Procedural Knowledge Ontology (PKO) (2503.20634v1)

Abstract: Processes, workflows and guidelines are core to ensure the correct functioning of industrial companies: for the successful operations of factory lines, machinery or services, often industry operators rely on their past experience and know-how. The effect is that this Procedural Knowledge (PK) remains tacit and, as such, difficult to exploit efficiently and effectively. This paper presents PKO, the Procedural Knowledge Ontology, which enables the explicit modeling of procedures and their executions, by reusing and extending existing ontologies. PKO is built on requirements collected from three heterogeneous industrial use cases and can be exploited by any AI and data-driven tools that rely on a shared and interoperable representation to support the governance of PK throughout its life cycle. We describe its structure and design methodology, and outline its relevance, quality, and impact by discussing applications leveraging PKO for PK elicitation and exploitation.

• The paper presents a formal semantic model for capturing and managing tacit industrial procedural knowledge.
• It delineates an ontological structure that separates abstract procedure specifications from concrete execution instances through real-world use cases.
• The paper demonstrates practical applications, including a PK elicitation tool and an AI-powered chatbot for real-time procedural guidance.

Procedural Knowledge Ontology (PKO): Structure and Application

The Procedural Knowledge Ontology (PKO) provides a formal semantic model designed to explicitly represent industrial procedures and their execution instances (2503.20634). Its primary objective is to capture and structure procedural knowledge (PK), which often remains tacit or is embedded in unstructured formats within industrial environments. By offering a shared, interoperable representation, PKO aims to support the governance of PK throughout its lifecycle, facilitating its elicitation, storage, retrieval, execution support, and analysis, particularly through AI and data-driven tools. The ontology was developed based on requirements derived from three distinct industrial use cases: safety procedures in a processing plant, commissioning of CNC machines, and managing mixed human-machine activities in electrical grid operations.

Ontological Structure and Core Concepts

PKO distinguishes fundamentally between the abstract representation of a procedure (the plan or specification) and its concrete instantiations during execution. This separation addresses limitations observed in some existing process modeling approaches.

• Procedure Specification: Modeled primarily using pko:Procedure (a subclass of p-plan:Plan) and pko:Step (related to p-plan:Step). A pko:Procedure consists of an ordered sequence of pko:Step instances. Procedures have associated metadata, including status (e.g., pko:ProcedureStatus like 'Draft', 'Approved'), version information, and responsible agents (prov:Agent). Steps can have associated properties like required pko:ExpertiseLevel, preconditions, postconditions, and pko:StepVerification requirements.
• Procedure Execution: Modeled using pko:ProcedureExecution (a subclass of prov:Activity) and pko:StepExecution. These classes represent specific instances of carrying out a procedure or a step at a particular time. Execution instances are linked to the corresponding specification (pko:Procedure or pko:Step) via properties like p-plan:isStepOfPlan and p-plan:isSubPlanOfPlan. Executions capture temporal information (start/end times, duration, using the Time Ontology), the executing agent (prov:wasAssociatedWith), execution status (pko:ProcedureExecutionStatus like 'InProgress', 'Completed', 'Failed'), and contextual events that occurred during execution.
• Contextual Information: PKO integrates various contextual elements crucial for industrial procedures.
  • Resources: Links procedures and steps to relevant documents, images, videos, etc., using dcat:Resource and DCMI terms for metadata.
  • Agents and Roles: Models agents (prov:Agent) involved (creators, executors) and their roles over time (pko:RoleInTime, using PRO).
  • Tools: Represents tools required for steps using m4i:Tool from the Metadata4Ing ontology.
  • Execution Events: Captures occurrences during execution, such as pko:IssueOccurrence (representing errors or problems encountered), pko:UserFeedbackOccurrence (recording user comments or ratings), and pko:UserQuestionOccurrence (logging questions asked by operators during execution).
• Industry-Specific Module (pko-ind): A separate module extends the core PKO with concepts tailored to the initial industrial use cases, including pko-ind:Machine, pko-ind:Device, pko-ind:Location, pko-ind:PersonalProtectiveEquipment, LOTO-specific items (pko-ind:Padlock), and pko-ind:EnergySource. This modularity allows for domain-specific extensions without altering the core ontology.

Design Methodology and Development

PKO was engineered using the Linked Open Terms (LOT) methodology, an iterative process suited for industrial ontology development.

1. Requirements Specification: Workshops with domain experts from the target use cases yielded user stories, relevant documentation, 54 initial competency questions (CQs), and 67 facts representing knowledge requirements. These were validated by experts to ensure alignment with practical needs. Examples of CQs include "What are the steps of procedure X?", "What tools are needed for step Y?", "Who executed procedure Z and when?", "What issues were reported during the execution of procedure W?".
2. Ontology Implementation: An initial conceptual model was developed, clustering concepts related to Procedure, Step, Status Change, Execution, Resource, and Agent. Existing ontologies were reviewed, and relevant terms were selected for reuse (soft reuse via URI referencing). The model was formalized in OWL using the Chowlk tool, facilitating diagram-to-code transformation. Modularity was maintained by separating pko-core from pko-ind.
3. Ontology Publication and Documentation: PKO is published online using persistent URIs managed by w3id.org. It is available in standard RDF syntaxes (Turtle, RDF/XML, JSON-LD, N-Triples) and licensed under CC BY-SA 4.0. Documentation was automatically generated using Widoco. Version control and issue tracking are managed via a public GitHub repository.
4. Ontology Maintenance: The GitHub repository serves as the platform for collecting feedback, bug reports, and suggestions for future extensions, ensuring the ontology can evolve based on further piloting and adoption.

Integration with Existing Ontologies

PKO leverages and extends several established ontologies to ensure interoperability and avoid redefining well-understood concepts:

• PROV-O: Provides foundational classes like prov:Activity (superclass for executions) and prov:Agent, and properties for provenance tracking.
• P-Plan: Extends PROV-O for plan representation. PKO subclasses p-plan:Plan for procedures and relates pko:Step to p-plan:Step, utilizing P-Plan's structure for relating plans, steps, and executions.
• DCAT & DCMI: Used for describing associated resources (documents, media) and their metadata.
• Time Ontology (OWL-Time): Provides vocabulary for representing temporal aspects of executions (instants, intervals, durations).
• ADMS: Used for metadata about ontology assets and distributions.
• Metadata4Ing (M4I): Provides the m4i:Tool class for representing tools used in steps.
• Publishing Roles Ontology (PRO): Used to model roles held by agents (pko:RoleInTime).

PKO extends these by providing a more granular model specifically for industrial procedures, notably the clear distinction between the specification (Procedure, Step) and execution (ProcedureExecution, StepExecution) aspects. It introduces domain-specific classes and properties required by the industrial CQs, such as different status types, execution event tracking (IssueOccurrence, UserFeedbackOccurrence, UserQuestionOccurrence), ExpertiseLevel, StepVerification, and the concepts within the pko-ind module.

Industrial Application and Relevance

PKO directly addresses practical industrial challenges related to managing tacit knowledge, ensuring compliance, reducing operational errors, and facilitating knowledge transfer. Its relevance is demonstrated through two implemented applications:

1. PK Elicitation Tool: A web-based form dynamically generated from the PKO structure. It guides domain experts through the process of documenting procedures step-by-step. User inputs are transparently mapped to PKO concepts, generating a PKO-compliant Knowledge Graph (KG) without requiring the user to have KG expertise. This tool facilitates the conversion of implicit or unstructured PK into explicit, structured, and semantically rich data.
2. PK Exploitation Chatbot: A conversational AI system designed to assist operators during procedure execution. It leverages a KG populated with PKO-modeled procedures. The chatbot employs a KG-empowered Retrieval-Augmented Generation (KG-RAG) approach. Operator queries are translated into SPARQL queries against the KG. The retrieved structured information (e.g., specific steps, required tools, safety warnings) is used to ground the responses generated by a LLM, mitigating hallucination risks and providing accurate, context-specific guidance. Steps can be presented interactively, for instance, as cards in the chat interface.

The paper also notes interest from external industrial actors in manufacturing and other sectors (media, cultural heritage, agriculture), indicating broader potential applicability, including potential integration with Digital Twins to describe service capabilities.

Impact on AI and Data-Driven Systems

PKO serves as a foundational semantic layer for developing more sophisticated AI and data-driven tools for PK management. Its potential impact includes:

• Interoperability: Provides a common vocabulary and structure, enabling seamless data exchange between different tools and systems involved in the PK lifecycle (e.g., authoring tools, execution guidance systems, analytics platforms).
• Enhanced Data Quality: Structuring PK according to PKO yields high-quality, semantic data readily consumable by AI algorithms for tasks like training predictive models (e.g., predicting execution time or failure likelihood), performing analytics on execution logs (e.g., identifying bottlenecks, common errors, compliance deviations), or grounding generative AI systems as demonstrated by the KG-RAG chatbot.
• Reasoning Capabilities: The OWL-based formalization enables logical reasoning and inference over the PK knowledge graphs. This can be used for validating procedure consistency, checking compliance against standards, optimizing procedure workflows, or providing more intelligent, context-aware execution support.
• Knowledge Reuse: Explicit modeling of procedures and steps facilitates their discovery, adaptation, and reuse across different contexts, machines, or product lines, promoting efficiency and standardization.
• Stimulating Tool Development: PKO acts as a reference model, encouraging the development of advanced PK management tools incorporating features like automated extraction of procedures from documentation or logs, semantic search over procedures, adaptive user guidance based on skill level or real-time conditions, and automated compliance verification.

Conclusion

In conclusion, the Procedural Knowledge Ontology (PKO) offers a robust and extensible semantic framework for representing industrial procedures and their executions. By leveraging existing standard ontologies and incorporating specific concepts derived from industrial requirements, PKO provides a structured, interoperable foundation for managing the procedural knowledge lifecycle. Its practical utility is demonstrated through applications for knowledge elicitation and AI-powered execution support, highlighting its potential to enhance operational efficiency, safety, and knowledge management in industrial settings through better integration with AI and data-driven systems.