The results of The Project CoMModO are available in form of engineering services. These are illustrated by the following exemplary usecases, which can also be seen as escalating steps in a comprehensive process scheme for the modeling of complex material behaviours.
Material Parameter Estimation and Prediction
Material tests come most of the time in the form of force-deflection curves like the one shown here. The aim of material modelling is the mathematical description of such curves dependent on all necessary parameters, like load conditions, geometry, material specs, etc.
As a first step one can try not to describe the complete curve but some of its characteristic features, such as the maximum force, the displacement attained at the time of the maximum force, some curve integrals associated with energies, etc. The application of machine learning models or other data mining procedures for this task helps to identify which are the main parameters that influence the material behaviour.
Often these characteristic features have some physical meaning and are themselves parameters needed by conventional analytical material models. The resulting models for the prediction of these parameters can then be used to calibrate and tune the analytical material models.
Material Model Validation, Exploration and Tuning
An analytical material model with values for its parameters is given. Is this material model with its settings (e.g. coming from parameter prediction models) able to fit some given material test data?
Such a material model validation can be done by means of stochastic simulation. Hereby one can find out if a given material model is able to accomplish the desired functional requirements and how the resulting parameters depend on the material model parameters and on each other. This exploration of the flexibility of the model allows to avoid unrealistic goals for a further parameter optimization or, in case of insufficient flexibility, supports the decision to proceed to advanced material modeling techniques.
The figure beside shows an example of an analytical material model that is not able to fit the given material tests. The consequence would be to relax the requirements by reducing the number of load cases or to proceed to the next step of advanced material modeling techniques. Material parameter optimization would not lead to satisfactory solutions in this case, because the given material model is not flexible enough.
If the material model would cover the test results then evolutionary strategies could be one method to find the best fitting material model parameters.
Advanced Material Modeling
If no known analytical material model is able to fulfill the desired behaviour, advanced material modeling techniques developed within CoMModO come into play. Hereby the material behaviour is described directly by much more flexible machine learning models, e.g. artificial neural networks. Besides the pure mathematical formulation also an adequate formulation of the training goals, special adaptions of the learning algorithms, and a requirements-driven development process are necessary to make the application of artificial neural networks feasible for material modeling.
CoMModO provides a comprehensive development environment for the design, tuning, implementation, validation and adaption of such material models.
Material Model Requirements Management and Test Design
A crucial issue for the development of models for complex materials is the consistent formulation of the functional requirements and the according specification of representative material tests. Both tasks are usually far from being trivial in the beginning of the material model development.
CoMModO provides an approach for evaluating, capturing and structuring the load patterns that occur in complex structures with the aid of new postprocessing algorithms that allow for an iterative acquisition of the functional requirements.
Hereby the applied loads and input parameters of the material models are continuously monitored and accompanying processes check if the material tests still represent the arising loads and functional requirements well enough. The coverage of new load conditions and input parameter values is the basis for a continuous adaption and improvement of the material models. WIth this approach, these models learn the desired material behaviour coping with the growing requirements and new load conditions.