Development of prediction method for SPR(Self Piercing Rivet) joint strength using machine learning

Kento Shimizu･Yuta Suzuki･Hirotaka Sakamoto･Toshiyuki Isono･Kohei Takahashi (Toyota Motor)･Narihisa Fujimoto (TOYOTA MOTOR KYUSHU)

SPR (Self Piercing Rivet) joint is enabled to join dissimilar materials, but it is difficult to predict fatigue strength without testing and detailed mesh simulation because, it varies from plate combination. Therefore, it is essential to develop a method to predict fatigue strength of SPR easily. This report describes the investigation of SPR joint fatigue strength prediction method using a machine learning model with random forests, based on specimen test result.

High-resolution topology optimization of instrument panel beams

Yuji Wada (Institute of Science Tokyo)･Takeshi Kashiyama･Kei Nagasaka (Suzuki Motor)･Koji Nishiguchi (Nagoya University, RIKEN)･Shigenobu Okazawa (University of Yamanashi)･Makoto Tsubokura (Kobe University, RIKEN)

The instrument panel beam at the front of the vehicle body is often made of hollow tubes, but in order to reduce the number of parts, one-piece molding has been considered in recent years. Voxel topology optimization is performed under the CUBE framework and the resulting optimal shape is compared with existing parts to study stiffness performance and manufacturability using a plate manufacturing constraint.

Machine learning surrogate models for real-time Tire contact Finite Element Analysis

Akira Wada･Ryoji Sekine (TOYO TIRE)

Potential fundamental characteristics of tires are analyzed by tire contact finite element analysis. It is necessary to find the optimal design values efficiently, however they have hundreds of design variables. In this paper particularly focuses on tire profile variables and presents cases where a surrogate model enabling real-time simulation and evaluated for predictive accuracy within a certain design space.

A Study on Enhancing the Accuracy of Crash CAE Predictions for Fiber-Reinforced Plastic Components through the Utilization of Surrogate Models

Yoshikazu Nakagawa･Osamu Ito (Honda Motor)

To rapidly predict the fiber orientation of resin parts, essential for automotive crash analysis, we developed a surrogate model by extending the machine learning algorithm pix2pix to 3D. A comparison with resin flow analysis results confirmed the model's high accuracy. This approach successfully reduces the substantial workload associated with the entire labor-intensive flow analysis process.

Shape Optimization of Aluminum Extrusion Considering Manufacturing Constraints

Yuto Komatsu･Shota Chinzei･Taiki Yamakawa･Narikazu Hashimoto (Kobe Steel)

Aluminum extrusion design offers significant flexibility due to variations in cross-sectional geometry and internal rib configurations. However, achieving a globally optimal design is challenging through manual methods. While shape optimization is increasingly applied, it often overlooks manufacturing feasibility. This study presents and validates a technology that incorporates manufacturing constraints of extrusion process into the AI-driven shape optimization of aluminum extrusions.

3D Shape Generative AI for Thin Sheet Metal Components under Geometric Boundary Shape Constraints

Takumi Sugiura (JSOL)･Isamu Hashiguchi (The University of Osaka)･Atsushi Takahashi･Nobuhiro Taki (JSOL)･Koji Nishiguchi (Nagoya University)･Kei Saito (JSOL)

This paper proposes a 3D shape generative AI method to enhance efficiency during the early structural design phase of vehicle development. For sheet metal components, the method is capable of generating multiple internal shape concepts under specified geometric boundary shape constraints. The hood is used to demonstrate the proposed method.

Tread Pattern Design Technology that Balances Functionality and Aesthetics Using Generative AI

Ryoichi Ishihara･Shuichi Karatsu (TOYO TIRE)

The tread pattern must satisfy not only physical performance criteria, such as ride comfort and low fuel consumption, but also aesthetic design requirements. In this study, we propose a framework for optimizing tread pattern designs that simultaneously balance physical performance and aesthetics. We utilized CAE to evaluate physical properties and applied AI models for generating tread patterns and assessing aesthetic qualities.