SCADA vs HMI: Architecture, Data Flow, and When to Use Each

01 — The One-Line Difference

An HMI (Human-Machine Interface) is a local display that shows the state of one machine or process. A SCADA (Supervisory Control and Data Acquisition) system collects data from many remote devices, stores it historically, and allows supervisory control across an entire site or network of sites.

The confusion exists because modern software packages blur this line — Ignition, iFIX, and Wonderware all market themselves as "SCADA/HMI platforms." But the underlying architecture is distinct, and getting it wrong leads to systems that can't scale, can't historian, or can't meet utility interconnect requirements.

02 — What an HMI Actually Is

An HMI is a touchscreen or panel display connected directly — usually via serial (RS-232/485) or Ethernet — to a single PLC, drive, or controller. It reads registers directly from that device and displays them as animated graphics, trend charts, and alarm lists.

HMI Characteristics

• Local connection only — typically within the same panel or within 1000 feet via RS-485
• No historian — data lives in the PLC registers; when power cycles, it's gone unless the PLC has SD card logging
• Protocol: Modbus RTU, Modbus TCP, or proprietary (Allen-Bradley DF1, Siemens MPI)
• Latency: sub-second — HMI polls the PLC register directly, usually every 250–500ms
• Security: usually none by design — HMIs typically sit on the same plant LAN as the PLC they read, often with no authentication on the local screen

03 — What SCADA Actually Is

SCADA is a supervisory architecture — a system of systems. It sits above the field devices (PLCs, RTUs, inverters, meters) and collects data from all of them into a centralized server that provides historian storage, alarm management, reporting, and remote control capability.

SCADA Characteristics

• Multi-device, multi-site — one SCADA system can monitor hundreds of RTUs across a region
• Historian database — time-series data stored at configurable resolution (typically 1s to 1min intervals)
• Protocols: DNP3, IEC 61850, OPC-UA, ICCP — designed for unreliable WAN links
• Alarm management: alarm annunciation, acknowledgment tracking, alarm prioritization
• Remote SCADA (EMS/DMS): utility-grade systems include state estimation and network analysis

04 — Architecture: How They Connect

In a proper industrial architecture, HMIs and SCADA exist at different layers of the ISA-95 / Purdue Model:

• Level 0–1: Field devices — sensors, actuators, drives
• Level 2: PLC / RTU with local HMI panel — machine-level control and display
• Level 3: SCADA / Historian — site-level supervision, data collection, alarming
• Level 4: ERP / Business systems — reporting, KPIs, enterprise integration

The HMI at Level 2 communicates with the PLC directly via serial or local Ethernet. The SCADA at Level 3 communicates with all Level 2 devices via routed Ethernet, cellular, or fiber — using protocols designed for reliability over unreliable links.

05 — Protocols: DNP3, Modbus, OPC

06 — Solar Plant Example: What Uses What

A typical 10 MW utility-scale solar plant uses all three layers simultaneously:

• Each inverter has a local touchscreen HMI — Modbus RTU, shows DC voltage, AC power, fault codes. Scope: one inverter.
• Plant controller (RTAC) collects from all 20 inverters via Modbus TCP/SunSpec, runs reactive power dispatch, and exposes a DNP3 outstation to the utility SCADA master.
• Revenue meter at POI connects to the RTAC via Modbus TCP — P, Q, V, I, PF at the point of interconnection.
• Utility SCADA master (at the control center, miles away) polls the RTAC via DNP3 over a fiber or cellular link — gets all the plant data aggregated into one DNP3 connection.
• Plant owner's SCADA (O&M platform) independently collects from the RTAC via a second DNP3 session or via OPC-UA to the plant historian.

07 — Decision Guide: Which Do You Need?

08 — Common Mistakes

The "HMI/SCADA platform" confusion

What it looks like. Project spec says "Ignition SCADA at the plant." Integrator installs Ignition, builds operator screens, plant goes live. Six months later the O&M team asks for a year-over-year availability report and finds nothing was historized. Inverter faults from three months ago aren't recoverable. The site has an HMI, not a SCADA.

How to find it. Check whether the Ignition tag database has Historical Logging enabled per tag, whether the Tag Historian module is installed and configured, and whether a real SQL or InfluxDB store is connected behind it. If "historized" is set to false on the production tags, you have an HMI configured to look like a SCADA. Same check applies to Wonderware, FactoryTalk, Citect — the platform supports both modes; configuration determines which one you actually have.

How to fix it. Enable historian on every operational tag (P, Q, V, I, breaker status, fault flags). Provision a database backend with enough retention for the contracted reporting period. Build the alarm pipeline. Add a redundant server pair. You are now running a SCADA. Going back and reconstructing history from inverter event logs after the fact is painful, sometimes impossible — configure historization on day one.

Other mistakes worth naming

• Using Modbus TCP for utility-facing data. Modbus has no SOE timestamping, no unsolicited reporting, no WAN error recovery. Utility SCADA requires DNP3 or IEC 61850. If your interconnect agreement says "DNP3" and you proposed Modbus, you will be redoing the work during commissioning testing, not before.
• Single points of failure. A SCADA system with no redundant server, no UPS, and no cellular backup is not a SCADA system. It is an HMI installed in a server room. The utility commissioning team will check for redundancy during the final acceptance test on most projects above 20 MW.
• Bridging owner SCADA to utility SCADA directly. These are separate systems with separate trust boundaries. The plant owner's SCADA reads operational data for O&M; the utility SCADA reads dispatch-relevant data. They never share a network segment without a DMZ firewall and explicit data diodes or one-way feeds. NERC CIP standards on the bulk power system formalize this requirement; even non-CIP plants follow the same pattern.