A factory is five layers, stacked from the spinning, heating, flowing physical process at the bottom to the boardroom at the top. What makes the stack work is the traffic between the layers: a single order flows down and turns into motion, while a single measurement flows up and turns into a business decision. In this lesson you watch both journeys happen, hop by hop, and see why the time-scale stretches from milliseconds to months as you climb.
The ISA-95 model (and its cousin, the Purdue model) stacks a plant into levels. Each level speaks only to its neighbours, never end-to-end, and each one works on a very different time-scale. Here is the cast, bottom to top, in the simulator below.
⬇️ Going down, an order decomposes. "Make 600 units" becomes a dispatch, becomes setpoints, becomes I/O bits, becomes a heater turning on. Each level translates the one above into something the one below can act on.
⬆️ Going up, data aggregates. A photo-eye pulse becomes a count, becomes a rate, becomes an OEE figure, becomes "we'll ship on time." Each level summarises the raw reality below into something the one above cares about.
Send an order down and watch it turn into something more concrete at every hop, or send a signal up and watch raw field data distil into a business KPI. The travelling token carries the message. Each level lights up as it arrives and shows what the value looks like there, with the latency that level adds. Tick the time-horizon box to see the scale stretch from milliseconds to months.
Everything in the pyramid is one of these two flows, and once you have them the rest of the model follows:
⬇️ Decomposition (top → bottom)
A vague, slow, high-level intent ("fulfil this order by Friday") is broken into ever more specific, ever faster instructions until, at the bottom, it is a single output bit flipping a contactor. Abstraction falls as you descend.
⬆️ Aggregation (bottom → top)
A flood of fast, raw, low-level signals is summarised into fewer, slower, more meaningful numbers until, at the top, it is a single sentence a manager reads. Abstraction rises as you ascend.
🧭 The golden rule: each level talks only to its neighbours. ERP never pokes a PLC directly; the field never calls the boardroom. Skipping levels leaves the hole that attackers and spaghetti integrations crawl through. That is why the IT/OT boundary and MES exist.
If you remember one thing from this section, make it this: everything changes scale as you climb. Time horizons lengthen and data volume shrinks, while the decision-maker shifts from a microchip to a manager.
| Level | Lives here | Time horizon | Data & decision |
|---|---|---|---|
| L4 · ERP | SAP, finance, planning | Days → months | Low volume, high meaning · humans & planning |
| L3 · MES | Execution, OEE, genealogy | Minutes → shift | Orders & KPIs · software + supervisors |
| L2 · SCADA/HMI | Screens, alarms, trends | Seconds | Live tags · operators |
| L1 · Control | PLCs, the logic | Milliseconds | I/O & logic · the controller |
| L0 · Field | Sensors, motors, valves | Now (physics) | Raw signals · the physical world |
This is why you don't run motion logic in the ERP and don't do monthly planning in a PLC: each tool is built for its own time-scale. Put work at the wrong level and it's either dangerously slow or hopelessly over-complex.
You'll hear both names. They're the same stack seen by different crowds:
📊 ISA-95 is the functional / data view, the one the MES and IT crowd prefer. It's about how levels exchange information (orders, schedules, results) and underpins MES↔ERP integration (B2MML).
🛡️ Purdue is the network / security view, the one the OT-security crowd prefers. It has the same levels but adds the IT/OT boundary and the DMZ, and it is the basis for segmentation and IEC 62443.
🧩 Same five levels. ISA-95 asks "what data moves between them?"; Purdue asks "where are the network walls?" Learn one and you understand the other.