Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).
Please note that all times are shown in the time zone of the conference. The current conference time is: 17th May 2024, 11:46:47am CEST
Session Chair: John Gero, UNC Charlotte, United States of America
Location:Congress Hall Ragusa
Presentations
Large language models in complex system design
Alejandro Pradas Gomez1, Petter Krus2, Massimo Panarotto1, Ola Isaksson1
1Chalmers University of Technology, Sweden; 2Linköping University, Sweden
This paper investigates the use of Large Language Models (LLMs) in engineering complex systems, demonstrating how they can support designers on detail design phases. Two aerospace cases, a system architecture definition and a CAD model generation activities are studied. The research reveals LLMs' challenges and opportunities to support designers, and future research areas to further improve their application in engineering tasks. It emphasizes the new paradigm of LLMs support compared to traditional Machine Learning techniques, as they can successfully perform tasks with just a few examples.
Automatic identification of role-specific information in product development: a critical review on large language models
In the era of digitization and the growing flood of information, the automatic, role-specific identification of information is crucial. This research paper aims to investigate whether the adaptation of LLM is suitable for classifying information obtained from standards for corresponding role profiles. This research reveals that with systematic fine-tuning, prediction accuracy can be increased by almost 100%. The validation was carried out using a two-digit number of standards for three predefined roles and demonstrates the significant potential of LM for labelling content with regard to roles.
Benchmarking AI design skills: insights from ChatGPT’s participation in a prototyping hackathon
Daniel Nygård Ege, Henrik H. Øvrebø, Vegar Stubberud, Martin Francis Berg, Martin Steinert, Håvard Vestad
Norwegian University of Science and Technology, Norway
This study provides insights into the capabilities and performance of generative AI, specifically ChatGPT, in engineering design. ChatGPT participated in a 48-hour hackathon by instructing two participants who acted out its instructions, successfully designing and prototyping a NERF dart launcher that finished second among six teams. The paper highlights the potential and limitations of generative AI as a tool for ideation, decision-making, and optimization in engineering tasks, demonstrating the practical applicability of generating viable design solutions under real-world constraints.
How good is ChatGPT? An exploratory study on ChatGPT’s performance in engineering design tasks and subjective decision-making
Wanyu Xu, Maulik Chhabilkumar Kotecha, Daniel A. McAdams
Texas A&M University, United States of America
This study explores how large language models like ChatGPT comprehend language and assess information. Through two experiments, we compare ChatGPT's performance with humans', addressing two key questions: 1) How does ChatGPT compare with human raters in evaluating judgment-based tasks like speculative technology realization? 2) How well does ChatGPT extract technical knowledge from non-technical content, such as mining speculative technologies from text, compared to humans? Results suggest ChatGPT's promise in knowledge extraction but also reveal a disparity with humans in decision-making.
Datasets in design research: needs and challenges and the role of AI and GPT in filling the gaps
Mohammad Arjomandi Rad, Tina Hajali, Julian Martinsson Bonde, Massimo Panarotto, Kristina Wärmefjord, Johan Malmqvist, Ola Isaksson
Chalmers University of Technology, Sweden
Despite the recognized importance of datasets in data-driven design approaches, their extensive study remains limited. We review the current landscape of design datasets and highlight the ongoing need for larger and more comprehensive datasets. Three categories of challenges in dataset development are identified. Analyses show critical dataset gaps in design process where future studies can be directed. Synthetic and end-to-end datasets are suggested as two less explored avenues. The recent application of Generative Pretrained Transformers (GPT) shows their potential in addressing these needs.
Nature’s lessons, AI’s power: sustainable process design with generative AI
Mas'udah Mas'udah, Pavel Livotov
Offenburg University of Applied Sciences, Germany
In the realm of process engineering, the pursuit of sustainability is paramount. Traditional approaches can be time-consuming and often struggle to address modern environmental challenges effectively. This article explores the integration of generative AI, as a powerful tool to generate solution ideas and solve problems in process engineering using a Solution-Driven Approach (SDA). SDA applies nature-inspired principles to tackle intricate engineering challenges. In this study, generative AI is trained to understand and use the SDA patterns to suggest solutions to complex engineering challenges.