AI for manufacturing, predictive maintenance, quality, supply visibility.
AI for manufacturing is the design and deployment of custom systems for predictive maintenance, quality inspection, supply visibility, and operations knowledge, with classification against the EU AI Act §6 conformity track wherever an AI component touches workplace safety, and against the Machinery Regulation 2023/1230 where the AI is part of a machinery safety function. Impetora builds these systems for discrete and process manufacturers across the EU. EU manufacturing GDP sits at around €2.4 trillion, and ~25% of EU manufacturers report active AI deployment.
How AI is reshaping manufacturing in 2026
Most manufacturing AI is not high-risk by default. The wins come from sensor-data quality, integration discipline, and clean §6 boundary management, not from autonomy.
Manufacturing is one of the most heterogeneous AI surfaces in this list. A condition-monitoring model on a non-safety-critical asset is a different regulatory animal from a vision system that is part of a machinery safety function, which is a different animal again from supply-chain analytics inside a back-office.
Most manufacturing AI is not high-risk under the EU AI Act by default. The triggers that lift a system to EU AI Act §6 high-risk are integration into a product covered by harmonised legislation (typically machinery, medical devices, or vehicles), employee monitoring, or biometric identification. Add the Machinery Regulation 2023/1230 for safety functions, the Cyber Resilience Act (Regulation 2024/2847) for products with digital elements, and the revised Product Liability Directive 2024 for software-as-product, and the regulatory map is well-defined.
The systems we ship treat the conformity assessment as a first-class deliverable wherever §6 applies, run shadow-mode rollouts before any safety-critical decisioning, and stay back-office-only on the rest.
About 25% of EU manufacturing enterprises report active AI deployment, with the strongest concentration in predictive maintenance and quality inspection.
Use cases we deliver for discrete and process manufacturers, industrial operators, plant teams
Predictive maintenance and condition monitoring
Unplanned downtime is the single biggest cost-of-quality lever in most plants. Vibration, temperature, and current-signature data is collected but underexploited.
Quality inspection and visual defect detection
Manual inspection misses 5 to 15% of defects on high-throughput lines. Vision systems exist but classification quality plateaus on novel defect patterns.
Supply visibility and shortage forecasting
Inbound shipment data lives across email, supplier portals, and EDI. Manual reconciliation is slow and shortage signals arrive too late to act on.
MES and shop-floor document processing
Quality records, batch sheets, calibration certificates, and work orders flow through paper, scans, and PDFs. Re-keying into MES drives errors and delay.
Internal ops and SOP knowledge AI
Operators and maintenance technicians reference SOPs, machine manuals, and process documents that live across SharePoint, Confluence, and binders. Onboarding new operators takes weeks.
Energy and yield optimisation analytics
Energy and yield levers are spread across PLCs, historians, and ERP. Analyst time on cross-source correlation eats into the optimisation budget.
How TRACE applies to manufacturing AI
Trust
Readiness
Architecture
Citations and evidence
Regulatory considerations for manufacturing AI
Manufacturing AI is regulated under multiple overlapping frameworks. Most systems are not high-risk by default. We map every engagement to §6, Machinery, Cyber Resilience, and PLD before code is written.
- 01
EU AI Act §6 - high-risk via harmonised legislation
AI systems integrated into products covered by harmonised legislation (Machinery Regulation, Medical Devices, Toys, Vehicles) are §6 high-risk. Conformity assessment, risk management, technical documentation, and post-market monitoring required.EUR-Lex - 02
Machinery Regulation 2023/1230
Replaces the Machinery Directive. In force January 2027. Explicit treatment of AI components in safety functions, with conformity-assessment routes and CE-marking obligations for self-evolving systems.EUR-Lex - 03
Cyber Resilience Act (Regulation 2024/2847)
In force December 2027 for products with digital elements. AI in industrial products falls in scope. Vulnerability handling, security-by-design, and 24-hour exploitation reporting obligations.EUR-Lex - 04
Product Liability Directive 2024
Revised PLD extends strict liability to software-as-product, including AI. Evidence-disclosure obligations on manufacturers, presumption of defectiveness in certain cases.EUR-Lex - 05
GDPR Articles 6, 9, and 88 - workplace data
Article 88 specific provisions on processing in the employment context. Where AI touches operator performance data, lawful basis, transparency, and works-council consultation matter on top of generic GDPR.GDPR-Info - 06
ISO/IEC 42001 AI management system
ISO 42001 certification track aligns naturally with §6 conformity-assessment evidence and Machinery Regulation documentation requirements. We track 42001 across our engagements; certification is on the roadmap.ISO
How we typically engage
Three phases. Discovery scopes the §6 vs §50 boundary, the Machinery Regulation interaction, and the Cyber Resilience Act posture before any code is written.
- 011 to 2 weeks
Discovery
Workflow audit across line, maintenance, and supply ops. Sensor-data baseline, label-quality review, integration map, written risk classification under §6 and Machinery Regulation.
- 024 to 14 weeks
Build
MES/SCADA/ERP integration, eval suite tied to your asset and product mix, shadow-mode rollout for quality and PdM surfaces, conformity-assessment scaffolding where §6 applies.
- 03Ongoing
Operate
Quarterly drift reports, recalibration tied to product-mix and process changes, post-market monitoring under EU AI Act Article 72, vulnerability handling under the Cyber Resilience Act.
What Impetora does not build
An honest list. These systems we will not build because they breach professional ethics, regulation, or our own risk policy.
Employee-monitoring biometric AI
Safety-critical PdM without §6 conformity
Pricing models with protected-class proxies
Closed-loop control without human override
How a manufacturing AI system flows
The typical value chain from input to audit log. Every node is a reviewable stage with guardrails.
Frequently asked questions
Is manufacturing AI considered high-risk under the EU AI Act?
Most manufacturing AI is not high-risk by default. The §6 trigger is integration into a product covered by harmonised legislation, most commonly the Machinery Regulation 2023/1230 for safety functions. Predictive maintenance on non-safety-critical assets, supply analytics, and back-office knowledge AI are not high-risk by default. We classify every system during discovery and build to the proportionate level the regulation requires.
How do you handle the Machinery Regulation 2023/1230?
Where the AI is part of a machinery safety function, we build to the conformity-assessment route the regulation requires from week one, including the technical documentation pack, the risk-assessment chain, and the CE-marking evidence. Where the AI is adjacent to but not part of a safety function (typical predictive maintenance), we document the boundary in writing so the conformity assessment of the machinery does not pull the AI into scope unnecessarily.
What about the Cyber Resilience Act?
The Cyber Resilience Act (Regulation 2024/2847) is in force December 2027 for products with digital elements, with AI in industrial products in scope. We integrate vulnerability-handling processes, security-by-design controls, and the 24-hour active-exploitation reporting workflow into the system from day one.
How do you handle workplace data under GDPR Article 88?
Article 88 sets specific rules on processing in the employment context. Where AI touches operator performance, throughput, or behavioural data, we document the lawful basis, run a written DPIA, and recommend works-council or staff-representative consultation paths where local law requires it. We do not build covert monitoring AI.
Can the system integrate with our MES, SCADA, and historian?
Yes. We integrate with the major MES platforms (Siemens Opcenter, Rockwell PharmaSuite, Apriso, AVEVA), SCADA stacks, OPC UA-based historians, ERP systems (SAP S/4HANA, Microsoft D365), and CMMS platforms. Idempotent writes, queue-based bridges for legacy systems, and append-only audit logs across the stack.
How accurate is predictive maintenance in production?
Production-grade predictive-maintenance deployments see lead-time-to-failure forecasts that beat scheduled-maintenance baselines by 20 to 40% on stable asset classes after the first 8 to 12 weeks of evaluation tuning. Accuracy depends on sensor quality, failure-class diversity, and label completeness. We baseline first, target a specific delta against your current process, and report against it monthly through the pilot.
What is the typical engagement scope and timeline?
First engagements target one workflow with a measurable baseline, run 4 to 14 weeks to production, and ship as a single signed-off system inside one MES or asset surface. Common scopes are: predictive maintenance across one to two asset classes; quality inspection across one to two product lines; supply visibility across one to two supplier categories. Submit a project with the workflow you have in mind.
What does a manufacturing AI engagement cost?
Pricing is set after the discovery sprint, against your specific workflow, asset count, and integration surface. Manufacturing AI engagements range from compact knowledge AI deployments to §6 conformity-assessment-grade safety-adjacent builds, so we do not publish a flat rate. Submit a project with the workflow and rough volume.
Considering AI for your manufacturing operation?
Tell us the workflow and asset class you have in mind and we come back within one business day with a discovery proposal.