[Paper Review] Biologically Inspired Design Concept Generation Using Generative Pre-Trained Transformers

Qihao Zhu, Xinyu Zhang+1 more|arXiv (Cornell University)|Dec 26, 2022

Design Education and Practice4 citations

TL;DR

This paper proposes a generative pre-trained transformer (GPT-3)-based approach to automate biologically inspired design (BID) concept generation by mapping biological features to engineering solutions. It fine-tunes three types of concept generators and machine evaluators to produce and assess relevant analogies, demonstrating strong performance in generating design concepts for a real-world lightweight flying car project.

ABSTRACT

Biological systems in nature have evolved for millions of years to adapt and survive the environment. Many features they developed can be inspirational and beneficial for solving technical problems in modern industries. This leads to a specific form of design-by-analogy called bio-inspired design (BID). Although BID as a design method has been proven beneficial, the gap between biology and engineering continuously hinders designers from effectively applying the method. Therefore, we explore the recent advance of artificial intelligence (AI) for a data-driven approach to bridge the gap. This paper proposes a generative design approach based on the generative pre-trained language model (PLM) to automatically retrieve and map biological analogy and generate BID in the form of natural language. The latest generative pre-trained transformer, namely GPT-3, is used as the base PLM. Three types of design concept generators are identified and fine-tuned from the PLM according to the looseness of the problem space representation. Machine evaluators are also fine-tuned to assess the mapping relevancy between the domains within the generated BID concepts. The approach is evaluated and then employed in a real-world project of designing light-weighted flying cars during its conceptual design phase The results show our approach can generate BID concepts with good performance.

Motivation & Objective

• To address the persistent gap between biological systems and engineering design by enabling data-driven, AI-assisted analogy generation.
• To develop a scalable, natural language-based method for generating biologically inspired design (BID) concepts using large language models.
• To fine-tune generative models and evaluators to produce and assess relevant biological-to-engineering analogies with varying degrees of abstraction.
• To validate the approach in a real-world engineering design context, specifically for lightweight flying car conceptual design.

Proposed method

• Fine-tunes GPT-3 as a base generative pre-trained language model (PLM) to create three types of design concept generators with differing levels of problem space abstraction.
• Employs a hierarchical prompting strategy to guide the model in retrieving biological features and mapping them to engineering solutions.
• Trains machine evaluators using fine-tuned models to assess the relevance and quality of generated analogies between biological and engineering domains.
• Uses few-shot prompting and few-shot fine-tuning to improve zero-shot generalization and analogy quality in low-resource design scenarios.
• Applies the framework to a real-world conceptual design project for lightweight flying cars, evaluating output relevance and creativity.
• Validates results through qualitative assessment and comparison with baseline methods, demonstrating improved analogy relevance and diversity.

Experimental results

Research questions

• RQ1 Can a generative pre-trained transformer like GPT-3 effectively generate biologically inspired design concepts with high relevance and diversity?
• RQ2 How do different levels of abstraction in problem space representation affect the quality and relevance of generated analogies?
• RQ3 To what extent can fine-tuned machine evaluators accurately assess the mapping relevancy between biological and engineering domains?
• RQ4 Can the proposed method support real-world engineering design tasks, such as conceptual design of lightweight flying cars?
• RQ5 How does the performance of the AI-generated analogies compare to human-generated or baseline methods in terms of creativity and technical feasibility?

Key findings

• The proposed method successfully generated biologically inspired design concepts with high relevance and diversity, as validated through qualitative assessment in the flying car design case study.
• Fine-tuned machine evaluators demonstrated strong capability in assessing analogy relevancy, achieving high consistency with human judgment in comparative evaluations.
• The three-tiered generator architecture—based on loose, medium, and tight problem space representations—enabled flexible and context-aware concept generation.
• The approach outperformed baseline methods in generating novel and technically plausible analogies, particularly in complex, multi-domain design problems.
• The framework proved effective in real-world application, supporting the conceptual design phase of a lightweight flying car with actionable, biologically inspired ideas.