Nonlinear Vibration Response Modeling and Performance Prediction of Air Suspension Using Machine Learning

Yoshihiro Atsumi･Yasukazu Sato (Yokohama National University)

A vibration response model incorporating the nonlinear characteristics of an air suspension is constructed using machine learning, and a method for accurately predicting ride comfort performance is proposed. The nonlinear characteristics are separated into static and dynamic components based on design specifications and experimental data. These components are then modeled. The prediction accuracy of the proposed method is evaluated through simulations in preparation for bench testing, and its effectiveness is demonstrated.

Estimation Technique of Differential Pressure–Flow Characteristics of Shock Absorber Valves Based on Machine Learning

Haruhito Kato･Peng Lyu･Kenji Hashirayama･Naofumi Harada (Astemo)

In the early design phase of shock absorbers, it is necessary to obtain the differential pressure–flow characteristics of the valves. The characteristics are generally estimated using computer simulations. It is difficult to increase both estimation accuracy and computational efficiency since the characteristics depend on valve geometry. In this paper, we develop the machine-learning-based estimation technique that is able to address the practical problem.

A Target Characteristic Setting Method for Tires and Suspensions Using Surrogate Models and FRF-Based Substructuring

Keito Akima･Masayuki Taketani･Hideki Kawai･Hiroyuki Seino･Hirotaka Shiozaki (Mitsubishi Motors)

Road noise development is facing increasing constraints due to shortened development schedules and fuel efficiency demands, making it crucial to set appropriate component targets in the early stages when design flexibility is high. This study proposes a method using surrogate models and FRF-based substructuring to derive specific targets such as component stiffness and resonance frequency from interior noise targets.

Occupant Head Motion Estimation Using an LSTM-Based Machine Learning Model

Kenji Nishida･Chikara Kawamura･Daichi Sato･Hiroto Nishimura･Hiroaki Miyatake･Kenji Takeda･Yu Chen･Toshiaki Aoki･Kosuke Maruyama･Toru Ishikawa (Mazda)

In vehicle development, it has been challenging to quantitatively predict, at the early design stage, how changes in vehicle requirements affect occupant head motion. In this study, an LSTM-based time-series model was developed using experimental data to predict occupant head motion from vehicle behavior. A key feature of the model is its ability to account for input history. The prediction accuracy of occupant head motion under variations in vehicle behavior was evaluated, and the applicability of the model to vehicle requirement analysis was assessed.

Multi performance parametric optimization for motor using machine learning models and Robust validation

Koichi Yakawa･Ryoichi Adachi･Ryotaro Nakayama (Mazda)

In this presentation, we will introduce a method for exploring specification and the evaluation of variation factors through parametric optimization using machine learning models. In the development of electric vehicle motors,manual adjustments take time to derive specifications that resolve conflicting multi-performance issues and comprehensive verification of variation factors. We created a machine learning model based on JMAG calculation results, incorporated it into an optimization tool, and performed optimization. We also used the created machine learning model to evaluate the impact of variation factors.

Optimization of an FCEV Air Exhaust System for Flow-Induced Noise Reduction and Cost Efficiency

Jong Yeol Lee･Seung Tae Gong･Jong Seung Won (Hyundai Motor)･Hyun Soo Kim･Joon Ho Jang (SJG SEJONG)

This study investigates an optimized air exhaust silencer for a fuel cell electric vehicle to reduce flow-induced noise while improving cost efficiency. Vehicle tests identified discharge noise in the 1-3 kHz range as the primary NVH issue, originating from APC valve-induced turbulence and structural limitations of the production silencer. To mitigate this, the noise attenuation zone was relocated forward and a Helmholtz resonator-based concept was applied. Transmission loss analysis showed that the optimal design achieved up to 16.6 dB improvement in TL RMS, maintained approximately 90% gas-liquid separation, and reduced dependence on absorptive materials.