Evaluating TRIZ with and without LLM support: an experimental study on engineering problem-solving

This paper examines integrating Large Language Models (LLMs) into the TRIZ contradiction matrix (TRIZ-C+LLM) to support engineering students in creative problem-solving. Experiments with three problems show that LLMs did not always improve design quality for complex tasks but reduced cognitive workload, improved understanding of contradictions, and increased perceived usefulness. Prompting experience strongly influenced outcomes, highlighting both the promise and limits of combining TRIZ with generative AI.