Industrial production, cyber-physical production system, model-driven engineering, ontology knowledge-base The evolution of manufacturing systems towards Industry 4.0, bringing flexibility of production processes, re-configurability of production system resources, and better use of industrial artificial intelligence, has fundamentally transformed the operational landscape of industrial production environments [1]. Industrial cyber-physical production systems conforming to Industry 4.0 represent software-intensive architectures, where industrial workcells, robots, and other resources incorporate decisive roles in system functionality, encompassing robot control programs, as well as Manufacturing Execution Systems or Manufacturing Operations Management systems that orchestrate production at the system level [2]. This transformation from centralized to distributed system architectures introduces unprecedented levels of flexibility, enabling mass customization and adaptive manufacturing [ 3, 4]. As a consequence, production engineering has to shift from a component viewpoint to a system viewpoint and to consider holistic principles of system engineering and analysis [5].
(P. Kadera) ∗Corresponding author. †These authors contributed equally. CEUR
ceur-ws.org
Few-AGVs Wrong-positioning
of-battery Remaining-screw
Condition Effects Product & Process Part of Knowledge Graph Product–Agnostic Description of Production System Resources with Resource-Specific Causes of Undesired Conditions Battery-not-arrivedin-time
EV
Battery Cover-unscrewing
failure Cover-removal
failure
Undesired Condition
Effects
Required-Capabiliy
Transport Required-Capabiliy
Unscrew Required-Capabiliy
Transport Required-Capabiliy Remove-Cover
Provided-Capability
Transport AGV-error
Fleet-Management
System Fleet-managementissue
AGV1
AGV2 Provided-Capability
Unscrew
Robot-Endeffector Robotic-Screwdriver
Robot
Robot-WC1-1 Screw-recognitionfailure
Wrong-calibration
Unscrewing-torque
limit-violated Provided-Capability Remove-Cover
Robot-Endeffector Robotic-Screwdriver
Robot
Robot-WC2-1 Pneumaticmalfunction
Wrong-positioning
Smear-or-adhesives Unscrew-Cover
Stage-1 Cover-Removal
Stage-2 Module1-Removal
Required-Capabiliy
Remove-Module
Module2-Removal Products
Processes
Required Capabilities
Provided Capabilities
Resources
Undesired Condition
Causes
Process flow Mapping links • Req. 1 – Model ability to represent flexible Industry 4.0 production of products with highly automated production resources, adopting principles of skill-based engineering; • Req. 2 – Model ability to represent undesired conditions relevant for production system engineering in terms of plausible causes and efects; • Req. 3 – Support for eficient interaction with human engineers and operators via LLMs/chatbots.
The PPR-AKG approach has been used for handing our laboratory prototype over to an industrial
vendor in a project: Automated robotic disassembly of used electric vehicle batteries and their
remanufacturing [
Processes and Systems with Undesired Conditions and their Causes
The proposed approach “Product–Process–Resource Asset Knowledge Graphs” (PPR-AKGs) extends
a Product–Process–Resource (PPR) model, established in [
Fig. 1 illustrates the proposed comprehensive PPR-AKG, which is structured into 4 columns. The core part of the knowledge graph is the blue region (i.e., second column), describing relationships between products (red circles) and processes (green boxes). To address Req. 1, processes are not assigned to resources, but processes are assigned to required capabilities (blue rounded boxes). Each required capability expresses a requirement for a resource, which allows to assign a resource flexibly.
Production resources, such as robotic workcells or autonomous vehicles, are specified in the yellow area by yellow boxes in Fig. 1. Resources provide one or more provided capabilities (purple rounded boxes). Whereas the products and processes in the blue region are described on the class level, meaning that instances of the product-and-process model can be instantiated and executed in parallel, production resources are described on the instance level. To execute production, the required capabilities have to be matched with the provided capabilities by solving a dynamic resource allocation task, which is supported by the PPR-AKG knowledge graph. An key benefit is the ability to add or remove capabilities on the fly, facilitating continuous engineering and improvement of the production systems.
For industrial practice and to meet Req. 2, it is very important to focus on the undesired conditions
(purple boxes) as well as their plausible causes (gray boxes). Plausible causes can be expressed
resourcespecifically (i.e., inside yellow boxes, which can be defined by the resource vendor), as well as on the
system level globally (right-hand column in Fig. 1). Such a scoping of causes for undesired conditions
has been recommended by practitioners and it is also inspired by procan.do method [
While the knowledge graph provides explicit computer-understandable specification, it is dificult to interpret or to modify by humans. This gap can be eficiently bridged by Large Language Models (LLMs) or LLM-based chatbots. The primary use of LLMs to satisfy Req. 3 is encapsulating the PPR-AKG with an intuitive natural language interface for factory operators, production planners, maintenance technicians, and engineers. The typical prompts found a feasible schedule by matchmaking/scheduling required and provided capabilities and skills, or identified plausible causal explanations of undesired conditions (e.g., the prompt “Why did the battery not arrive in time” pinpointed possible causes to technicians). The secondary LLM use instantiates ontology individuals according to available production resources, but this topic required a cross-validation by judgment with diferent LLM types/versions. 3. Conclusions and Future Work This paper reports on demonstrating the potential of semantic knowledge graphs for modeling and improving flexible Industry 4.0 production combined with analysis of undesired conditions in industrial production with plausible causes. The description logic in the proposed PPR-AKG ontology provides machine-understandable model of relevant knowledge, and the access to this model via LLMs facilitates natural human-oriented interaction with operators and engineers. The electric vehicle battery remanufacturing validation confirms practical applicability and substantial performance improvements.
While the PPR-AKG was found useful for mitigating undesired conditions in flexible production, the efective and eficient validation of the PPR-AKG content with LLMs remains a challenge.
In future work, we plan to systematically address the problem of instantiating ontology individuals with one LLM and judgment with another LLM. In addition, we plan to design an optimization method taking into account cost functions for undesired and desired conditions.
This work was co-funded by the European Union under the project ROBOPROX (reg. no. CZ.02.01.01/00/22_008/0004590) and by the Rockwell Automation Laboratory for Distributed Intelligent Control (RA-DIC).
Declaration on Generative AI The author(s) have not employed any Generative AI tools.