A cyber-physical system (CPS) is a machine or process where physical hardware and computing are engineered together: sensors read the physical world, a computer decides, and actuators change the physical world back, in a closed loop. The U.S. National Institute of Standards and Technology defines them as smart systems built from interacting networks of physical and computational components. They are the building block of Industry 4.0.
The phrase sounds academic, but you almost certainly run dozens of them. A temperature-controlled oven that measures its own heat and modulates a burner is a cyber-physical system. So is a CNC machine that adjusts feed rates from force sensors, and a bottling line that slows itself when a downstream buffer fills. What is new is not the idea, it is that these systems are now networked, data-rich, and increasingly modeled in software, which is what turns a collection of smart machines into a smart factory.
What makes a system "cyber-physical"?
The closed loop. A plain sensor that logs a temperature is not cyber-physical; a plain motor that spins on command is not either. A cyber-physical system is the loop that joins them: it senses a real quantity, a computation reasons about it, and an actuator acts on the physical world, then the result is sensed again, and the loop repeats. That feedback is the whole idea. The physical and the digital are not two systems bolted together; they are one system that happens to have a physical half and a computational half.
Walk it through on one machine. A plastic extruder has to hold a melt temperature within a few degrees or the product goes out of spec. A thermocouple senses the actual barrel temperature many times a second. A controller compares that reading to the target and computes how much to open or close the heater, not just on/off, but proportional to how far off it is and how fast it is drifting. The heater actuates. The temperature moves. The thermocouple senses the new value, and the loop runs again. Nobody stands there turning a knob; the physical and computational halves do it between them, thousands of times an hour. That is a cyber-physical system doing exactly one job. A modern plant is thousands of these running at once.
How is a CPS different from automation, IoT, or embedded control?
These terms overlap and get used loosely, so here is a working distinction. Ordinary PLC-based automation closes a control loop but has historically been an island, deterministic, local, and blind to the rest of the plant. The Internet of Things adds connectivity, it gets data off the device and onto a network, but connectivity alone is not control. An embedded system is a single device with a chip inside it. A cyber-physical system is the union: local control, network connectivity, and a computational model that can reason across devices and time, all in a feedback loop with the physical process.
| Term | What it emphasizes | What it is missing on its own |
|---|---|---|
| Embedded system | Computing inside one device | Networking and system-level view |
| PLC automation | Deterministic local control | Connectivity and a wider model |
| Industrial IoT | Getting data onto the network | Control and closed-loop action |
| Cyber-physical system | The whole sense-compute-actuate loop, networked | Nothing structural, it is the union |
In practice the boundaries blur, and that is fine. The useful takeaway is that a CPS is defined by the closed loop and the computation inside it, not by any one component. The IIoT layer feeds it, the controller executes it, and the model gives it something a lone PLC never had: memory and context.
What are the parts of a cyber-physical system?
Every CPS, from a single smart oven to a whole connected line, has the same anatomy. Naming the parts makes it easy to see what a given plant already has and what it is missing.
- The physical process. The actual physics being controlled, heat, pressure, flow, torque, position, chemistry. This is the ground truth everything else serves.
- Sensing. Instruments that turn physical quantities into signals: thermocouples, load cells, encoders, flow meters, vibration sensors. No sensing, no cyber-physical system.
- Networking. The path that carries signals off the device and commands back, from field buses up through the plant network. This is the layer IoT contributed.
- Computation and modeling. The decision-making core: control logic, analytics, and increasingly a software model of the system that predicts and explains behavior.
- Actuation. The devices that push back on the physical world, valves, drives, heaters, robots, closing the loop the sensors opened.
Where do digital twins fit?
A digital twin is the computational half of a cyber-physical system taken to its logical end: a living software model of the physical asset, fed by its sensors, kept in step with reality. If a basic CPS reacts, sense, decide, act, a twin lets the system rehearse. It can predict where the process is heading, test a change in software before committing it to steel, and explain why the physical thing behaved the way it did. The twin does not replace the control loop; it sits alongside it, turning raw feedback into foresight. That is why the two ideas are so often discussed together: the digital twin is what makes a cyber-physical system smart rather than merely reactive.
What does a cyber-physical system look like on a plant floor?
Concretely, a modern line is a hierarchy of cyber-physical systems nested inside a bigger one. At the bottom, each machine closes its own tight loop under a PLC. Above them, a SCADA layer supervises many machines, and a plant-level layer coordinates the lines against orders and quality targets. Each level senses, computes, and acts on a slower clock than the one below it, milliseconds at the machine, minutes across the line, hours across the plant. The plant itself becomes one large cyber-physical system whose "sensors" are all the machines' data and whose "actuators" are the decisions that get pushed back down. Machine monitoring is often the first rung: reading the existing signals to give the higher loops something to reason about.
This hierarchy is why so many plants are further along than they think. The bottom loops already exist and have for decades, every controlled machine is a working cyber-physical system. What is usually missing is the higher loops: the plant-level sensing and computation that would let one line's data inform another's decisions. Those loops rarely require new hardware. They require connecting signals that already exist to computation that can reason across them, which is a data and software problem, not a controls-rip-and-replace one.
The nesting has a consequence worth naming: timing is a first-class concern, not an afterthought. Each loop runs on the clock its physics demands. A safety interlock on a press has to react within a scan, milliseconds, or someone gets hurt. A line-balancing decision can take seconds. A "which product ran worst this week" question can take hours and lose nothing. Trouble starts when a loop is asked to run faster than it can, or slower than it must. NIST devotes an entire volume of its cyber-physical systems framework to timing for exactly this reason. When you evaluate a plant's digital ambitions, the honest question is not "can we collect this data" but "at what speed does each loop actually need to close", and to build the fast, safety-critical loops on deterministic control while the slower, context-rich loops run in the analytics layer above.
By the Numbers
The concept has a formal foundation. NIST convened a public working group in 2014 and published its Framework for Cyber-Physical Systems (SP 1500-201) which defines a CPS as engineered, interacting networks of physical and computational components and organizes the design around "aspects" such as functional, timing, data, and trustworthiness. NIST also frames cyber-physical systems as the core of Industry 4.0, and flags that connecting the physical and digital worlds expands the cybersecurity surface that has to be defended (NIST Manufacturing Innovation Blog). Where Harmony fits: Harmony operates on the computational side of these loops, it connects to the sensors and PLCs a plant already owns, computes true OEE and context from source signals, and layers search and approvable AI actions on top, without touching the deterministic control at the machine. See the connected systems module or the CLS case study.
Why are cyber-physical systems the foundation of Industry 4.0?
Because Industry 4.0 is, at bottom, the claim that factories get smarter when their physical and digital halves are fused and networked. Every headline technology of the movement is a cyber-physical system or a service built on one: the smart factory is a plant of coordinated CPS, the digital twin is the modeling layer, and agentic AI is what happens when the computational half gets good enough to recommend and take action. The through-line is the loop. A plant advances not by buying more sensors but by closing more loops, connecting the data it already generates to computation that can act on it, safely, at the right speed. For the next layer up, see how these pieces assemble into a manufacturing operating system.