AIARD
FrançaisEnglish

Research

Research

Can a machine help formulate an inventive problem, find the knowledge that resolves it, and argue for a solution an engineer would trust?

Research themes

T1

Neuro-symbolic inventive reasoning

Language models are fluent; TRIZ is rigorous. Can a machine be both?

  • Established

    Can an LLM execute ARIZ step by step and produce concepts that experts judge novel?

T2

Industrial knowledge and problem graphs

Before solving, map the problem.

  • Ongoing

    How do engineering teams formulate problems, as recorded in FINDER graphs?

T3

Patent and scientific literature intelligence

Millions of solved problems, written in a language machines must learn.

  • Established

    Do patent-specialised embeddings retrieve solutions better than generic ones?

T4

Interdisciplinary knowledge discovery

The answer often lives in another discipline.

  • Established

    Can interdisciplinarity be quantified from the text of an article?

T5

Human–AI collaboration in invention

The engineer remains the judge.

  • Ongoing

    How should AI suggestions be presented so that teams keep control of the reasoning?

T6

Tech scouting in energy

Who is inventing what, and how do we know?

  • Established

    Can fine-tuned NER models detect new technologies with limited French data?

Evidence levelEstablished — Published or demonstratedOngoing — Research under wayAmbition — Part of the programme, not yet started

Our scientific approach

From problems to inventions.

AIARD connects problems to knowledge, and knowledge to invention. Each step below is a research object in its own right, with its own models, data and evaluation.

Start from real situations brought by partners, with their constraints, history and vocabulary.

Problem situation

● The Invention Graph — Shortening injection-moulding cycles

Choose a stage or a node to read why it is there.

Fig. 01 [Methodological illustration] Shortening injection-moulding cycles. Each piece of knowledge rests on real patents and articles (document icons): hover over it, or step to 04, to see them. The directions shown are documented in the engineering literature; they illustrate the reasoning, they are not AIARD results.

One research chain, six research objects

  1. 01

    Industrial data

    Problem situations, documents and expertise from partners.

  2. 02

    Knowledge extraction

    Patents and articles turned into problems, parameters and solutions.

  3. 03

    Problem modelling

    Graphs of problems, partial solutions and parameters.

  4. 04

    Contradiction reasoning

    Detecting and qualifying the contradictions that block progress.

  5. 05

    Concept generation

    Connecting contradictions to knowledge, near and far.

  6. 06

    Human evaluation

    Engineers judge, select and refine; their judgement is data too.

Research axes announced since the Chair was created

  • A1Extraction and digitisation of knowledge from all forms of documents
  • A2Automatic pairing between problem models and solution models built from that knowledge
  • A3Permanent enrichment of the data

Research projects and theses

Each project states its research question, its method and, separately, what it has actually shown.

PhD thesis · 2014Completed

(Meta)Knowledge modeling for inventive design

Doctoral thesis in computer science, defended on 7 February 2014 at the University of Strasbourg, supervised by Denis Cavallucci and Pierre Collet (theses.fr 2014STRAD006).

PhD thesis · 2019Completed

Experience capitalization to support inventive design studies

Doctoral thesis in industrial engineering (design), defended on 18 February 2019 at the University of Strasbourg, supervised by Denis Cavallucci and Cecilia Zanni-Merk (theses.fr 2019STRAD002). Followed by a postdoc for the Chair.

PhD thesis · 2025Completed

Patent intelligence and LLM orchestration of ARIZ

A doctoral thesis on retrieving the patents that answer an inventive problem, and on running ARIZ with language models to generate concepts anchored in those patents. Defended in December 2025.

PhD thesis · –2026Ongoing

Text mining and topic modelling for innovation analysis

A doctoral thesis on detecting technologies and innovative start-ups in renewable energy with NER and topic models, and on a transparent innovation index. Defence scheduled on 5 November 2026 at 2 pm.

PhD thesis funded byEDF

PhD thesis · 2025–Ongoing

Towards an innovative methodology for knowledge fusion from patent datasets: application to inventive design

A doctoral thesis in computer science, started in 2025 and supervised by Denis Cavallucci and Hicham Chibane, on fusing the knowledge held in patent datasets to support inventive design. The thesis designs the graph-building algorithm that turns texts into problem graphs, the engine of TRIZBench; first work on extracting problem graphs and causal maps with language models.

PhD thesis funded byArcelorMittal
All publicationsPeople