Cyber-physical Production Systems: Connecting Physical Operations and Digital Control

What cyber-physical production systems are

The decisive factor is the integration of computation and physical process. A locally controlled machine can also form a cyber-physical system; a higher-level platform is not mandatory. Human participation may be part of the design, but does not have to occur in every individual feedback loop.

The NIST framework describes cyber-physical systems as systems of interacting digital, analog, physical, and human components whose function arises through integrated physical and logical relationships.^[1] Applied to production, this includes more than machines and sensors. Materials, products, work areas, employees, rules, and decisions can also be parts of the system.

Within Industry 5.0, technical coupling is not an end in itself. It is intended to support people in resilient and sustainable production. The European Commission therefore places human-centricity, sustainability, and resilience alongside a purely efficiency- and productivity-based approach.^[5]

Example: Command accepted, effect still pending

An illustrative example distinguishes three times: a sensor measures a temperature deviation at 10:00:00. The higher-level application receives the message at 10:00:05. A later measurement at 10:00:20 shows the state after an intervention was initiated. The times illustrate the sequence; they do not define acceptable latency.

For this case, a local protective function is provided that responds independently of the higher-level application. A further corrective adjustment requires human approval. Automatic protection is thus separate from the action requiring approval. Which protective state is suitable must be determined and tested for the equipment.

Acceptance of the corrective adjustment command by the controller initially demonstrates only that the command was accepted. Only suitable feedback on execution and the observed state allows its effect to be assessed. The new temperature measurement must also be assigned to the correct sensor, object, and time.

If this feedback is missing, the status is not “effect confirmed.” The specific design must define how pending or contradictory feedback is handled. A website description demonstrates neither the protective function nor an actual equipment connection.

Time and synchronization are part of substantive meaning

Physical processes run continuously, while digital systems process information in individual measurements, messages, and state changes. Delays, different system clocks, or lost messages can therefore change the interpretation.

A temperature measurement can only be meaningfully interpreted when its measurement time, transmission time, and associated process phase are known. If a limit violation is not detected until several minutes later, the physical situation may already be different. Likewise, a late-arriving message must not be treated without checking as though it originated in the order of technical processing.

Process events structure observed actions and state changes into analyzable units. For cyber-physical systems, they must additionally represent the temporal relationship between observation, transmission, processing, and intervention. Only then can it remain clear whether a response occurred in time and which actual state it was based on.

NIST treats timing, data, composition, boundaries, lifecycle, and trustworthiness as interconnected aspects of a cyber-physical system.^[1] Time is therefore not merely technical metadata, but part of system function.

The cyber-physical control loop

The basic operation of a cyber-physical production system can be understood as a control loop:

1. Observe: A relevant physical state is captured through sensors, machine data, or a human observation.
2. Assign: The information is connected to the affected object, task, time, and applicable conditions.
3. Assess: Software compares the state with requirements, rules, models, or known relationships.
4. Decide: Depending on risk and system design, a controller makes a limited automatic decision or presents action options to a person.
5. Intervene: A machine, actuator, or person changes the physical process.
6. Check the effect: The new state is captured again and compared with the expected effect.

Edward A. Lee characterizes cyber-physical systems as the integration of computation and physical processes, in which embedded computers and networks monitor and control physical processes and both sides are connected through feedback.^[2] This feedback distinguishes a functioning operational system from retrospective data collection.

A control loop does not have to be fully automated. The decision may deliberately remain with a suitably qualified person. What matters is that observation, assessment, action, and effect are retained as a connected history.

Feedback must belong to the intervention that was triggered

A new temperature message may originate from another vessel or an old transmission buffer. In that case, it does not confirm the effect of the intervention just approved. Object assignment and measurement time must therefore be checked together.

The context model describes the domain relationships. In the control loop, these relationships are specifically needed to connect the trigger, addressed controller, and later observation. An intervention code alone is insufficient.

Where several system clocks are involved, it must also be known how their times are made comparable and what uncertainty remains. Without this relationship, supposedly later measurements may actually have originated before the intervention.

Which components work together

Cyber-physical production systems are not a single product. They arise from the coordinated connection of several layers:

• Physical objects: Systems, equipment, tools, materials, products, samples, and spatial areas.
• Capture: Sensors, measuring devices, cameras, machine messages, and structured human inputs.
• Communication: Local fieldbuses, industrial networks, interfaces, gateways, and higher-level platforms.
• Digital processing: State models, rules, event processing, analyses, and operational knowledge models.
• Intervention: Actuators, machine controllers, approvals, tasks, or instructions to people.
• Human participation: Execution, observation, review, decision, approval, and correction.

The Industrie 4.0 Working Group described cyber-physical production systems as the foundation of connected industrial value creation, in which embedded systems, production resources, and logistics communicate and can connect across company boundaries.^[3] In regulated operations, however, technical connectivity alone is insufficient. It must also remain clear which object was affected, which requirement applied, and who was responsible for an intervention.

Not every feedback loop leads to autonomous control

Several stages lie between pure display and fully automatic intervention:

• Observation: The system captures and visualizes the state without suggesting a response.
• Warning: A defined condition triggers a notification.
• Recommendation: The system provides context-specific action options and their basis.
• Intervention requiring approval: A prepared action is executed only after human confirmation.
• Limited automation: A preapproved rule autonomously triggers an action within clear limits.
• Safe state: In the event of a fault, uncertainty, or loss of communication, a defined state is established or a controlled handoff is triggered.

The appropriate stage depends on risk, reversibility, time criticality, and evidence needs. A minor corrective adjustment may occur automatically, while product release or deviation assessment remains reserved for a responsible person.

Workflow models can specify which task, review, or approval follows a detected state. The cyber-physical production system supplies the actual trigger and incorporates the subsequent effect back into the observed context.

Human interventions must take effect within the time window

For a corrective adjustment requiring approval, achievable response time is part of the design. If the physical process is faster than a realistic human review, a person cannot be planned as the sole protective measure.

With Human in the Loop, the necessary information and options must be available, and the response to a missing decision must be defined. In a cyber-physical system, the approved intervention must also be executed in time and its actual effect observed.

Reliability is established across the entire control loop

A cyber-physical production system can only be as reliable as the connections between its components. A calibrated sensor offers little benefit if its values are assigned to the wrong object. A correct rule can cause harm if an actuator executes it too late or in an unsuitable operating state. A complete record is insufficient if unauthorized people can trigger control commands.

Central requirements therefore include:

• known identities and states of all relevant components,
• unambiguous communication and responsibility boundaries,
• checks of data quality, temporal context, and completeness,
• permissions for observation, decision, and intervention,
• detection of failures, conflicts, and implausible states,
• defined safe responses to uncertainty or loss of connection,
• traceable changes to software, rules, and equipment configurations.

Lee notes that physical components bring safety and reliability requirements that differ qualitatively from ordinary application software.^[2] A software process may run formally correctly yet respond too late for the actual operation or under incorrect physical assumptions.

A cyber-physical system and a digital twin are not the same

A digital twin represents relevant properties and states of a real production object or process. ISO 23247-1 describes general principles and requirements for a digital twin framework for manufacturing for this purpose.^[4]

The cyber-physical production system, by contrast, encompasses the entire functional relationship between physical components, capture, communication, digital processing, control, and human participation. A digital twin can supply the current and historical representation within it. It is not, however, automatically the sensors, network, actuators, or responsible organization.

Conversely, a locally controlled machine can form a cyber-physical subsystem without a comprehensive operational twin existing. A reliable production architecture therefore needs a clear distinction between what acts physically, what represents digitally, what decides, and which connection holds the two sides together.

Distinguish related concepts clearly

• An embedded system is a computer-based component within a device. It can be part of a cyber-physical system.
• The Internet of Things emphasizes connected physical objects and their communication. A cyber-physical system additionally emphasizes the functional relationship between action and feedback.
• Industrial automation performs defined control tasks. It can be part of a cyber-physical production system but does not have to be extensively context-aware or connected.
• A digital twin digitally represents relevant properties and states of real objects or processes. It can form the information basis of a cyber-physical system.
• A cyber-physical production system encompasses physical production, capture, digital processing, intervention, and human participation as a connected system.

What cyber-physical production systems do not accomplish

• More sensors do not automatically mean more insight. Data need a purpose, context, and a substantively justified use.
• A model does not replace physical reality. Wear, faults, and unrecorded conditions remain possible.
• A successful intervention does not prove a general cause. Its effect initially applies to the observed context.

Command acceptance is not yet a confirmed process effect

In the example, the message at 10:00:05 describes a state measured earlier. An accepted corrective adjustment and a later observation are further, separate pieces of information. Their assignment determines which effect can be assessed at all.

Cyber-physical production systems therefore require a view of the entire temporal relationship between actions and effects. Reliability concerns capture, transmission, intervention, and feedback, including the defined responses to failure.