SCADA, Supervisory Control and Data Acquisition, is the software layer that lets operators monitor and control an industrial process from screens: it collects live readings from PLCs and remote equipment, displays them, raises alarms, logs history, and sends operator commands back down to the machines.
The name says the whole job. Data acquisition: poll every controller and instrument for current values. Supervisory control: let a human watch the whole process and intervene, change a setpoint, start a pump, acknowledge an alarm, without standing at the equipment. SCADA does not make the millisecond control decisions; that is the PLC's job. SCADA supervises.
SCADA vs HMI vs PLC vs MES: which does what?
These four get tangled constantly. The untangling:
- PLC controls the machine. A ruggedized computer running logic in milliseconds: if the sensor says X, fire the actuator. It works whether or not anyone is watching.
- HMI, one window into one machine. The touchscreen on the panel. Often one operator, one machine or cell. Every SCADA system includes HMIs; not every HMI is part of a SCADA system.
- SCADA, supervises many controllers at once. Plant-wide (or region-wide) monitoring, alarming, trending, and command across dozens or thousands of PLCs and RTUs.
- Historian, the memory. A time-series database, usually bundled with or beside SCADA, logging every tag so you can ask what temperature the oven was at 2:14 a.m. last Tuesday.
- MES manages the production business. Orders, routings, genealogy, quality: what should be made, what was made, was it good. MES cares about the order; SCADA cares about the process making it.
A useful mental model: the PLC is the reflex, the HMI is the gauge cluster, SCADA is the control room, the historian is the logbook, and MES is the production office.
What does a SCADA architecture look like?
Four things worth noticing. First, SCADA earns its keep at scale and distance it was born in industries like water, pipelines, and power, where equipment is spread across miles and RTUs (remote terminal units) report back over radio or cellular; in a single plant, the same pattern supervises many lines from one room. Second, the alarm engine is the safety-critical part a flood of badly configured alarms is a known cause of missed real events, which is why alarm rationalization is a discipline of its own. Third, the historian quietly becomes the most valuable component: years of second-by-second process history is the raw material for every later analytics, machine monitoring and digital twin effort. Fourth, redundancy is standard SCADA servers typically run in failover pairs because losing visibility of a live process is not acceptable.
Where does SCADA sit in the control pyramid?
The industry's shared map is the Purdue model, formalized in the ISA-95 / IEC 62264 standard for enterprise-control integration. SCADA and HMIs occupy level 2, above the PLCs, below the operations systems.
The pyramid explains most integration behavior you will see on a floor: each level summarizes upward and commands downward, with reaction times stretching from milliseconds at level 1 to days at level 4. It is also the reference frame for industrial network security, the ISA/IEC 62443 standards build their zones-and-conduits security architecture on these same levels, and the enduring rule is that control layers stay segmented from office networks and the internet. More on that in our IIoT deep dive.
How do you evaluate or modernize a SCADA setup?
- Inventory the tags. What controllers exist, what tags they expose, what is actually polled, and what naming convention (if any) holds it together.
- Check the historian. Is history being logged at useful resolution, for how long, and can anyone besides the controls engineer query it?
- Rationalize alarms. Count alarms per operator-hour. If operators acknowledge without reading, the alarm system is training them to miss the real one.
- Map who consumes the data. If process data stops at the control-room screen, the plant is sitting on an unread archive, supervisors, quality, maintenance, and planning all have questions the historian can answer.
- Open a read-only path upward. Modern analytics and AI layers should read from SCADA and the historian without write access to controls, data flows up, commands stay in the control room.
- Segment before you connect. Any new upward connection follows the Purdue/62443 pattern: through a firewall, read-mostly, inventoried, no default credentials.
Where does an AI layer fit with SCADA?
Above it, reading from it, not replacing it. SCADA and the PLCs beneath it are excellent at their jobs: deterministic control and real-time supervision. What they do not do is context and action beyond the process: SCADA knows the filler stopped and which interlock tripped; it does not know which order was running, what the stop cost, whether the same pattern hit last month, or who should do what about it. That is the layer above, where process data from SCADA and the historian gets joined with orders, quality records, downtime reasons, and the knowledge in operators' heads, and where software can draft the work order or flag the pattern instead of just displaying the tag. That is precisely where Harmony operates: it connects to PLCs, sensors, and existing systems, computes true OEE from source signals rather than estimates, and layers search, dashboards, and approvable AI actions on top, with the control layer untouched. No rip-and-replace (see the connected systems module).
The stat that matters here is architectural, not promotional: the standards bodies themselves, ISA-95 for integration, ISA/IEC 62443 for security, define exactly this pattern of layered, segmented data flow (ISA). An AI layer that respects the pyramid inherits fifty years of hard-won operational discipline; one that bypasses it inherits the outages.
Where did SCADA come from?
Supervisory control predates the acronym: utilities were remotely supervising substations with relay-and-tone telemetry by mid-century, and the modern pattern solidified in the 1960s–70s as minicomputers replaced hardwired supervisory panels and the newly invented PLC (1969) gave SCADA something standard to talk to at the machine level. The architecture has survived every technology wave since, mainframes to minis to PCs to virtualization, because the job it does, letting few humans safely supervise many machines, has not changed. For how SCADA fits the broader modernization picture, see smart factory technology.