A failed link to the monitoring station can leave the person on shift without a new measurement while the local controller keeps running its loop. Industrial IoT is the collection and analysis of measurements from industrial equipment, and industrial automation is the control loop that commands an actuator such as a valve, a breaker, a switch or a motor.

This page is for an owner or an operations lead who has to separate those jobs before a project connects a machine.

IoT is used in industrial automation to collect measurements from sensors and machines, move those measurements among devices, controllers, gateways and software, and feed an analysis step. IBM’s explainer of 27 August 2026 defines the Industrial Internet of Things as smart sensors, devices and software that collect and analyze data in manufacturing and industrial operations, and its action examples are a maintenance request, a work order or an alert to an operator. NIST Special Publication 800-82 Revision 3 (September 2023) describes a separate mechanism: a controller turns the measurement and a set point into commands for actuators, and a typical monitoring system has no direct connection from the sensors to any actuators. Name the reporting device and the commanding device before the machine is connected.

More tool explainers sit in the Technology section.

Job What the opened page says Dominant failure mode Prevention the source names
Measurement IBM: sensors collect data from assets such as a turbine blade or flow in a pipe, including temperature on a motor, pressure on a pipeline and vibration on a pump. NIST: a sensor measures a physical property and sends that value on. The next station or controller receives no new measurement on that path. Neither page prints a timeout. IBM: separate operational technology networks from corporate IT networks when possible, with strong authentication and role-based access.
Monitoring NIST: a typical monitoring system has no direct connection between the sensors and any actuators. Values go to a station where a person analyzes them. The person loses that view. In the typical case NIST describes, this path does not command the actuator. NIST’s examples send a person to a forced door, or have control-center staff start an auxiliary cooling unit.
Control loop NIST: the controller applies an algorithm and a set point, then sends commands to actuators such as control valves, breakers, switches and motors. A new measurement fails to arrive, or the manipulated variables fail to go out. NIST leaves the process result to the domain. Where impact is greater, systems often need redundant controls or a way to run in a degraded state. Where timing is tight, put the computation close to the sensors and actuators.
Supervisory link In NIST’s distributed-control example, the supervisor sends set points and requests data. Field controllers command actuators from those commands and from sensor feedback. Loss of the supervisory link. The guide does not say whether an unnamed field controller holds the last set point. An intelligent electronic device, in most cases, can act without a direct instruction from the control center. A fieldbus can run control algorithms so that not every operation is routed back to the PLC.
Diagram of a monitoring path from a sensor to a person at a station, with no line to an actuator, beside a control loop from sensor to controller to actuator
Diagram of the typical case in NIST SP 800-82 Rev. 3. The monitoring path ends at a person. The control loop continues to an actuator. A plant can contain both, and this drawing does not fill the interruption cells.

How is IoT used in industrial automation?

Industrial IoT, on the pages read here, is the measurement path and the analysis that follows it. The control loop is the part that moves the process. A project uses IoT in industrial automation when it carries measurements from the machine to a place that can schedule work, show a screen or, where the design says so, pass a set point down to a controller that already commands the actuators.

IBM lays out a four-step sequence: collection, communication, processing and analysis, and action. Collection is the sensor on the asset. Communication moves data among devices, controllers, gateways and software. IBM lists Ethernet, Wi-Fi, cellular connectivity, or industrial protocols such as Message Queuing Telemetry Transport (MQTT), as options that depend on the environment. The page names MQTT. It does not define the protocol or print a speed. Processing can run on an edge device fixed to the sensor, or farther away in a data center. IBM says local analysis helps reduce latency because the data is processed closer to its source instead of being sent to a data center or a cloud. The page prints no latency number.

IBM’s examples for the action step are scheduling a maintenance request, issuing a work order, or alerting an operator. Devices are often tied to a computerized maintenance management system that can generate work orders, inventory requests and operator alerts. A work order changes the next task a person picks up. In the manufacturing paragraph, IBM says connected production equipment provides real-time information to a manufacturing execution system, which that page calls the software that controls manufacturing operations on a factory floor. What a plant does with that record, once a job and a maintenance decision are in play, is the subject of how a reading becomes a job record and a maintenance decision on the factory floor.

A closed gray metal box with four white cables, mounted on a steel column in a factory aisle
A closed metal box with four white cables on a steel column. The photograph does not show the protocol, or whether the cables carry measurements, commands, or both.

The OPC Foundation’s Unified Architecture page describes one published way to expose that data. The page says OPC Unified Architecture (OPC UA), released in 2008, is a platform-independent architecture that runs from an embedded microcontroller to cloud infrastructure. Clients can read and write current and historical data. A subscription monitors data and reports by exception when values change, using criteria the client sets. The page prints no poll interval. A client can also execute a method the server defines. That is a program call the server has published. It is a different mechanism from the continuous loop NIST describes, in which the controller keeps issuing manipulated variables from the latest measurement and the set point. The same page describes a publish-subscribe option in which a publisher sends messages to middleware and does not know which subscribers exist. That pattern distributes notifications. It does not, by itself, identify the device that commands an actuator.

IBM also separates consumer IoT from the industrial case. Its examples are a consumer sensor that measures heart rate or distance, and an industrial sensor that measures pressure or flow in a pipeline. IBM says consumer applications collect data every few minutes, while industrial applications must collect continuously and respond with low latency. The page states that requirement and prints no latency number, so it is not a measured response time for a named device.

Takeaway: Write down the action the project will take. A work order moves a person. An actuator command moves a valve, a breaker, a switch or a motor.

Where a control loop differs from a screen

NIST SP 800-82 Rev. 3 describes a typical operational technology system as control loops, human-machine interfaces, and remote diagnostics and maintenance tools, built on network protocols and layered networks. Some critical processes also include safety systems.

In the control loop, the sensor sends a measurement to the controller as a controlled variable. The controller interprets it and generates manipulated variables from a control algorithm and target set points, then transmits those to the actuators. The actuators change the process. The glossary, adapting an ISA dictionary entry, defines cycle time as the time, usually in seconds, for one pass: sensor signals are read into memory, the algorithms run, control signals go to the actuators, the process changes, and new sensor signals identify the new state.

An open gray control cabinet beside a machine, with relays, terminal blocks and bundles of blue, yellow and gray wire
An open cabinet beside a machine, with relays, terminal blocks and wire bundles. Small markings on the modules are not legible. The photograph does not show which wires command an actuator and which wires only report a measurement.

People use the human-machine interface to monitor the process, change set points and algorithms, and see status and history. Diagnostics and maintenance utilities are how the system is supposed to prevent, identify and recover from abnormal operation or failures. NIST’s glossary calls diagnostics information about known failure modes and their characteristics, used to pinpoint a cause and choose a corrective measure. Supervisory loops and lower-level loops can be nested, and the guide says those loops run continuously with cycle times on the order of milliseconds to minutes.

A SCADA arrangement in the same guide does both jobs in one system. Field information is collected, sent to a control center and shown to an operator, who can monitor or control from that center in near real time. The local remote terminal unit or programmable logic controller controls the local process. Control of a given task can be automatic, or it can be an operator command. Which one applies is a property of that system. An intelligent electronic device can be polled by the control server and, in most cases, has local programming that lets it act without a direct instruction from the control center.

In the distributed-control example, a supervisory controller sends set points to field controllers and requests data from them. The field controllers command their actuators from the control-server commands and from sensor feedback. NIST notes that a machine controller in the example interfaces with sensors and actuators over point-to-point wiring, while other field devices use a fieldbus. The guide says a fieldbus can carry diagnostics and can run control algorithms in the field, avoiding a route back to the PLC for every control operation. It names Modbus and fieldbus as protocols often used on control networks and fieldbus networks. The guide does not rank them and does not print a bit rate.

Takeaway: Name the device that commands the actuator, and name the device that only displays the measurement.

What stays on the local controller

Rule of thumb. If the actuator command has to be produced on a cycle a person on shift cannot meet, keep that command on the local controller and use the wider network for measurements, set points and alarms. That split follows NIST’s timing paragraph: humans may be unable to meet high-speed, consistent timing, so automated controllers may be necessary, and some systems need the computation performed close to the sensors and actuators so the control action happens on time.

For an existing line, choosing industrial IoT sensors for factory retrofits compares mounting, sampling and isolation before a monitoring connection is added, and machine vision inspection line handoff details how optical exposure timing and encoder-tracked PLC reject verification stay on the deterministic local control loop.

The same design section places a small local PLC-controlled process at one end of geographic distribution and pipelines or power grids at the other. Wider distribution, NIST says, typically implies wide-area networking. In that passage, distance is a reason for the wide-area path, and the actuator command remains with the field controller.

Cycle time T, in seconds, is one completed pass as the glossary defines it. If each pass takes T seconds and the next pass starts immediately, passes per minute = 60 / T.

Worked example, arithmetic only. Assume T = 1 second, an input placed inside the range NIST describes as on the order of milliseconds to minutes. Passes per minute = 60 / 1 = 60. That line is arithmetic on the assumption. It is not a cycle time printed for a named controller, and it is not a sample interval for the monitoring path. IBM’s page prints no industrial sample interval, so that cell stays empty rather than borrowing this control-loop arithmetic.

A fieldbus that runs the algorithm locally is the prevention NIST names when the design would otherwise send every control operation back to the PLC. Holding the last set point after the supervisory link drops is a different question. The opened sections do not answer it for an unnamed controller, so it stays in the blank table below.

What to leave blank when the path stops

NIST says the failure of a control function can have substantially different impacts across domains. Systems with greater impact often need redundant controls or a degraded state. In the safety discussion, the guide says OT systems typically use a fail-to-a-known-state design so that an unexpected situation or a component failure places the process in a safe state. A loss of network communications could trigger those events. Organizations should define the thresholds at which components keep operating with reduced capability, including lost network communications. In the HIGH baseline note on alternate communications paths, NIST says organizations consider which systems need another path so that a loss of communication does not lead to an unacceptable loss of view or control, or to a safety event.

None of those sentences names the loops in a specific plant. The cells below stay blank.

Layer While the path is up If the path stops When the path returns
Sensor Measures a physical property and sends the value to a controller or a monitoring station (NIST SP 800-82 Rev. 3).
Local controller Turns the measurement, the algorithm and the set point into actuator commands (NIST). A local intelligent electronic device can, in most cases, act without a control-center instruction (NIST).
Wider network Moves data. IBM names Ethernet, Wi-Fi, cellular connectivity or protocols such as MQTT. OPC UA can report by exception or carry a defined method call (OPC Foundation).
Analysis Can run on an edge device or in a data center (IBM, 27 August 2026). IBM states that local processing reduces latency and prints no latency figure.
Person on shift Watches status and history on the human-machine interface and can change set points (NIST).

The next table is a checklist. The answer column is blank on purpose. Fill it from the architecture you have, not from this page.

Question for the engineer Answer in the plant documents
Which control loops keep running when the wider network is down?
Which loops stop, and which enter the fail-safe state the builder defined?
Which screens and alarms go dark?
Who is told, and by what signal that does not use the failed path?
After the path returns, who is allowed to change a set point, and from which screen?
Which unsafe conditions must the local system detect and act on without that path?
A dark monitor on a metal desk with an empty chair, and piping and process equipment in the room behind
A dark monitor and an empty chair, with piping and equipment behind the desk. The photograph does not show a network failure, a stopped machine or a safe shutdown.

A dark screen and a dead path are different facts. For a locked terminal rather than a process controller, what a remote kiosk check can see, and which faults still need someone on site separates a reachable device from a viewable screen and from a peripheral that software cannot test.

Takeaway: Copy the questions into the plant documents. Leave a cell empty until those documents answer it.

Where this page does not hold

Use the tables as questions. They are not a control-system design, a wiring diagram or a safety certificate. NIST says that in most safety-critical operations, human oversight of a dangerous process is essential, and that the system must be able to detect an unsafe condition and trigger an action that returns it to a safe one. This page does not specify that action for any plant.

The millisecond-to-minute range is NIST’s description of supervisory and lower-level loop cycle times in general. The one-second worked example is arithmetic inside that description. IBM’s “low latency” has no number. The OPC Foundation page names a binary transport and JSON over WebSockets and prints no speed. No model numbers or prices are used. A vendor sentence is labeled as that vendor’s account. The photographs show hardware in a factory room. They do not identify a brand, a protocol or a failed network.

How this page was checked

The passages above come from three publisher pages. From NIST SP 800-82 Rev. 3: the control-loop, monitoring and human-machine interface section; the design considerations on timing, geography and failure impact; the SCADA note on local intelligent electronic devices; the distributed-control and fieldbus passages; the glossary entries for control loop, cycle time and diagnostics; the fail-safe paragraph; and the alternate-communications discussion. From IBM: the definition, the consumer contrast, the four-step sequence, the network-separation paragraph, and the manufacturing execution system sentence. From the OPC Foundation: the Unified Architecture description of platform range, read and write, subscriptions, methods and publish-subscribe. How this publication selects and attributes sources states the rule for which sources can appear on a page.