Skip to main content
Lumiar
Home/Services/Industrial AI
SERVICE

Industrial AI

AI creates new opportunities for better decisions, efficient operations, and more effective use of industrial data. When solutions are deployed in or near control systems, these opportunities must be assessed alongside architecture, accountability, safety, and consequences for physical operations.

Where can AI create value?

In industry, AI can support process analysis, condition monitoring, maintenance, technical documentation, and decision support. What matters is not the model alone, but whether the solution addresses a relevant problem, uses appropriate data, and works with the plant and its operational processes.

Do any of these sound familiar?

You want to use AI on operational data, but are unsure who owns the risk if the model is wrong.
Results from AI are to be written back to the control system, and you need a secure architecture for it.
Management wants speed on AI, while the OT and security teams want control.
Leaders and specialists need a shared understanding of what AI can and cannot do in industry.

What we do

Lumiar provides independent assessments of AI applications, technical architecture, integration, risk, and verification. Our advice starts with what the organization aims to achieve and the consequences solutions may have in real operations.

01Assessment of AI use that can affect operations, process, or safety
02Architecture and requirements for data flow between control systems, data platforms and AI services, including when results are written back. We build on NAMUR NOA and IEC PAS 62443-1-6.
03Barriers and limits: what AI is allowed to do, and what must always remain in independent safety functions
04Training for leaders and technical teams
INDUSTRIELL KI · NAMUR NE 178
Tilbakeskriving fra KI i NAMUR Open Architecture (NOA)

KI kan lese data fritt, men en endring når kontrollsystemet bare som en forespørsel som er kontrollert i flere steg.

Sentral M+Osentral overvåkingog optimaliseringpsM+Oanleggsspesifikk overvåkingog optimaliseringCPCkjerneprosesskontrollSky-KI · modeller på tvers avanleggAnalyse · digital tvilling · KIog MLSikkerhetsgateway eller diode · NE 177PLS · DCSFeltutstyrdatavia psM+OSIS ogsikkerhetsrelatertesløyferTilbakeskriving når aldriSIS ellersikkerhetsrelatertesløyfer.VoR · NAMUR NE 1781 Autentisering og autorisasjonHvem sender, og er forespørselen uendret?2 Verifisering av forespørselRimelig, gjennomførbar, akseptabel og trygg?3 MappingHvilke CPC-verdier berøres? Endrer ingenting.4 PropageringInn til CPC. Skjer alltid, og før aksept.5 AkseptPasser endringen med prosessens tilstand nå?6 Verifisering av mappingInnenfor grenser og stabilt? Inne i CPC.7 Endringen tas i brukI sin helhet, eller forkastes i sin helhet.forespørsel· Alle steg er sporbare· Begrenset tilbakemelding til avsender· Avslører ikke driftsstatus
Data ut (NE 177)CPCPLS · DCSSikkerhetsgateway eller diodepsM+O · SENTRAL M+OAnalyse · digital tvilling · KI og ML ·Sky-KI · modeller på tvers av anleggForespørsel inn (NE 178)Forespørsler fra sentral M+O går via psM+O.Det finnes ingen direkte vei til CPC.1Autentisering og autorisasjonHvem sender, og er forespørselen uendret?2Verifisering av forespørselRimelig, gjennomførbar, akseptabel ogtrygg?3MappingHvilke CPC-verdier berøres? Endreringenting.4PropageringInn til CPC. Skjer alltid, og før aksept.5AkseptPasser endringen med prosessens tilstandnå?6Verifisering av mappingInnenfor grenser og stabilt? Inne i CPC.7Endringen tas i brukI sin helhet, eller forkastes i sin helhet.Alle steg er sporbare · Begrensettilbakemelding til avsender · Avslører ikkedriftsstatus.SIS og sikkerhetsrelaterte sløyferTilbakeskriving når aldri SIS ellersikkerhetsrelaterte sløyfer.

Basert på NAMUR-anbefalingene NE 175, NE 177 og NE 178 (versjon 2025-03-10). Optimalisering kan gi verdi, men sikker og pålitelig drift skal aldri avhenge av den.

Lumiar-synteseLumiar
INDUSTRIAL AI · NAMUR NE 178

Write-back from AI in NAMUR Open Architecture (NOA)

AI can read data freely, but a change reaches the control system only as a request that is checked in several steps.

Central M+Ocentral monitoringand optimizationpsM+Oplant-specific monitoringand optimizationCPCcore process controlCloud AI · cross-site modelsAnalytics · digital twin · AI andMLSecurity gateway or diode · NE 177PLC · DCSField devicesdatavia psM+OSIS and safety-relatedloopsWrite-back never reachesthe SIS or safety-relatedloops.VoR · NAMUR NE 1781 Authentication and authorizationWho is sending, and is it unchanged?2 Request verificationReasonable, feasible, acceptable, and safe?3 MappingWhich CPC values? Changes nothing yet.4 PropagationInto CPC. Always, and before acceptance.5 AcceptanceDoes it fit the current process state?6 Mapping verificationWithin limits and stable? Inside CPC.7 Change appliedIn full, or discarded in full.request· Every step is traceable· Limited feedback to the sender· Doesn’t reveal operating status
Data out (NE 177)CPCPLC · DCSSecurity gateway or diodepsM+O · CENTRAL M+OAnalytics · digital twin · AI and ML ·Cloud AI · cross-site modelsRequest in (NE 178)Requests from central M+O go via psM+O.There is no direct path to CPC.1Authentication and authorizationWho is sending, and is it unchanged?2Request verificationReasonable, feasible, acceptable, and safe?3MappingWhich CPC values? Changes nothing yet.4PropagationInto CPC. Always, and before acceptance.5AcceptanceDoes it fit the current process state?6Mapping verificationWithin limits and stable? Inside CPC.7Change appliedIn full, or discarded in full.Every step is traceable · Limited feedbackto the sender · Doesn’t reveal operatingstatus.SIS and safety-related loopsWrite-back never reaches the SIS orsafety-related loops.

Based on the NAMUR recommendations NE 175, NE 177, and NE 178 (version 2025-03-10). Optimization can add value, but safe and reliable operation must never depend on it.

Lumiar synthesisLumiar
INDUSTRIELL KI · MYNDIGHET
Evne, myndighet og konsekvens

Risikoen avhenger mer av hvilken myndighet KI-en har over beslutninger med konsekvens enn av hvor avansert den er.

KI-evne: anbefale, optimalisere, lage forespørsel, tilpasse seg innenfor grenser.Evne gir ikke myndighet.Lett styringAutomasjonsstyringBeslutningssikringAssurance på ingeniørnivå oguavhengig beskyttelseInformereAnbefaleGodkjenne og utføreAutonom handlingLavBetydeligSafety ogstorulykkeMyndighet →Konsekvenspotensial →Menneskelig godkjenning er bare en barriere hvis den som godkjenner hartid og grunnlag til å vurdere forslaget.
KI-evne: anbefale, optimalisere, lageforespørsel, tilpasse seg innenforgrenser.Evne gir ikke myndighet.Myndighet →Informere → Anbefale → Godkjenne og utføre →Autonom handlingStyringsnivåLav konsekvens, lav myndighetLett styringLav konsekvens, høy myndighetAutomasjonsstyringHøy konsekvens, lav myndighetBeslutningssikringHøy konsekvens, høy myndighetAssurance på ingeniørnivå og uavhengigbeskyttelseMenneskelig godkjenning er bare en barrierehvis den som godkjenner har tid og grunnlagtil å vurdere forslaget.

Merknaden om godkjenning bygger på NAMUR NE 178.

Lumiar-synteseLumiar
INDUSTRIAL AI · AUTHORITY

Capability, authority, and consequence

The risk depends more on how much authority the AI has over consequential decisions than on how advanced it is.

AI capability: recommend, optimize, create requests, adapt within limits.Capability doesn’t grant authority.Light governanceAutomation governanceDecision assuranceEngineering-grade assuranceand independent protectionInformRecommendApprove and executeAutonomous actionLowSignificantSafety andmajor accidentAuthority →Consequence potential →Human approval is only a safeguard if the approver has the time andinformation to assess the proposal.
AI capability: recommend, optimize,create requests, adapt within limits.Capability doesn’t grant authority.Authority →Inform → Recommend → Approve and execute →Autonomous actionGovernance levelLow consequence, low authorityLight governanceLow consequence, high authorityAutomation governanceHigh consequence, low authorityDecision assuranceHigh consequence, high authorityEngineering-grade assurance andindependent protectionHuman approval is only a safeguard if theapprover has the time and information toassess the proposal.

The note on approval is based on NAMUR NE 178.

Lumiar synthesisLumiar

Methodological basis

NAMUR NE 175, NE 177, and NE 178 (NOA)
IEC PAS 62443-1-6 (IIoT and cloud services)
IEC 62443-3-2
IEC 61511 (independent safety functions)

Experience

Over several years, we have advised on the use of AI calculations on operational data and writing the results back to the control systems at an operator on the Norwegian continental shelf.

Courses and meeting places

OPEN COURSE · TEKNA (IN NORWEGIAN)
AI i industrien – grunnleggende bruk og muligheter
Open →
CONFERENCE · 25–NOVEMBER 26, 2026
KI-dagene 2026
Kenneth Titlestad is program lead and host of KI-dagene 2026, November 25–26 at Oslo Kongressenter, organized by Phenix Events.
Open →

Questions and answers

Is this general AI consulting?
No. We work with AI where it affects operations, process, and critical infrastructure, and where safety and barriers must be maintained.
Do you recommend specific AI tools or platforms?
No. We are vendor-independent and do not sell technology.

Is AI coming into your operations?

Tell us briefly about your needs, and we’ll take it from there.

Talk to us about AI in operations
OTHER SERVICES
Independent review and assuranceStrategy and technical leadershipRisk and barriersArchitecture and engineering assuranceAssurance, verification, and complianceCapability, training, and exercises