From SCADA to edge, without the rip-and-replace.

Most SCADA estates are older, deeper and more entangled than the documentation suggests. Before any migration plan, map the truth: every PLC, every screen, every protocol and every analog cable that keeps the plant running.

• Asset inventory: PLCs, RTUs, HMIs, drives, gateways — with firmware versions and end-of-life dates.
• Protocol survey: Modbus, Profibus/Profinet, EtherNet/IP, DNP3, OPC Classic — including non-routable links.
• Tag census: how many points are actively scanned, how many are stale, how many are duplicated across servers.
• Operator interview round: which screens are used hourly, which are decorative, which are silently broken.

A modern edge architecture lives or dies on its data model. Standardise an ISA-95 hierarchy and a unified namespace (UNS) on paper first — migrating tags into a clean model is far cheaper than migrating them twice.

• Adopt an ISA-95 site → area → line → cell → asset hierarchy as the canonical topology.
• Define a Unified Namespace that exposes context, not just raw tags (e.g. site/area/line/asset/metric).
• Pick a semantic model — OPC UA Companion Specs, Sparkplug, or vendor-neutral JSON — and enforce it.
• Agree units, engineering ranges and quality flags as part of the tag contract, not as comments in code.

The lowest-risk migration runs the new edge alongside the existing SCADA, not in place of it. Edge gateways subscribe to the same PLCs, publish into the new UNS, and prove themselves for months before any control hand-over.

• Deploy edge gateways (Ignition Edge, HiveMQ Edge, Litmus, AWS Greengrass or equivalent) read-only first.
• Bridge legacy PLCs via OPC UA where available; fall back to native Modbus/EtherNet/IP drivers otherwise.
• Publish to an MQTT broker using Sparkplug B for store-and-forward, birth/death and quality semantics.
• Run the edge for at least one full production cycle before any downstream system depends on it.

A big-bang SCADA replacement is the textbook way to bring a plant to its knees. Run old and new in parallel, migrate one area or line at a time, and keep the legacy HMI authoritative until the new stack has earned trust.

• Choose one non-critical line as the pilot — ideally with a recent control upgrade and good documentation.
• Mirror critical screens on the new HMI/visualisation layer; operators decide when to switch primary view.
• Keep historian dual-feed during transition so reports remain comparable across the cutover window.
• Define explicit rollback criteria: latency, packet loss, alarm parity, operator confidence score.

Once the UNS is the source of truth, the value cases your old SCADA could never serve become straightforward: condition monitoring, OEE, energy attribution, batch genealogy and ML-driven anomaly detection.

• Stream UNS topics into a time-series store (InfluxDB, TimescaleDB, Snowflake, Fabric) with retention tiers.
• Push events to CMMS (SAP PM, Maximo) and ERP via an event bus rather than point-to-point integrations.
• Stand up condition-monitoring and OEE dashboards on the new layer — leave the legacy SCADA for control.
• Treat ML models as first-class deployable artefacts with versioning, rollback and drift monitoring.

Edge gateways, MQTT brokers and cloud bridges materially expand the attack surface. IEC 62443 zones and conduits, signed firmware and least-privilege access have to be designed in — bolting them on after go-live is a project of its own.

• Define IEC 62443 zones across Levels 2–3.5 with documented conduits and protocol whitelists.
• Terminate every north-south flow at an industrial DMZ; no direct PLC-to-cloud connections.
• Use mTLS and per-device certificates for MQTT clients; rotate via an internal PKI, not shared secrets.
• Mirror OT events into the corporate SIEM — Splunk, Sentinel, Elastic — with OT-aware detections.

The migration ends not when the old SCADA is decommissioned, but when an empowered platform team owns the new stack: roadmap, SLAs, change control, training and a clear backlog of value cases.

• Named platform owner sitting across IT, OT and operations — with budget authority and a backlog.
• Observability across edge gateways, brokers and historians: uptime, lag, message loss, queue depth.
• Change control covers tag additions, dashboard publishing and ML model promotion — not just PLC code.
• Training plan for control engineers, reliability and operations — analytics tools, not just dashboards.