Capabilities and limitations of large language models (LLMs) in the calculation and optimization of welded joints

Abstract

The rapid development and integration of Artificial Intelligence (AI)-based tools, such as large language models (ChatGPT, Gemini, Claude), open up new opportunities for the automation and optimization of engineering processes. This paper analyzes the potential application of contemporary AI tools in the field of mechanical structures, with a specific focus on the load-bearing capacity calculation of welded joints in accordance with current standards. Through a series of tested engineering scenarios and advanced prompt engineering, the accuracy and reliability of the generated analytical expressions were examined, specifically analyzing the capability of AI tools to precisely calculate the load-bearing capacity of welded joints. In addition to the computational aspect, the paper explores the capabilities of these tools in formulating optimized recommendations for the geometric design of welded joints, aiming to improve both load-bearing capacity and manufacturability. The results indicate that although AI tools significantly accelerate the preliminary design and conceptual shaping phases, their application requires rigorous engineering verification due to their tendency to generate incorrect numerical values ("hallucinations"). In conclusion, the paper recommends a hybrid approach that combines the rapidity of artificial intelligence with the reliability of traditional engineering calculation software.