IndustryHub
AI

AI — Artificial intelligence in industry

PILLAR · INDUSTRIAL AI

AI — Artificial intelligence in industry

AI is changing how industrial sites are designed, operated and maintained: vision for quality, predictive maintenance, engineering copilots, energy optimization. But a plant is not a web app — real time, safety, scarce and noisy data, and a new regulatory frame (EU AI Act, ISO/IEC 42001) change the rules.

Domains of industrial AI

Why a plant is not a web app

  • Scarce and noisy data — failures are (thankfully) rare: few labelled examples, imbalanced classes, drifting sensors. Data quality beats model choice.
  • Real time & edge — inference on the line is measured in milliseconds; training lives in the cloud, inference at the edge (IPC, embedded GPU), close to the process.
  • Safety & explainability — a model does not decide an emergency shutdown alone: AI advises, the safety function (SIS) stays deterministic and certified. Keep a human in the loop.
  • Drift & MLOps — a model ages: wear, seasons, material change. Drift monitoring, retraining and versioning of data and models are essential.
  • Cybersecurity & sovereignty — connecting OT to cloud models opens an attack surface; segmentation, secure remote access and EU-hosted data (IEC 62443, GDPR).

Regulation & standards

  • EU AI Act — Regulation (EU) 2024/1689: first comprehensive, risk-based frame (unacceptable / high / limited / minimal), phased application 2025-2027, rules for general-purpose models.
  • ISO/IEC 42001:2023 — AIMS — AI management system. The ISO 27001 equivalent for AI: policy, roles, risk management and continual improvement.
  • ISO/IEC 23894 & 22989 — AI-specific risk management (23894) and foundational concepts/vocabulary (22989).
  • NIST AI RMF 1.0 — US voluntary framework: Govern / Map / Measure / Manage functions. A practical complement to the EU AI Act.

Related standard page on IndustryHub

Platforms & players

Industrial vendors

Siemens (Industrial AI, Senseye), Schneider (EcoStruxure AI), ABB (Genix), AVEVA, Honeywell Forge, Rockwell.

Hyperscalers

AWS (IoT TwinMaker, SageMaker), Microsoft (Azure AI, Copilot), Google Cloud Manufacturing.

Vision & edge

Cognex, Keyence, NVIDIA (Jetson, Metropolis), Landing AI, Intel OpenVINO.

MLOps platforms

Databricks, Seeq, TwinThread, Dataiku, MLflow.

Landmark facts

MilestoneYearWhy it matters
AlexNet — deep-learning breakthrough2012AlexNet's ImageNet win made deep-learning vision viable — the origin of modern industrial quality inspection.
Transformers ("Attention is all you need")2017The architecture behind today's large language models and engineering copilots.
ISO/IEC 42001 & generative copilots2023First AI management standard; generative copilots enter engineering workflows.
EU AI Act2024First comprehensive, risk-based AI law. In force 2024, phased application through 2027.

Go further