection, backend API, account system, model-training runtime, or automated deployment authority. P-101 and all displayed experiment values are fictional, deterministic teaching fixtures.

The 4 September 2026 editorial review aligns the research map with current digital-twin-system, industrial-agent, simulation-interoperability, data-access, and AI-governance developments. It distinguishes the EU AI Act's general application date from the amended schedule for high-risk provisions and references Modelica 3.7 as its language specification. The Phase 1 runtime and authority boundary remain unchanged: static publication, deterministic local fixture, no backend, and no control path.

Place in the AI Learning System

ITL is the English-language industrial-twin research publication within the aserdargun.com AI Learning System (opens in a new tab). The portfolio provides Turkish and English descriptions; the ITL articles and controls are currently in English. Its Physical AI learning path connects world-model and swarm research to industrial twins and humanoid engineering. These are learning relationships between independent applications.

The 21 September 2026 portfolio review distinguishes three learning experiences: ITL for research questions and authored evidence fixtures, PDT for pump anatomy, and DTR for computed synthetic experiments. Sharing the P-101 teaching name does not establish a common model, telemetry stream, or compatible replay format.

Companion laboratories

computes a fictional pumping process, separate observations, and an independent twin in the browser. It compares three alternatives and separates candidate review, approval, application to the simulation, and prediction–outcome comparison. Its intelligence layer uses deterministic logic, without a runtime AI model. DTR's JSON export is a separate experiment record and cannot be imported into ITL's replay demonstrator. Neither application grants field control authority.

The companion PDT — P-101 Interactive Digital Twin (opens in a new tab) application provides an interactive 3D centrifugal-pump teaching exhibit with component inspection, sensor views, and synthetic fault illustrations. It complements ITL’s research and experiment narrative with spatial exploration. PDT is a separate educational application, not completion of the runtime, data-import, or industrial-connector phases below.

Roadmap

1. Phase 1 — Manifesto + Architecture Atlas: establish the thesis, safety boundary, typed knowledge layer, P-101 fixture, and conceptual demonstrator.
2. Phase 2 — Interactive Twin Capsule: edit and inspect bounded asset representations.
3. Phase 3 — Synthetic Experiment Workbench: configure controlled, explicitly synthetic investigations.
4. Phase 4 — Real Dataset Import: introduce governed, traceable dataset ingestion.
5. Phase 5 — Python Experiment Runtime: execute reproducible analytical workloads outside OT control.
6. Phase 6 — MLflow / Experiment Registry: preserve experiment and model lineage.
7. Phase 7 — FMU / Modelica Simulation: connect validated simulation artifacts through explicit interfaces.
8. Phase 8 — Local LLM AI Scientist: orchestrate tools and evidence locally without control authority.
9. Phase 9 — Industrial Connectors: add controlled industrial data access under a separate security design.
10. Phase 10 — Fleet Intelligence: test portability and organizational learning across assets.

The roadmap documents extension points; it is not hidden partial implementation or a promise that every phase is appropriate for every plant. Any control-facing capability requires its own hazard analysis, security boundary, validation case, approval process, and accountable human authority.