From a device value to usable process information

IoT Gateway: Assigning Device Data to the Work Step

An IoT gateway connects devices or technical data sources with a downstream application, for example by bringing interfaces together or translating protocols. For process execution, that connection alone is not enough: a received value also has to belong to the correct object, time and work step.

System KnowledgeAuthor: Published: Last updated:

In industrial applications, the term Industrial IoT Gateway is also used. NIST describes an IoT device as having at least one sensor or actuator and a network interface.[1] Process execution adds a second task: the technically received information has to become usable in the correct operation.

Sensor among plants; device value, IoT gateway, assignment and work step shown as graphic stages above

What an IoT gateway connects — and what it does not

An IoT gateway mediates between technical systems. Depending on the architecture, it can bring interfaces together or translate protocols. OPC UA describes a Gateway Server as an intermediary for one or more servers and explicitly names protocol conversion as one possible function.[2]

The identifier of a scale does not establish whether its value belongs to a weighing step, a check measurement or a return weighing. A scanned container identifier does not establish whether that container is the one expected in the current step. A technically valid temperature value likewise does not tell the process which product or process state it is relevant to.

The received information therefore has to be connected with the process. The context model defines which objects, tasks and states belong together for the respective work step.

The gateway brings the information to the application. The process logic determines what it belongs to.

Start with the task, then choose the device

Device integration should therefore not start with the longest possible list of supported interfaces. The first question is which task is to be solved in the work step.

Device or workplace toolTask in the processWhat to establish before use
ScaleCapture a quantity for a work stepMeasured value, unit, device, object and meaning of the measurement
Barcode or RFID readerSelect material, a container or equipmentIdentifier read, expected object and handling of an incorrect selection
Sensor or data loggerObserve a condition over timeQuantity measured, unit, source, time reference and acceptable age of the information
Terminal or smart glassesDisplay instructions and enable inputStep displayed, person acting, expected input and its effect

These tasks are fundamentally different. With industrial smart glasses, display and interaction at the person are central. An automatically received sensor value serves a different purpose.

Each process should therefore specify separately which information can be received automatically and where human assessment or confirmation is still required.

Example: weighing into the correct operation

In the following fictional example, a weighing is recorded for order A-204. Scale W-03 reports 1.250 kg. The active step requires the quantity for container B-17.

The value alone is not enough. The application also has to know that the message came from W-03, that the unit matches the expected entry and that B-17 is actually selected in the current work step.

This creates a chain: value → source → object → work step. Only this assignment turns the received weight value into usable process information.

01DeviceW-03 · 1.250 kg
02IoT gatewayReceive message
03AssignmentA-204 · B-17
04Work stepRecord weighing
Fictional weighing example: technical receipt and assignment in the process are separate tasks. Schematic design requirement, not a 420+ screen workflow.

This becomes especially important when a value does not arrive immediately. If a message reaches the application later, it must not be assigned to the operation that is active by then merely because the device identifier is still the same.

Which information belongs together

For process assignment, a measured value is better treated as part of a related set of device information. Which fields are technically available depends on the device, interface and application. For the process-level assessment, the following questions are particularly useful:

InformationQuestion to resolve
ValueWhat was transmitted?
UnitIn which unit is the value to be understood?
SourceWhich device or technical source did the information come from?
Status or qualityIs there a technical indication of whether the value can be used?
SourceTimestampWhen was the value timestamped by its data source?
Receipt or transfer timeWhen did the information reach the next system layer?
Object referenceWhich material, container, product or other object does it belong to?
Work stepIn which specific operation is the information to be used?

This overview is not a universal data model for IoT systems. It is a checklist for a different question: Which pieces of information have to fit together before a device value can be used in the process?

The order in which the information becomes available is not arbitrary either. An application may already have received a value technically even though the associated object reference has not yet been established. Conversely, an object may already be selected even though no new device value is available for the current step. Process design should therefore define when the related information is complete enough for the step to continue.

Object assignment should not be inferred silently from the workplace. One device may be used successively for different orders, materials or process steps. A unique device identifier identifies the source — not automatically the process object the message belongs to.

Assess SourceTimestamp, ServerTimestamp and status together

For measured values, the question is not only which value is present, but also when it was generated or processed.

OPC UA makes this distinction explicit. A DataValue can contain a value together with a StatusCode, a SourceTimestamp and a ServerTimestamp.[3]

The SourceTimestamp reflects the time applied to the value by the data source. If the value is forwarded by another OPC UA Server, the source timestamp is passed on unchanged. By contrast, the ServerTimestamp reflects when the respective Server received the value or knew it to be accurate.[3]

Two time references can therefore exist side by side without contradicting each other. Whether the clocks of the devices involved are aligned with one another is a question of time synchronization. For the work step, this creates concrete questions: Is the value still current for this task? Did it come from the intended source? Does its time reference fit the active operation?

The StatusCode can additionally provide a technical indication of whether the value can be provided or used. This technical assessment does not replace process assignment. Data integrity and fitness for use in the specific process step therefore have to be considered together while remaining conceptually distinct.

When preparing measurement time series for AI analysis, a transmission gap must be distinguished from an actual change in the measured process variable.

Distinguish missing, delayed and duplicate messages

Three situations are particularly easy to conflate in device integration: an expected message is missing, a message arrives late, or the same message is delivered again. These cases can have different consequences for the process.

If a message is missing, the process has to make clear that the expected value is unavailable. A previous measurement should not silently appear as a new one. If a message arrives late, the process has to check whether it still belongs to the active object and active step. Receipt time alone is not enough.

If a message is delivered again, the same process-level information must not silently become two measurements. MQTT 3.1.1 shows why applications have to account for these cases: with QoS 0, message loss can occur; with QoS 1, duplicate delivery can occur.[4]

Those statements concern message transport. The resulting process events must then make clear which information was actually assigned to an operation and used there.

If the received measurements are subsequently assessed by a trained model, the time until the analysis result is available remains a separate measure alongside the time of receipt.

Reading is not controlling

Device integration can operate in two fundamentally different directions. In the first case, an application reads information from a device or receives a message. Examples include a measured value, a device status or an identifier read from a Data Matrix code. In the second case, a system sends a command or setpoint to a device or an industrial control system.

These two directions should not be treated as architecturally equivalent.

An outgoing command introduces additional states: Who or what is allowed to trigger it? Which command was intended? Was it accepted technically? And does the device state observed afterwards actually correspond to the intended state?

A technical receipt acknowledgement is not automatically evidence that a physical action was carried out. The ability to read device data therefore does not imply a general ability to control devices. A write connection has to be designed, constrained and tested separately for the specific use case.

Include an outage in the trial

An initial trial does not have to cover an entire fleet of devices. A limited work step with a clearly identified device and a clear object reference is more useful.

That scope can be used to test different states deliberately: the expected value arrives correctly, the wrong object is selected, an old or delayed message arrives, the connection is interrupted, or the same message is delivered again.

For each case, the process should make clear what state the person at the workplace sees and how work can continue in a controlled way. The failure cases are particularly informative.

Such a test also shows whether normal operation and failure handling use the same assignment logic. After an interruption in particular, the process should not silently fall back to the last known value when the current step expects a new measurement. Restart behavior is therefore part of process design, not merely a question of whether the connection is technically available again.

How does the process establish that this exact value belongs to the current object, time and work step?

A device identifier does not establish the person acting.

This does not imply blanket plug-and-play compatibility with arbitrary devices or protocols.

Primary sources and further reading

  1. [1] · Technical reference IoT device NIST. CSRC Glossary, definition referencing NISTIR 8425. Sensor/actuator and network interface NIST CSRC Glossary ↑ back to text
  2. [2] · Technical reference OPC Unified Architecture — Part 4: Services OPC Foundation, v1.05.07, section 3.1.8, Gateway Server. Intermediation and possible protocol conversion Gateway Server ↑ back to text
  3. [3] · Technical reference OPC Unified Architecture — Part 4: Services OPC Foundation, v1.05.07, section 7.11, particularly 7.11.3 to 7.11.5. DataValue, StatusCode, SourceTimestamp and ServerTimestamp DataValue ↑ back to text
  4. [4] · Standard MQTT Version 3.1.1 Plus Errata 01 OASIS (2015), 10 December 2015. Abstract and sections 4.3.1–4.3.2: delivery guarantees and transport acknowledgements MQTT 3.1.1 Plus Errata 01 ↑ back to text