IT/OT Integration Architecture Guide

Introduction

Plant data is useful when it improves a controlled decision: identify a downtime event, release a production step, generate a maintenance inspection, reconcile material, trace a lot or alert an accountable role. Connecting every tag to the cloud is not an integration strategy.

Table of Contents

1. Start with use cases
2. Map layers and ownership
3. Architecture patterns
4. Security and safety
5. Data context and quality
6. Integration to ERP, MES and CMMS
7. Pilot and operation
8. Vendor evaluation
9. FAQs

Start With a Decision Use Case

Define:

• User and decision/action.
• Asset/process and consequence.
• Source signals and context.
• Required latency and availability.
• Validation/human review.
• Destination workflow.
• Evidence/retention.
• baseline and success.

Example:

When a critical compressor’s validated operating condition crosses a defined engineering threshold for a sustained period, a reliability engineer reviews the event and creates an inspection in the CMMS with the asset, time window and evidence. The objective is actionable warning, not raw-tag collection.

Prioritize by value, feasibility, security and support. Avoid vague “single pane of glass.”

Map the Environment

Inventory:

• Physical assets and control boundaries.
• Sensors/instruments.
• PLC/DCS/SCADA.
• Historians.
• MES/QMS/LIMS.
• CMMS/EAM.
• ERP/WMS/data platform.
• Networks/zones/conduits.
• Edge gateways/remote access.
• Vendors and support lifecycle.
• Identity/time synchronization.
• Protocols and data ownership.

Document criticality, firmware/software, owner, backup/configuration, connectivity and change process. Do not scan or test OT using unapproved IT methods.

Layer and System Ownership

A conceptual model:

• Control: PLC/DCS executes control.
• Supervisory: SCADA/HMI monitors/operates.
• Operations: historian/MES/quality/maintenance context.
• Enterprise: ERP, business applications, analytics.

Use ISA-95/IEC 62264 concepts where helpful, but fit the actual architecture.

Assign:

• Machine/line/work-center mapping.
• Production order/product/lot context.
• Asset ID.
• Downtime event.
• Maintenance work.
• Quality result.
• Inventory movement.
• User/shift.

One physical machine may have PLC tag, historian path, MES resource and CMMS asset IDs. Govern the mapping.

Architecture Patterns

Direct point-to-point

Fast for one interface but creates brittle coupling at scale. Use only with clear lifecycle/monitoring.

Edge gateway

Collects/translates/buffers at plant boundary, can support store-and-forward and protocol isolation. It becomes critical infrastructure needing hardening, monitoring and support.

Historian/data platform

Historian preserves industrial time-series; enterprise data platform combines context for analytics. Do not make a lake the control system.

Event/message integration

Publishes meaningful events to consumers, reducing tight coupling. Requires schemas, identity, delivery semantics and error handling.

MES as context layer

MES connects production order/operator/material/equipment context and exchanges summarized transactions with ERP/control. Not every factory needs full MES.

Choose based on latency, resilience, protocol, security, volume, context and support—not trend.

Security and Safety

Follow current organizational policy and recognized primary guidance such as the NIST Guide to Operational Technology Security (verify current revision at publication).

Core practices may include:

• Asset/network inventory.
• Zones/conduits and segmentation.
• Least privilege and separate admin accounts.
• Controlled remote/vendor access with MFA and monitoring.
• Allowlisted/approved data flows.
• Secure configuration and credential/certificate management.
• Logging/time synchronization.
• Patch/vulnerability process adapted to safety/availability.
• Backup of configurations and tested recovery.
• Incident response coordinated with plant safety.
• Change management and maintenance windows.
• Supplier/component risk.

Never connect business applications directly to control devices without approved architecture. Analytics output should not autonomously change high-consequence control unless formally engineered and validated.

Data Context and Quality

Raw tag T_102 is meaningless without asset, unit, location, sensor quality, sampling, timestamp, operating state and calibration.

Define:

• Semantic name and unit.
• Source and quality/status.
• Timestamp/time zone/synchronization.
• Asset/component mapping.
• Operating/product/order/shift context.
• Sampling/aggregation.
• Missing/outlier handling.
• Retention.
• owner and change version.

Monitor sensor drift, data gaps, duplicates and topology changes. Preserve lineage from dashboard/model to source.

ERP, MES, CMMS and Dashboard Integration

ERP

Send business transactions or summarized confirmed events—not uncontrolled raw tags. ERP may provide order/item/resource context and receive production/material/cost status.

MES

Can consume equipment states and connect them to execution, quality and genealogy. Define boundary with SCADA and ERP.

CMMS

Asset/meter/condition/downtime events may inform inspection/work. Confirm event before generating noise, prioritize by risk and capture technician findings back. Titan MMS behavior must be verified.

Dashboard/AI

Design metric/action, not just data. For predictive maintenance, define failure mode, label, warning, reviewer and work response. Monitor false alerts and drift.

Integration Contract

For each flow:

• Producer/consumer and business owner.
• Schema/unit/key.
• Frequency/latency.
• Authentication/encryption.
• Delivery/ordering/duplicate semantics.
• Validation and quality.
• Buffer during outage.
• Retry/dead-letter/manual correction.
• Monitoring/alert/SLAs.
• environments/testing.
• change/version.
• support escalation.

Test normal, source outage, network loss, delayed/out-of-order/duplicate, schema change and recovery. Reconcile business outcomes, not only messages.

Pilot and Scale

1. Select bounded use case and plant area.
2. Conduct safety/security/architecture review.
3. Baseline decision and data.
4. Map asset/context and configure edge/integration.
5. Run shadow monitoring.
6. Validate events with operations/maintenance.
7. Integrate human/action workflow.
8. Measure reliability, alert burden, response and business effect.
9. Document reusable pattern and prerequisites.
10. Scale through governed change windows.

Have rollback and manual fallback. Do not use production control as a sandbox.

Operating Model

Roles across IT, OT controls, security, and plant operations must be clearly defined for monitoring, incidents, patching, and device replacement, with escalation paths consistent with plant risk profiles.

Review interface health and data quality. A successful pilot that nobody supports becomes a hidden operational dependency.

Vendor Evaluation

• Supported protocols and architecture.
• Security evidence and lifecycle.
• Edge/offline buffering.
• Scale/performance.
• Data semantics/context.
• APIs/events and integration tools.
• Monitoring/support.
• Reference use case under similar constraints.
• Product roadmap/end-of-life.
• Data export/portability.
• Implementation partner capability.

Sensitive network and architectural diagrams should never be exposed during early public procurement phases.

Common Mistakes

• Collecting all data without use cases.
• Flattening OT into IT architecture.
• Direct uncontrolled PLC-to-cloud/business links.
• Ignoring time/units/context.
• Using dashboards as control.
• Auto-generating too many work orders.
• No owner for failed messages/certificates.
• Piloting on nonrepresentative lab data only.
• Scaling before security/support.

FAQs

What is IT/OT integration?

Controlled connection of enterprise information systems with industrial operations/control data and workflows.

Does IT/OT convergence mean one network/team?

No. It means coordinated architecture/governance while respecting safety, availability and security boundaries.

Which protocol should be used?

Depends on equipment, latency, semantics, security and support. Specialists should assess; protocol alone does not define architecture.

Can OT data trigger CMMS work?

Yes with supported integration, asset mapping, validation, risk rules and response workflow. Avoid alert floods.

What should go to cloud?

Only data/functions justified by use case and approved security/architecture; control remains within engineered boundaries.

How should a pilot be measured?

Data reliability, event validity, warning/latency, response/action, support burden, security guardrails and business outcome.