LLM-Supported Formal Knowledge Representation for Enhancing Control Engineering Content with an Interactive Semantic Layer
Abstract
The rapid growth of research output in control engineering calls for new approaches to structure and formalize domain knowledge. We therefore propose an LLM-supported method for semi-automated generation of a formal knowledge representation that combines readability by humans with machine interpretability and high expressiveness. Based on the Imperative Representation of Knowledge (PyIRK) framework, we demonstrate how language models can assist in transforming natural-language descriptions and mathematical statements (available as LaTeX source code) into a formalized knowledge graph. As a first application we present the generation of an "interactive semantic layer" to enhance the source documents in order to facilitate comprehensibility and knowledge transfer. From our perspective this contributes to the vision of easily accessible, collaborative, and verifiable knowledge bases for the control engineering domain.
Full text
LLM-Supported Formal Knowledge Representation for Enhancing Control Engineering Content with an Interactive Semantic Layer Julius Fiedler, Carsten Knoll , Klaus Röbenack TU Dresden GAMM FA RSE & RDM Kickoff Meeting 2025-12-05
LLM-Supported Formal Knowledge Representation for Enhancing Control Engineering Content with an Interactive Semantic Layer, Institut of Control Theory, TU Dresden, Julius Fiedler, Carsten Knoll, Klaus Röbenack GAMM RSE & RDM Kickoff Meeting 2025-12-05 Slide 2/14 Why Formal Knowledge Representation? Control theory has a wide spectrum of methods •PID controller, sliding mode, backstepping, model predictive control, … Control theory has a wide range of application domains •process engineering, robotics, civil engineering, automotive, … Permanent growth of knowledge (publications) enforces specialization Knowledge transfer is non-trivial •within control theory •into application domains Problem: Current representation of knowledge is suboptimal •natural language text, formulas, figures →Usage of semantic technologies as complement
LLM-Supported Formal Knowledge Representation for Enhancing Control Engineering Content with an Interactive Semantic Layer, Institut of Control Theory, TU Dresden, Julius Fiedler, Carsten Knoll, Klaus Röbenack GAMM RSE & RDM Kickoff Meeting 2025-12-05 Slide 3/14 What Is Formal Knowledge Representation? Goal: •formally represent knowledge make it machine processible→ Formalization approach: Ontology •„Formal, explicit specification of a shared conceptualization [of a knowledge domain].“ [Studer et. al. 1998] •Which concepts exist? How are they related? Representation as directed graph (“knowledge graph”) •nodes: concepts •edges: semantic relations →modeling of subject-predicate-object triples integer number real number Scalar is a is a R. Studer, V.R. Benjamins, D. Fensel, Knowledge engineering: Principles and methods, Data & Knowledge Engineering, Volume 25, Issues 1–2, 1998, Pages 161-197.
LLM-Supported Formal Knowledge Representation for Enhancing Control Engineering Content with an Interactive Semantic Layer, Institut of Control Theory, TU Dresden, Julius Fiedler, Carsten Knoll, Klaus Röbenack GAMM RSE & RDM Kickoff Meeting 2025-12-05 Slide 4/14 Outline 0. Motivation 1. Our Framework for Knowledge Representation – PyIRK 2. Semi-automated Formalization Workflow 3. Example
LLM-Supported Formal Knowledge Representation for Enhancing Control Engineering Content with an Interactive Semantic Layer, Institut of Control Theory, TU Dresden, Julius Fiedler, Carsten Knoll, Klaus Röbenack GAMM RSE & RDM Kickoff Meeting 2025-12-05 Slide 5/14 Imperative Representation of Knowledge – PyIRK Motivation for developing our own framework [Knoll et al. 2024] •existing technologies (Web-Ontology-Language OWL): limited expressive power (deliberately) •far goal: control engineering assistant with reliable transparent collaboratively managed knowledge base Understandable to humans and machines •every node (item) and edge (relation) has a unique identifier + „human-readable“ label: e.g. I35["real number"] Hurdles: •formalizing knowledge (in the aspired depth) requires significant effort •open question: How to generate utility value with limited effort? Knoll et al.: Imperative Formal Knowledge Representation for Control Engineering: Examples from Lyapunov Theory.NMachinesN2024,N12, 181. https://doi.org/10.3390/machines12030181
LLM-Supported Formal Knowledge Representation for Enhancing Control Engineering Content with an Interactive Semantic Layer, Institut of Control Theory, TU Dresden, Julius Fiedler, Carsten Knoll, Klaus Röbenack GAMM RSE & RDM Kickoff Meeting 2025-12-05 Slide 6/14 Problem Knowledge Formalized Knowledge application? far goal: assistance system near goal: enriching source documents paper or book ontology/knowledge graph formalize use query language SPARQL, ... semi-automated process
LLM-Supported Formal Knowledge Representation for Enhancing Control Engineering Content with an Interactive Semantic Layer, Institut of Control Theory, TU Dresden, Julius Fiedler, Carsten Knoll, Klaus Röbenack GAMM RSE & RDM Kickoff Meeting 2025-12-05 Slide 7/14 Semiautomatic Formalization Process (1) •PDF-to-LaTeX-Conversion: possible but nontrivial •End-to-End-approach not viable •„Formalized Natural Language“ with triple-structure as intermediate step
LLM-Supported Formal Knowledge Representation for Enhancing Control Engineering Content with an Interactive Semantic Layer, Institut of Control Theory, TU Dresden, Julius Fiedler, Carsten Knoll, Klaus Röbenack GAMM RSE & RDM Kickoff Meeting 2025-12-05 Slide 8/14 Semiautomatic Formalization Process (2) Prompt: •introduction and purpose •allowed statements •already formalized statements •current LaTeX snippet •instruction: formalise its content … Allowed Statements: - There is a class: <arg1>. - There is a property: <arg1>. - There is a relation: <arg1>. - There is a general operator: <arg1>. …
LLM-Supported Formal Knowledge Representation for Enhancing Control Engineering Content with an Interactive Semantic Layer, Institut of Control Theory, TU Dresden, Julius Fiedler, Carsten Knoll, Klaus Röbenack GAMM RSE & RDM Kickoff Meeting 2025-12-05 Slide 9/14 Semiautomatic Formalization Process (3) Let ${\mathbb{R}}$ be the set of real numbers. - There is a class: 'real number' - There is a class: 'set of real numbers' - 'set of real numbers' is an instance of 'set' - 'set of real numbers' 'has element type' 'real number' Let ${\mathbb{R}}$ be the \setref{set of real numbers}{label:set_of_real_numbers}.
LLM-Supported Formal Knowledge Representation for Enhancing Control Engineering Content with an Interactive Semantic Layer, Institut of Control Theory, TU Dresden, Julius Fiedler, Carsten Knoll, Klaus Röbenack GAMM RSE & RDM Kickoff Meeting 2025-12-05 Slide 16/14 Appendix – Current OCSE Knowledge Graph > 2000 nodes > 160 relations > 3000 subject-predicate-object triple