Page 12 - IJAMD-1-3

International Journal of AI for
Materials and Design
Smart cockpit design with generative models

The rest of the paper is organized as follows: Section 2 engagement, and collaborative information tools, are
6
20
reviews the relevant studies. Section 3 outlines the system vital in achieving personalized design, particularly in
framework and implementation details of the proposed the context of new product development. Nevertheless,
personalized design framework. In Section 4, a case study of current strategies often restrict consumers to limited
personalized vehicle seat design for smart vehicle cockpits choices, such as altering product colors or selecting
is presented. Limitations and future work are discussed in configuration components. These approaches fall short
Section 5, and conclusions are provided in Section 6. of fully addressing consumers’ affective and functional
needs. This study aims to empower consumers to engage
2. Literature review proactively and immersively in personalized product
The main focus of this research is achieving personalized design through natural language interaction, focusing on
design for smart vehicle cockpits through generative models, both functions and design style aspects.
MAS, and HDT models. The state-of-the-art research 2.2. Applications of generative models in design
is reviewed from three main aspects: (i) Personalized
design methods, (ii) applications of generative models in Large language models (LLMs), including ChatGPT,
design, and (iii) HDT-enabled design. Finally, the research BERT, 21 LLaMA 22, and other GAI techniques
identifies existing gaps in the literature. (e.g., DALL·E, Midjourney), belong to the category
of generative models. These models can efficiently
2.1. Personalized design methods produce high-quality, multimodal digitalized content at
6
With advancements in technology and production a rapid pace by interpreting human-guided instructions.
techniques, competition in an ever-changing and Consequently, they present significant opportunities for
23
unpredictable global market has compelled manufacturers design innovation. For instance, Wang et al. introduced
to transform their production models. They have a task-decomposed approach that integrates LLMs with
12
shifted their focus from products to customers, offering the function-behavior-structure (FBS) model to inspire
personalized design solutions and enabling customer designers during the design conceptualization phase. Jiang
participation in the product design process. Concurrently, et al. proposed AutoTRIZ, a design ideation tool powered
24
numerous studies on personalized design have been by LLMs, to facilitate automated design and ideation
25
conducted. For instance, Dou et al. introduced an processes. In addition, Yin et al. utilized Midjourney, a
5
interactive genetic algorithm that allows users to actively text-to-image generative model, to support collaborative
engage in the personalized product design process, applying design in product-service systems.
it to a personalized design system for automobile wheel However, leveraging a single LLM or multi-foundation
hubs. The experimental results indicated that the algorithm model sometimes fails to effectively accomplish complex
enhanced system performance. Moreover, configuration tasks. Consequently, generative model-based MAS are
has been recognized as a mainstream approach to gaining popularity. These systems consist of multiple
15
achieving personalized design. 13,14 Dong et al. proposed agents enhanced by LLMs or generative models that
a knowledge graph-based configuration approach for mass collaborate or compete to handle complicated design
personalization, considering users’ affective and functional tasks more effectively. AutoGen is an emerging multi-
26
demands. Hu et al. employed a heuristic process planning agent architecture for building generative applications,
16
method to achieve personalized product configuration supporting customized and conversational agents
17
design. Ren et al. introduced a proactive interaction composed of generative models, human inputs, and
method to recommend personalized services based on tools. This enables the system to assign different roles
context-aware prediction in the targeted service scenarios. in processing domain-specific human expertise and
In addition, Zheng et al. presented a three-model-based integrating robust generative capabilities. For instance,
18
27
generic framework for UX-oriented product development Chen et al. combined generative models (i.e., LLM and
28
for mass personalization, demonstrating its feasibility Stable Diffusion 2.1) with the Who, What, Where, When,
through a case study of a personalized smart wearable Why, How (5W1H) method, FBS model, and Kansei
product. Zheng et al. proposed a universal data-centric Engineering to generate innovative design solutions.
19
approach for developing personalized smart connected Celen et al. introduced I-Design, a personalized LLM-
29
products within a cloud-based environment. based interior designer that converts user prompts into
The essence of personalized products lies in addressing personalized interior design scenes, further developing
customer demands and enhancing user satisfaction. them into 3D design scenarios. Xu et al. employed LLM-
30
Customer-oriented strategies, such as configuration, enabled MAS to simulate design sessions among different
interactive genetic algorithms for proactive user stakeholders during the collaborative design process.

Volume 1 Issue 3 (2024) 6 doi: 10.36922/ijamd.4220