AMS has formulated two inverse algorithms to determine material curves from experimental data. The result are material function that accurately reproduce the experimental tests. A complete description of these algorithm can be found in AMS-publications. Our algorithms are able to determine material functions, such as plastic stress strain curve or the the softening curve, by means of an iterative calculation that successively transforms the material function reducing the differences between the numerical results and experimental data. The degree of approximation obtained with the AMS algorithm is greater than other equivalent techniques. A briefly description of the two algorithms:


  • AMS inverse algorithm to determine the plastic stress strain curve of a material. The result is a material curve that accurately reproducing the experimental test. Our technicians, in collaboration with universities in the Community of Madrid, have developed and validated this inverse algorithm that has been successfully applied to different geometries as: ring tension test, ring compression test, notched specimens under tension and nanoidentation.

  • AMS inverse algorithm to determine the concrete softening curve. Fracture in concrete can be explained using the cohesive zone model which formulation is based on a material curve called softening curve. The algorithm determines this curve that predicts the real behaviour in fracture.


Artificial intelligence and data-driven models are a powerful and promising of our working lines. AMS works on the application of the machine learning and Deep Learning techniques to Material Science constructing hybrid models. Some of our developments can be found in the following list:

  • Application of invertible neural networks in materials science The application of invertible neural networks in materials science is a powerful methodology that allows for the determination of material properties and the evaluation of measurement errors. This technique combines experimental results, numerical calculations, statistical techniques, and a specific type of neural networks known as invertible networks. At ADVANCED MATERIAL SIMULATION, we have applied this innovative methodology to characterize the poroacoustic parameters of cellular concrete using an absorption acoustic curve. Once the experimental curve is obtained, a range of possible parameters predicting the experimental results is calculated.
  • INN

  • Real-time modelling. Combining numerical "physical" models with reduced models based on neural networks enables the execution of complex and extensive simulations in real-time. This innovative approach allows for the accurate capture of fundamental aspects of physical systems while leveraging the efficiency and speed of neural network models to streamline simulation processes. This hybrid approach not only ensures the accuracy of results but also significantly enhances responsiveness and simulation speed, providing our clients with a powerful and effective tool to address a wide range of real-time challenges.

  • Real time models

  • Determination of mechanical properties in gradient regions from images. The automatic analysis of microstructures images is a promising element to assess the physical properties of non-homogeneous regions where a material gradient exists. A solution to evaluate precisely the property variation is to generate a Convolution Neural Network to establish a data-based correlation between the image and the properties. The network is trained with homogeneous images where the numerical values of the properties are known and applied to a gradient region to obtain the real data.



  • Virtual fracture surfaces from transfer learning. The application of transfer learning and Generative Adversarial Network to generate virtual fracture surfaces is a promising and novel methodology to mix complex mechanical effects. The virtual result of addition or subtraction of thermo-mechanical treatments can be visualized using these techniques.


  • Automatic image processing. Automatic image processing is an essential complement to Deep Learning. The combination of image data with standard morphological information led to hybrid physics informed models. Automatic procedures to determine surface fraction, morphological descriptor, or critical elements as for example the critical radial defect have been developed in AMS.


┬ęCopyright AMS SIMULATION 2024. All rights reserved.