To be able to use finite element (FE) simulations in structural component development, experimental investigations for the characterization of the material properties are required to subsequently calibrate suitable material cards. In contrast to the commonly used computational and time-consuming method of parameter identification (PI) by using analytical and numerical optimizations with internal or commercial software, a more time-efficient method based on machine learning (ML) is presented. This method is applied to simulate the material behavior of additively manufactured specimens made of acrylonitrile butadiene styrene (ABS) under uniaxial stress in a structural simulation. By using feedforward artificial neural networks (FFANN) for the ML-based direct inverse PI process, various investigations were carried out on the influence of sampling strategies, data quantity and data preparation on the prediction accuracy of the NN. Furthermore, the results of hyperparameter (HP) search methods are presented and discussed and their influence on the prediction quality of the FFANN are critically evaluated. The investigations show that the NN-based method is applicable to the present use case and results in material parameters that lead to a lower error between experimental and calculated force-displacement curves than the commonly used optimization-based method.

A neural network (NN)-based method is presented in this paper which allows the identification of parameters for material cards used in Finite Element simulations. Contrary to the conventionally used computationally intensive material parameter identification (MPI) by numerical optimization with internal or commercial software, a machine learning (ML)-based method is time saving when used repeatedly. Within this article, a self-developed ML-based Python framework is presented, which offers advantages, especially in the development of structural components in early development phases. In this procedure, different machine learning methods are used and adapted to the specific MPI problem considered herein. Using the developed NN-based and the common optimization-based method with LS-OPT, the material parameters of the LS-DYNA material card MAT_187_SAMP-1 and the failure model GISSMO were exemplarily calibrated for a virtually generated test dataset. Parameters for the description of elasticity, plasticity, tension–compression asymmetry, variable plastic Poisson’s ratio (VPPR), strain rate dependency and failure were taken into account. The focus of this paper is on performing a comparative study of the two different MPI methods with varying settings (algorithms, hyperparameters, etc.). Furthermore, the applicability of the NN-based procedure for the specific usage of both material cards was investigated. The studies reveal the general applicability for the calibration of a complex material card by the example of the used MAT_187_SAMP-1.

Increasing product requirements in the mechanical engineering industry and efforts to reduce time-to-market demand highly accurate and resource-efficient finite element simulations. The required parameter calibration of the material models is becoming increasingly challenging with regard to the growing variety of available materials. Besides the classical iterative optimization-based parameter identification method, novel machine learning-based methods represent promising alternatives, especially in terms of efficiency. However, the machine learning algorithms, architectures, and settings significantly affect the resulting accuracy. This work presents a comparative study of different machine learning algorithms based on virtual datasets with varying settings for the direct inverse material parameter identification method. Multilayer perceptrons, convolutional neural networks, and Bayesian neural networks are compared; and their resulting prediction accuracies are investigated. Furthermore, advantages in material parameter identification by uncertainty quantification using the Bayesian probabilistic approach are examined and discussed. The results show increased prediction quality when using convolutional neural networks instead of multilayer perceptrons. The assessment of the aleatoric and epistemic uncertainties when using Bayesian neural networks also demonstrated advantages in evaluating the reliability of the predicted material parameters and their influences on the subsequent finite element simulations.

Finite element (FE) simulations can be used both in the early product development phase to evaluate the performance of developed components as well as in later stages to verify the reliability of functions and components that would otherwise require a large number of physical prototype tests. This requires calibrated material cards that are capable of realistically representing the specific material behavior. The necessary material parameter identification process is usually time-consuming and resource-intensive, which is why the direct inverse method based on machine learning has recently become increasingly popular. Within the neural network (NN) the generated domain knowledge can be stored and retrieved within milliseconds, which is why this method is time and resource-efficient. This research paper describes advantages and potentials of the direct inverse method in the context of the product development process (PDP). Additionally, arising transformation opportunities of the PDP are discussed and an application scenario of the method is presented followed by possible linkage potentials with existing development methods such as shape optimization.