Source-linked AI summary
Cyber-Physical Digital Factory Architecture as the Enabler of Disembodied Work
Tero Kaarlela, Ivan Ruchkin, Jose Outeiro, Souradeep Dutta
TL;DR
Manufacturing technologies are often applied to individual processes rather than integrated into a unified operational environment, leaving disembodied work largely unsupported. The paper proposes a cyber-physical digital factory architecture combining synchronized DTs, cloud-edge AI, IIoT, and XR teleoperation, and demonstrates feasibility across three manufacturing applications. The resulting framework provides a reusable basis for human-AI-controlled digital factories, while factory-wide orchestration remains unvalidated.
Problem
A unified operational CPS architecture integrating heterogeneous resources, DTs, AI control, and human teleoperation to support disembodied work has not been established.
Method
The paper integrates synchronized DTs, hierarchical cloud-edge AI, autonomous resources, human supervision, teleoperation, scheduling, runtime contracts, and dynamic control handoff.
Results
The architecture’s feasibility was demonstrated through CNC machining, robotic-assisted abrasive finishing, and EVB disassembly, including DT synchronization and remote supervision capabilities.
Takeaways & Limitations
The implementations collectively provide validated building blocks for a unified digital factory supporting disembodied manufacturing work.
Takeaways & Limitations
The systems were developed and evaluated separately, and continuous factory-wide coordination through a single scheduler was not experimentally validated.
Abstract
from arXiv · showhide
Digital Twins (DTs), Artificial Intelligence (AI), and Industrial Internet of Things (IIoT) technologies have significantly advanced manufacturing digitalization. However, these technologies are typically applied to individual manufacturing processes rather than integrated into a unified cyber-physical manufacturing environment. This paper proposes a cyber-physical digital factory architecture that enables disembodied work, where manufacturing systems can be supervised and operated remotely through eXtended Reality (XR) user interfaces in collaboration between AI-based control and human operators. The architecture integrates synchronized DTs, hierarchical cloud-edge AI, IIoT, and XR teleoperation interfaces into a cyber-physical manufacturing environment. The proposed approach is validated through representative manufacturing operations, including CNC machining, robotic-assisted abrasive finishing, and robotized disassembly. The results demonstrate the feasibility of the proposed architecture for disembodied manufacturing work and provide a reusable cyber-physical framework for future human-AI-controlled digital factories.
I. INTRODUCTION
Manufacturing operators remain tied to physical facilities because supervision, exception handling, and complex decisions require direct interaction. The paper frames disembodied work as a way to enable safe, flexible, location-independent operation through coordinated humans, AI, and physical systems.
- Manufacturing operators remain physically tied to facilities because supervision, exception handling, and complex decisions require direct interaction.
- Disembodied work involves humans supervising, collaborating with, or intervening in physical manufacturing systems without being co-located with them.
- Disembodied work permits decision authority to shift dynamically between autonomous systems and remote human operators through synchronized digital twins.
- The proposed paradigm targets safe, flexible, and location-independent operation of manufacturing resources.
- Location-independent work could widen the labor pool while keeping physical capital, tooling, material flows, and production onshore.
B. State-of-the-art
Digital factories increasingly combine digital twins, IIoT, AI, cloud-edge computing, and teleoperation, but these technologies provide complementary capabilities rather than one unified account of manufacturing operation.
- Digital factories have evolved from integrated engineering, planning, simulation, and lifecycle-management environments toward cyber-physical manufacturing environments.
- Digital twins continuously synchronize physical and digital machines, processes, and manufacturing systems.
- AI supports process monitoring, anomaly detection, predictive maintenance, scheduling, autonomous operation, and decision-making with real-time IIoT data.
- Teleoperation enables location-independent supervision and control, while XR and human–robot interaction improve situational awareness and remote manipulation.
- DTs represent CPS state, IIoT exchanges information, AI supports decisions, and teleoperation enables human participation.
C. Reference Architectures and Standards
Existing standards and reference architectures address manufacturing digital twins, CPS structure, interoperability, production management, or cloud-edge orchestration. The paper identifies a remaining gap in a unified factory-scale operational architecture supporting disembodied work.
- ISO 23247, the 5C architecture, RAMI 4.0, the Asset Administration Shell, and IEC 62264 address distinct digital-twin, CPS, interoperability, or production-management concerns.
- Table I positions the proposed architecture against other approaches using primary-focus, partial-addressing, and not-fully-addressed categories.
- Existing digital-factory implementations mainly support production planning and simulation, while related digital-twin work often targets individual processes and systems.
- A unified operational CPS architecture integrating heterogeneous factory resources, DTs, AI control, and human teleoperation has not been established.
- The proposed contributions include integrated cyber-physical architecture, human-in-the-loop control-authority allocation, and validation through machining, finishing, and EVB disassembly.
II. PROPOSED DIGITAL FACTORY CONCEPT
The proposed concept organizes physical resources, edge intelligence, cloud coordination, and human XR supervision into synchronized feedback loops. Its execution core supports scheduling, runtime contract checking, safety monitoring, and dynamic handoff between autonomous and human control.
- The architecture targets an operational cyber-physical manufacturing environment in which distributed humans, intelligent software, and physical systems function as one coordinated environment.
- Synchronized DTs and common communication support real-time information exchange across multiple manufacturing resources and workflows.
- Three layers span physical execution, edge perception and control, and cloud-level coordination with human supervisory teleoperation.
- XR interfaces support real-time teleoperation for novel or unstructured tasks and autonomous operation using learned task modules.
- The execution layer schedules concurrent activities and uses contracts plus continuous checking to validate state transitions and prevent conflicting operations.
- The mode switcher allocates authority among autonomous, hybrid, and full-remote modes according to task complexity, confidence, safety, and communication conditions.
- An independent safety monitor continuously enforces collision avoidance, safe linear speed, and protection distances across concurrent tasks.
B. Cyber-Physical State Synchronization
The architecture uses continuously synchronized cyber-physical states and MQTT publish/subscribe communication to connect manufacturing resources, digital twins, cloud services, and remote operators.
- B. Cyber-Physical State Synchronization: Continuous synchronization connects physical resources, digital twins, cloud services, and remote operators for disembodied work.MQTT publish/subscribe decouples software components and supports scalable integration of heterogeneous manufacturing systems.
C. Artificial Intelligence Operation
The architecture distributes AI across cloud and edge layers while retaining human supervision and teleoperation when confidence, safety, accuracy, or decision-making limits are reached.
- C. Artificial Intelligence Operation: Hierarchical AI distributes decision-making between cloud orchestration and edge perception-action loops near manufacturing systems.Cloud AI manages factory-level coordination, while edge AI supports low-latency interaction with physical systems.
- C. Artificial Intelligence Operation: Cloud AI handles resource allocation, scheduling, operator management, knowledge management, and factory-wide decision-making.Its work modules describe manufacturing processes at a higher level rather than specifying how individual tasks are performed.
- C. Artificial Intelligence Operation: Physical AI combines machine vision and machine learning to localize EVB components and guide autonomous robotic detachment.This provides the perception, reasoning, and execution required for low-latency autonomous operation.
- C. Artificial Intelligence Operation: Human control resumes through XR teleoperation when predefined confidence or safety criteria are violated.In EVB disassembly, autonomous operation pauses when corroded or broken screws prevent component identification or detachment.
III. VALIDATION AND DISCUSSION
The architecture was evaluated across CNC machining, robotic-assisted abrasive finishing, and robotized EVB disassembly using independently developed systems that realize subsets of the proposed design.
- III. VALIDATION AND DISCUSSION: Three manufacturing operations—CNC machining, robotic-assisted abrasive finishing, and robotized EVB disassembly—were used for evaluation.The systems were developed and evaluated independently rather than operated together as an integrated factory.
- III. VALIDATION AND DISCUSSION: The implementations demonstrate technological building blocks for disembodied work and contribute to the unified digital factory architecture.Figure 2 presents the corresponding XR teleoperation interfaces.
- III. VALIDATION AND DISCUSSION: The systems provide proof-of-concept evidence that heterogeneous manufacturing resources can be incorporated into a common operational digital factory architecture.The demonstrated capabilities include twin synchronization, remote teleoperation, autonomous operation, cloud services, and edge-based AI.
A. Digital Twin Synchronization
The validation demonstrates synchronized digital twins, human teleoperation, autonomous operation, and edge AI within a cloud-connected cyber-physical factory architecture.
- A. Digital Twin Synchronization: 431 ms to 1.6 s round-trip latency, averaging 563 ms, supported synchronized remote operation over cloud infrastructure.The implementations maintained continuous synchronization between physical and virtual manufacturing systems.
- A. Digital Twin Synchronization: CNC, EVB disassembly, and abrasive finishing systems synchronized operational states between physical equipment and digital twins.Synchronized variables included machine coordinates, tool states, spindle speed, robot joints, battery pose, and process execution.
- A. Digital Twin Synchronization: Human operators can teach unknown EVB disassembly operations through teleoperation before storing sequences for later autonomous execution.This hybrid paradigm combines human intervention for unfamiliar tasks with subsequent autonomous operation.
- A. Digital Twin Synchronization: 98.1% precision and 96.5% recall were achieved for connector detection, while wire segmentation achieved 74% precision and 82% recall.Edge AI performs machine vision, component recognition, and connector localization directly on the manufacturing system.
- A. Digital Twin Synchronization: Cloud services coordinate digital twins, AI, scheduling, and persistent data, while edge services execute operations and enforce local safety.The architecture uses interacting feedback loops to coordinate decisions across physical and cyber domains.
C. Validation of the Digital Factory Architecture
The implementations validate the proposed digital factory’s principal capabilities across synchronized DTs, interoperable communication, cloud-edge intelligence, and human teleoperation. Validation remains preliminary because factory-wide orchestration and concurrent industrial operation were not experimentally assessed.
- The architecture combines synchronized DTs, MQTT communication, cloud coordination, edge AI, and teleoperation for human intervention when autonomous execution reaches operational limits.These capabilities collectively support the proposed digital factory’s operating principles.
- The implementations are application-independent across different physical processes, sensing modalities, and autonomy levels while retaining common architectural principles.Shared principles include synchronized DTs, publish/subscribe communication, hierarchical cloud-edge AI, and human supervision and teleoperation.
- Cybersecurity is treated as a continuous lifecycle activity, with IDSs and IPSs complementing vulnerability scans and software maintenance.Connected-manufacturing attacks may affect physical equipment and operator safety.
- The results provide preliminary validation of separate architectural components rather than experimentally demonstrated factory-wide orchestration through a single scheduler.Continuous coordination among all manufacturing resources remains unvalidated.
- Persistent work storage would retain synchronized operational data, workpiece geometry, manufacturing parameters, detected anomalies, timestamps, and complete workpiece histories.Scalability under concurrent resources and robustness to latency, disruptions, and dynamic allocation require further investigation.
IV. CONCLUSIONS
The paper proposes a unified cyber-physical digital factory architecture for location-independent manufacturing operations. Representative applications demonstrate feasibility and validate key capabilities, while broader scheduling, orchestration, and multi-cell coordination remain future work.
- The architecture enables disembodied work by integrating synchronized DTs, hierarchical cloud-edge AI, teleoperation, and autonomous resources into a continuously synchronized environment.It provides an operational framework for location-independent manufacturing rather than primarily supporting planning and simulation.
- The architecture provides an operational framework for location-independent manufacturing operations.
- Feasibility was demonstrated through CNC machining, robotic-assisted abrasive finishing, and EVB disassembly.
- These implementations validated DT synchronization, remote supervision, AI-assisted autonomous operation, and secure cloud-edge communication despite being developed independently.
- Future work targets persistent data storage, factory-wide scheduling and resource orchestration, and validation in a multi-cell manufacturing environment.The target environment must support real-time coordination across heterogeneous manufacturing systems.