Scholarlink

<p>Elastomers' hyperelastic properties make them popular for mitigating vibration and shock in a variety of engineering applications. These properties are commonly modeled using hyperelastic models, which require the determination of material constants. In the conventional approach, multiple material test data, such as uniaxial, biaxial, planar, and volumetric data, of each material are required to obtain material constants of a hyperelastic model. Elastomer experimental testing is an expensive and time-consuming procedure. These limitations prompt the development of neural networks that can accurately determine the hyperelastic constants without requiring multiple tests. Therefore, this study focuses on developing a deep neural network (DNN) to determine the material constants of the Ogden third-order model for elastomers. Finite element analyses of uniaxial tension and compression tests are performed using ABAQUS software for randomly generated samples to generate a dataset for training the DNN model. The designed DNN model consists of three hidden layers with tanh activation functions and is trained using the AdamW optimizer. The trained DNN model is validated for hydrogenated nitrile butadiene rubber, polychloroprene rubber, and natural rubber (NR) under uniaxial conditions, and for neoprene and silicone rubbers under uniaxial, planar, and O-ring multi-contact tests. Furthermore, simulation responses using DNN-predicted constants for neoprene rubber at 50&#xa0;°C and 80&#xa0;°C closely match the experimental data. Additionally, simulation responses with DNN-predicted constants and experimental data for a shock test case study on NR dampers exhibit close agreement. Thus, the proposed DNN model can determine material constants of the hyperelastic model for any kind of rubber using direct component uniaxial data, eliminating the need for coupon tests.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Determination of hyperelastic material constants for elastomers using a deep neural network approach

  • 期刊论文

  • 发表日期：

    2025年10月28日

查看全文
  • Romi Dhakad,
  • Sunil Kumar,
  • Anil Kumar

摘要

Elastomers' hyperelastic properties make them popular for mitigating vibration and shock in a variety of engineering applications. These properties are commonly modeled using hyperelastic models, which require the determination of material constants. In the conventional approach, multiple material test data, such as uniaxial, biaxial, planar, and volumetric data, of each material are required to obtain material constants of a hyperelastic model. Elastomer experimental testing is an expensive and time-consuming procedure. These limitations prompt the development of neural networks that can accurately determine the hyperelastic constants without requiring multiple tests. Therefore, this study focuses on developing a deep neural network (DNN) to determine the material constants of the Ogden third-order model for elastomers. Finite element analyses of uniaxial tension and compression tests are performed using ABAQUS software for randomly generated samples to generate a dataset for training the DNN model. The designed DNN model consists of three hidden layers with tanh activation functions and is trained using the AdamW optimizer. The trained DNN model is validated for hydrogenated nitrile butadiene rubber, polychloroprene rubber, and natural rubber (NR) under uniaxial conditions, and for neoprene and silicone rubbers under uniaxial, planar, and O-ring multi-contact tests. Furthermore, simulation responses using DNN-predicted constants for neoprene rubber at 50 °C and 80 °C closely match the experimental data. Additionally, simulation responses with DNN-predicted constants and experimental data for a shock test case study on NR dampers exhibit close agreement. Thus, the proposed DNN model can determine material constants of the hyperelastic model for any kind of rubber using direct component uniaxial data, eliminating the need for coupon tests.