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
Predictive maintenance
Vibration, thermography, ML signatures: anticipate failure before it happens and move from calendar-based to condition-based.
→Machine vision & quality
2D/3D inspection, deep-learning anomaly detection, industrial OCR and robot guidance on the line.
→Generative AI for engineering
Automation-code copilots, specification and documentation generation, procedure and diagnostic assistants.
→Regulation & governance
EU AI Act, ISO/IEC 42001, NIST AI RMF: classify risk, document, monitor — and keep a human in the loop.
→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
| Milestone | Year | Why it matters |
|---|---|---|
| AlexNet — deep-learning breakthrough | 2012 | AlexNet's ImageNet win made deep-learning vision viable — the origin of modern industrial quality inspection. |
| Transformers ("Attention is all you need") | 2017 | The architecture behind today's large language models and engineering copilots. |
| ISO/IEC 42001 & generative copilots | 2023 | First AI management standard; generative copilots enter engineering workflows. |
| EU AI Act | 2024 | First comprehensive, risk-based AI law. In force 2024, phased application through 2027. |