Bringing Together Engineering Problems and Basic Science Knowledge, One Step Closer to Systematic Invention

Abstract

Since its origins, TRIZ theory has been concerned with the use of fundamental knowledge of physics as a means of solving engineering problems. The three decades of TRIZ history have seen the emergence of methodological tools such as substance-field analysis combined with databases that have become increasingly computerized in line with advances in computer science. However, the current revival of artificial intelligence calls into question everything that has been done previously in terms of classification and allows us to think about the pairing of engineering problems and knowledge of physics not from closed databases, but in real time from online data sources and according to the versatility of web content. This article presents a new approach to pairing called PhysiSolve based on Artificial Intelligence techniques. We used natural language processing models like transformers based on attention to boost learning which allows us to outperform classical models for downstream tasks and unlock technical language understanding to automate data classification and facilitate semantic search for better ideas generation. Our research has led us to develop an online tool whose first results are presented and discussed from a perspective of measuring the efficiency of conducting an inventive activity. These results reinforce our belief that artificial assistance to inventiveness in R&D is no longer just possible but paves the way for a new era of digital tools for engineers and industrial companies.