Enhancing Conceptual Design with LLM-Augmented Morphological Analysis: A New Era in Human-AI Collaboration

In the ever-evolving landscape of product design, conceptual design remains a pivotal phase where abstract ideas are transformed into structured solutions. However, traditional methods often place heavy cognitive demands on designers and may limit creative exploration due to reliance on individual expertise. A groundbreaking study presented at the DRS2024 Biennial Conference proposes an innovative solution: an LLM-augmented morphological analysis approach that enhances creativity, reduces mental workload, and improves human-AI collaboration.

This article explores the methodology, experimental validation, and implications of this new framework for conceptual design, drawing from the research paper titled "An LLMs-augmented Morphological Analysis Approach for Conceptual Design" by Liuqing Chen et al. (2024), published in the DRS Digital Library .

---

What is Morphological Analysis?

Morphological analysis (MA) is a systematic method used in conceptual design to break down complex problems into manageable sub-functions, generate potential solutions for each, and combine them into holistic design concepts. This technique, originally introduced by Fritz Zwicky in 1943, has been widely applied across engineering and design disciplines due to its structured yet exploratory nature (Zwicky, 1948; Pahl & Beitz, 2013).

However, MA traditionally relies heavily on the designer’s domain knowledge and experience. It also suffers from issues like subjective function decomposition and limited scalability when dealing with unfamiliar or highly complex problems (Fiorineschi et al., 2016a).

---

The Rise of Large Language Models in Design

The emergence of Large Language Models (LLMs) such as GPT-4 has opened new avenues for augmenting human creativity. These models can generate diverse, contextually rich content, synthesize cross-disciplinary knowledge, and support idea generation beyond individual cognitive limits (Zhao et al., 2023; Dwivedi et al., 2023). In design, LLMs have already shown promise in concept ideation, text-to-image synthesis, and even narrative-based design storytelling.

Building on these capabilities, the authors propose a novel integration of LLMs into the morphological analysis process, enhancing both the depth and breadth of design exploration.

---

The MA-GPT Framework: A Three-Stage Process

The proposed MA-GPT approach integrates LLMs into the three core stages of morphological analysis: decomposition , generation , and combination . Each stage is supported by tailored prompt templates designed to guide AI assistance while maintaining human control and intent.

1. Decomposition: Structuring the Problem Space

At this stage, the goal is to decompose a design problem into functional components using a standardized vocabulary. The researchers employed the functional basis method (Stone & Wood, 1999), which uses a "verb + noun" format to define functions (e.g., “generate airflow”) and categorizes them under material flow , energy flow , and information flow .

Prompt Template:

“Help me design a {design task} that outlines its primary functions in terms of material flow, energy flow, and information flow in accordance with the functional decomposition theory.”

By standardizing the language of decomposition, the system ensures consistency and clarity between the designer and the LLM, reducing ambiguity and improving downstream solution quality.

2. Generation: Ideating Sub-Function Solutions

Once the functional structure is defined, the next step involves generating possible means (i.e., solutions) for each sub-function. Traditional brainstorming is often constrained by personal biases and prior experiences. LLMs overcome this by rapidly proposing a wide range of alternatives informed by vast datasets.

Prompt Template:

“For the {Function}, brainstorm potential means that align with the product's design goals.”

Designers then curate and refine these suggestions based on feasibility and relevance, ensuring alignment with project constraints.

3. Combination: Synthesizing Holistic Concepts

The final stage combines selected solutions into coherent design concepts. Given the combinatorial explosion of options, evaluating all permutations manually is impractical. Here, LLMs act as evaluators, assessing combinations for feasibility , efficiency , and conflict resolution .

Prompt Template:

“Given the selected means for each function {means}, evaluate the combination for feasibility, efficiency, and potential design conflicts. Provide suggestions or modifications to improve the combination.”

This feedback loop enables designers to iteratively refine their concepts, leveraging AI insights without losing creative agency.

---

Experimental Validation: How Does MA-GPT Perform?

To assess the effectiveness of the MA-GPT framework, the researchers conducted a controlled experiment involving 15 professional designers tasked with designing an automated home cat feeder. Participants were divided into three groups:

• MA Group : Traditional morphological analysis without AI.

• Free-GPT Group : Unstructured interaction with GPT-4.

• MA-GPT Group : The proposed LLM-augmented morphological analysis.

Key Evaluation Metrics:

• Design Quality : Measured using a modified Pugh Matrix (Güler & Petrisor, 2021), considering novelty , feasibility , usability , and functionality .

• Cognitive Load : Assessed via NASA-TLX (Hart, 2006).

• Human-AI Collaboration Experience : Evaluated through a survey measuring controllability , transparency , solution diversity , and creative stimulation .

Results:

The MA-GPT group significantly outperformed both baseline groups in overall design quality and demonstrated a marked reduction in cognitive load. Furthermore, participants reported higher satisfaction with the transparency and controllability of the MA-GPT process compared to free-form AI interaction.

---

Why MA-GPT Works: Insights from Designer Feedback

Semi-structured interviews revealed several key advantages of the MA-GPT approach:

1. Structured Creativity : Designers appreciated the balance between guided prompts and open-ended exploration.

2. Knowledge Expansion : LLMs provided access to cross-domain knowledge, enabling more informed decision-making.

3. Iterative Refinement : The ability to provide feedback and receive refined suggestions allowed for deeper customization.

4. Reduced Mental Fatigue : By offloading repetitive tasks and idea generation, designers could focus on strategic decisions.

One participant noted:

“MA-GPT makes me feel like I'm participating in this design, rather than me mentioning requirements the LLM executes and generates the solution directly.”

---

Implications for Human-AI Co-Creation in Design

The success of MA-GPT underscores a broader trend: the shift from AI as a passive tool to AI as a collaborative partner in creative processes. Effective co-creation requires not just computational power but also interpretability , interactivity , and adaptability —qualities embedded in the MA-GPT framework.

Moreover, the structured use of prompt engineering allows for better control over AI outputs, making it a viable tool for novice designers and students who may lack deep domain expertise.

---

Future Directions

While the current implementation focuses on textual input and output, future work aims to extend MA-GPT to multimodal interactions , including visual inputs and generative image outputs. Researchers also plan to incorporate personalized prompting strategies and real-time feedback loops to enhance adaptability.

As the team notes:

“We will develop a robustly interactive, multimodal input-supporting tool that goes beyond merely supporting textual input. It will generate visual form matrices that deftly transform text into images.”

---

Conclusion

The integration of LLMs into morphological analysis represents a significant leap forward in conceptual design methodologies. By combining structured problem-solving with AI-driven ideation and evaluation, the MA-GPT framework offers a powerful tool for enhancing creativity, reducing cognitive load, and fostering meaningful human-AI collaboration.

This research not only validates the practical benefits of LLMs in design but also sets a precedent for how AI can be strategically integrated into creative workflows—enhancing rather than replacing human ingenuity.

---

References

• Chen, L., Tsang, Y., Jing, Q., & Sun, L. (2024). An LLMs-augmented Morphological Analysis Approach for Conceptual Design . DRS2024 Biennial Conference.

• Dwivedi, Y. K. et al. (2023). So what if ChatGPT wrote it? Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy. International Journal of Information Management.

• Fiorineschi, L., Rotini, F., & Rissone, P. (2016a). A new conceptual design approach for overcoming the flaws of functional decomposition and morphology. Journal of Engineering Design.

• Güler, K., & Petrisor, D. M. (2021). A Pugh Matrix based product development model for increased small design team efficiency. Cogent Engineering.

• Hart, S. G. (2006). NASA-Task Load Index (NASA-TLX); 20 years later. Proceedings of the Human Factors and Ergonomics Society Annual Meeting.

• Stone, R. B., & Wood, K. L. (1999). Development of a functional basis for design. Journal of Mechanical Design.

• Zhao, W. X. et al. (2023). A survey of large language models. arXiv preprint arXiv:2303.18223.

• Zwicky, F. (1948). The morphological method of analysis and construction. California Institute of Technology.

---

About the Authors

• Liuqing Chen – ZJU-100 Young Professor at Zhejiang University, focusing on AI-driven design and human-computer interaction.

• Yiyan Tsang – Master student researching AI for design and human-computer interaction.

• Qianzhi Jing – PhD student interested in human-computer interaction.

• Lingyun Sun – Professor working on artificial intelligence, design intelligence, and AI-generated content.