INFORMATION SYSTEM FOR DETERMINING THE STRESS-STRAIN STATE OF ELASTOMERS STRUCTURES BASED ON A NEURAL-NETWORK
- 29 December 2025
- journal article
- Published by National Technical University Kharkiv Polytechnic Institute in Bulletin of the National Technical University «KhPI» Series: Dynamics and Strength of Machines
- No. 2,p. 47-55
Abstract
This work presents an information system designed to determine the mechanical characteristics of elastomeric materials through the integration of artificial neural networks with classical numerical analysis methods. A hybrid framework is proposed in which a neural network module approximate the experimental loading-unloading curve, including the Mullins effect as a key manifestation of material hysteresis. The predicted stress-strain response is incorporated into an iterative Newton-Raphson scheme to compute the tangent modulus of elasticity in the absence of analytical derivatives. Numerical stability is ensured by implementing an adaptive step-control mechanism, constraints on deformation corrections, and a local recovery procedure that prevents the propagation of divergence throughout the global solution. The system architecture is implemented as a cross-platform desktop application based on PySide6, combining interactive editing of the finite-element model structure, dynamic data validation, and maintenance of topological consistency through the delegation pattern. The model supports serialization and deserialization in CSV format, including field validation, automatic type conversion, and restoration of empty entries, thereby enabling reproducible calculations and reliable model exchange between users. The outputs of the system include both a graphical representation of the deformed structural configuration and a detailed analytical report containing nodal displacements and reactions, internal forces, strain measures, and resulting effective elastic moduli. The proposed system provides high computational stability and demonstrates practical applicability for modelling elastomers with pronounced nonlinear behaviour. The results highlight the potential of combining machine learning with classical mechanical techniques to advance modern tools for the engineering analysis.Keywords
This publication has 13 references indexed in Scilit:
- A Physics‐Informed Neural Network Model for the Anisotropic Hyperelasticity of the Human Passive MyocardiumInternational Journal for Numerical Methods in Engineering, 2025
- A framework for neural network based constitutive modelling of inelastic materialsComputer Methods in Applied Mechanics and Engineering, 2023
- Introduction of a recurrent neural network constitutive description within an implicit gradient enhanced damage frameworkComputers & Structures, 2023
- Predicting the mechanical properties of biopolymer gels using neural networks trained on discrete fiber network dataComputer Methods in Applied Mechanics and Engineering, 2021
- A physics-informed deep learning framework for inversion and surrogate modeling in solid mechanicsComputer Methods in Applied Mechanics and Engineering, 2021
- Learning constitutive relations using symmetric positive definite neural networksJournal of Computational Physics, 2021
- Machine learning-based constitutive model for J2- plasticityInternational Journal of Plasticity, 2021
- SciANN: A Keras/TensorFlow wrapper for scientific computations and physics-informed deep learning using artificial neural networksComputer Methods in Applied Mechanics and Engineering, 2021
- Neural network constitutive model for crystal structuresComputational Mechanics, 2020
- A neural network constitutive model for hyperelasticity based on molecular dynamics simulationsInternational Journal for Numerical Methods in Engineering, 2020