ENVIZNIndustrial automation, from the plant floor up
Most automation is not robots. It is millions of largely invisible control loops holding temperatures, pressures, and positions steady across factories worldwide. This track teaches the machine behind the machines.
Read the Purdue stack like a floor plan · explain PID and dead time from feel, not formulas · judge any automation claim with OEE math.
Two worlds: OT and IT
Industrial automation makes physical processes operate safely, repeatedly, and with limited continuous intervention. Discrete manufacturing produces countable items such as vehicles, electronics, and packages. Process manufacturing transforms continuous materials such as chemicals, fuels, metals, and water. PLCs and robots are common in the first; distributed control systems are common in the second. The equipment differs, but both systems measure, decide, and act.
The deeper divide is between operational technology (OT) and information technology (IT). IT environments typically prioritize data, confidentiality, and frequent change. OT environments prioritize safe physical operation and availability, sometimes across equipment that cannot stop casually. A server reboot may interrupt information; a controller reboot may interrupt a process.
Cybersecurity, AI adoption, and data integration often require these two operating cultures to negotiate. The Purdue model, formalized through ISA-95, makes that negotiation visible by organizing the system from millisecond sensors and controls to business planning measured in days or months.
Six layers, from milliseconds to months
Click a layer. The dot shows its timescale: note the logarithmic axis: nine orders of magnitude between the bottom of the stack and the top.
Run a PID loop yourself
Strip away everything and automation reduces to one idea: measure, compare, correct, repeat. A sensor reads the process, the controller compares it to the setpoint, and an actuator pushes back on the error: thousands of times per minute, forever. The algorithm doing the comparing in most of the world's loops is PID, essentially unchanged for a century.
Below is a live simulation of a canonical industrial control problem: a first-order process with dead time. Your action takes effect 1.6 seconds later, much as heat moves through a vessel or material moves down a pipe. Dead time makes control genuinely difficult because every correction is based on old information.
Start in Manual and try holding the level against a drifting load. Then switch to Auto and tune the controller. Set the gain too low and the response becomes sluggish; push it beyond the stability boundary and the loop enters sustained oscillation: the edge Ziegler and Nichols turned into a tuning method in 1942. Your score tracks time in specification and alarm count.
Hold the level
setpoint 70% · In spec , · Alarms 0The words on the floor
Where the robots actually are
Robots are automation's most visible slice, and a rapidly growing one, but still a slice. Four families cover much of the market: articulated arms weld and paint in automotive plants; SCARA and delta robots handle high-speed pick-and-place work; cobots operate beside people within ISO 10218 / TS 15066 safety limits; and AMRs move material through warehouses. The IFR's World Robotics reports place global robot density at roughly 150–160 units per 10,000 manufacturing employees, with about half a million new industrial robots installed annually in recent years. Korea and Singapore sit far above the average, while automotive and electronics remain the dominant buyers.
And the scoreboard. The number a plant manager wakes up to is OEE: three multiplied fractions that expose where production actually went. Set the sliders to any real plant's numbers and watch how fast multiplication punishes mediocrity:
OEE: overall equipment effectiveness
The classic TPM benchmark treats ~85% as world-class for discrete manufacturing; many surveyed plants run in the 40–60% range. Three good-looking numbers multiply into an honest one.
Common misconceptions
Robots are the visible minority. Most automation is control loops you'll never see: valves, drives, and PLC logic holding processes steady. A refinery can be almost fully automated and contain barely a robot.
Fully dark factories exist but remain rare showcases. Real plants run with people everywhere: maintenance, quality, changeovers, exceptions. Automation shifts human work; it hasn't emptied the building.
The priorities invert. IT ranks confidentiality first; OT ranks availability and safety first. Patching, downtime, and remote access all work differently when the computer is holding a chemical reaction steady.
Deterministic, certified, millisecond control isn't a prediction problem. AI is arriving above the loop (optimizing setpoints, predicting failures, scheduling) while the loop itself stays boring on purpose.
Four questions before you go
Where to go next
Sources & methodology
Figures are teaching values, hedged where practice varies. Reference points: the Purdue Reference Model (T. Williams, Purdue University) and ISA-95 (level architecture); IEC 61131-3 (PLC languages); ISA-101 and ISA-18.2 (HMI and alarm management); ISA/IEC 62443 and NIST SP 800-82 (OT security, zones and conduits); IEC 61508/61511 (functional safety, SIL); ISO 10218 and ISO/TS 15066 (robot and cobot safety); IFR World Robotics (robot density and adoption figures); classic TPM literature (the ~85% world-class OEE benchmark); pump affinity laws (VFD energy claim). Control history: Minorsky's 1922 ship-steering analysis (PID origins) and Ziegler–Nichols (1942) tuning. ISA-88 (batch control model). The simulator is a first-order-plus-dead-time (FOPDT) process (the canonical industrial approximation) with actuator lag, stochastic load, and PI control with anti-windup; illustrative dynamics, not a specific plant.