Artificial Inventiveness: Towards AI supported Model-Driven Systematic Conceptual Design

Abstract

The Conceptual Design tools are between their historical qualitative if not disruptive changes. Being historically characterized as creativity and non-standard thinking, the ability to come up with inventive design decisions have been significantly strengthened with systematic and formal Conceptual Design tools during the last decades while still not yet fully parameterized or formalised. Many different softwares and algorithms based on the automatic analysis of the patents, scientific publications and other textual databases as well as attempts to digitalise cause-effect chains, problem and solution models are already being industrially adopted. However, the last decade of the exponential growth of the artificial intelligence approaches and especially word embeddings paves the way for a conceptually new approach to the Conceptual Design tools. There are many attempts to characterise and parametrise the inventing process with the help of AI however it is only the initial stage of the inventing capability of the AI, the main power of which will still be unleashed in future.

The present research aims to investigate the capability of integration between systematic Conceptual Design tools with the modern AI technologies. The research includes the investigation of the word embeddings for parameterization and characterisation of the patents and scientific publications databases and navigation interface through such a semantic model of extracted concepts. The methods used in the research includes qualitative analysis and approbation of the heuristic and systematic Conceptual Design models such as TRIZ in case studies in different engineering fields. The analysis of such an heuristic approach is also studied with the group of physicists in the field of Quantum mechanics. Finally, the quantitative research including statistical methods, and AI-models including word embeddings were utilised to build a custom database of patents and scientific publications and to explore such a database opportunities for a model-driven systematic Conceptual Design. The achieved results of extracting meaningful and insightful semantic models of invention concepts based on the natural language processing and machine learning techniques including word embeddings analysis contributes to the investigation of the fully machine guided inventing process.