Development of MBD method that simultaneously considers power management, thermal performance, vehicle motion performance, etc by using the cross-domain model
-Part II: A Multivariate Optimization Case Study Based on Cross-Domain Model Results-

Aoto Utsumi･Shigemitsu Takahashi･Mayuka Ojima･Masayuki Kiyono (Nissan Motor)

In this study, we present a case study on the application of multivariate optimization to overall vehicle design that accounts for trade-offs among multiple performance attributes, using output results obtained from the cross-domain model CAEM (Collective Automotive Engineering Model), reported in Part 1, which integrates multiple performance domains, such as fuel economy and handling stability, belonging to largely different physical domains. The study confirms that overall vehicle design can be optimized with respect to the selection of design variables and modules.

Development of an MBD Method for Global Design Optimization Using Cross-Domain Models
-Report 3 Efficiency Improvement of Analysis, Evaluation, and Data Management Using LLMs-

Shigemitsu Takahashi･Aoto Utsumi･Masayuki Kiyono (Nissan Motor)

Global optimization utilizing cross-domain models significantly increases the amount of knowledge and decision-making speed required from engineers, making human capability the limiting factor. To address this issue, we structured data and interfaces on the premise of enabling both LLMs and engineers to access knowledge and execute analyses, thereby improving overall engineering efficiency.

Improving Bracket Design Efficiency Through CAE and AI Integration

Mikito Kawamura･Toshiki Terabe･Chika Kanba･Makoto Okawa･Takumi Sano (Toyota Motor)

In bracket design, reducing the development cycle while simultaneously satisfying multiple performance criteria such as noise and vibration (NV) and thermal stress is essential. Conventional CAE-driven design workflows, however, depend heavily on iterative trial-and-error evaluations and thus provide limited scope for time reduction. This study expedites the exploration of exhaust manifold bracket designs by integrating surrogate modeling and generative shape optimization with CAE simulations. We demonstrate that the combined approach significantly accelerates design space exploration and improves solution discovery, and we discuss the method’s practical benefits, limitations, and operational challenges encountered during implementation.

Development of a Design Space Exploration Method for Structural Members under Manufacturing Constraints

Tetsuya Ishigai Ishiga･Kohei Shintani･Mashio Taniguchi･Tomohito Sono･Sho Yamanaka･Hayata Morita (TOYOTA Mortor)

In recent automotive development, the use of shape optimization and machine learning has been increasingly explored to shorten development lead time. However, for practical vehicle development, a systematic process has not been fully established to generate feasible cross-sectional shapes within manufacturing constraints and to present multiple effective design candidates for decision-making. This paper proposes a design support process that integrates cross-section shape generation and design exploration. By using a generation method capable of producing realistic part-like shapes composed of flat regions and filleted regions, the proposed process enables efficient presentation of multiple design candidates in the early design stage.

Simultaneous Material and Shape Optimization of EV Battery Plates Using Surrogate Models Trained on Limited Samples

Hiroyuki Saito･Takayuki Michishita･Kotaro Kawajiri (AIZOTH)

In designing EV battery plates responsible for heat dissipation, both thermal and mechanical strength requirements must be satisfied simultaneously. This study presents a method that enables real-time prediction and simultaneous material and shape optimization by utilizing a high-accuracy surrogate model for thermo-structural physical field prediction, constructed through limited-sample learning.

Target Cascading of Tire Performances using Machine Learning

Yeonsang Yoo (Hyundai Motor)･Benjamin Schaefer (RWTH Aachen University)･Yongdae Kim･JinSil Kyeong (Hyundai Motor)

The automotive industry is shifting towards Model-Based Systems Engineering (MBSE) to manage increasing complexity in development. However, virtual tire development has a significant limitation: setting tire performance targets in early stage is difficult without physical testing. To overcome this situation, this study proposes a Target Cascading (T/C) process to derive tire performance targets directly from vehicle performance requirements and validated the process. Previously developed performance prediction models were improved for refined T/C process. Additionally, a GUI was developed to integrate the prediction models and T/C process for user convenience.