The proper prediction of failure onset concerning load levels and spatial distributions is fundamental in the performance analysis of composites. A common approach is based on evaluating failure indexes obtained from combinations of stress components at any given point, and the resolution of the stress field affects the failure onset prediction. In composite structures, the complete 3D stress field – in-plane and transverse components – is necessary for many practical applications due to the anisotropic nature of the material. A typical example is the stress field around free edges where transverse shear and axial stress are dominating. However, proper numerical models to detect such fields tend to be cumbersome due to the necessity of using highly refined 3D meshes. An alternative approach is based on refined structural models to build 1D or 2D finite elements and relax aspect ratio constraints. The use of Neural Networks (NN) in structural problems is increasing due to their flexibility and ability in dealing with nonlinearities. This paper proposes a novel use of NN in combination with Node Dependent Kinematics structural models in which each node of the finite element discretization can assume a different structural theory. NDK is useful to augment the resolution of the model locally. NN is trained using different distributions of nodes to predict failure indexes. The output is the optimal refinement of the model concerning the structural model's order and where to use it. The use of the trained NN is helpful to localize the most critical areas to be modelled and set the proper structural model to use.

Local Refinement of Structural Kinematics for Failure Onset Analysis via Neural Networks / Petrolo, M.; Pagani, A.; Iannotti, P.; Carrera, E.. - ELETTRONICO. - (2022). (Intervento presentato al convegno 15th World Congress on Computational Mechanics (WCCM-XV) and 8th Asian Pacific Congress on Computational Mechanics (APCOM-VIII) tenutosi a Yokohama nel 31/07/2022 - 05/08/2022).

Local Refinement of Structural Kinematics for Failure Onset Analysis via Neural Networks

M. Petrolo;A. Pagani;P. Iannotti;E. Carrera
2022

Abstract

The proper prediction of failure onset concerning load levels and spatial distributions is fundamental in the performance analysis of composites. A common approach is based on evaluating failure indexes obtained from combinations of stress components at any given point, and the resolution of the stress field affects the failure onset prediction. In composite structures, the complete 3D stress field – in-plane and transverse components – is necessary for many practical applications due to the anisotropic nature of the material. A typical example is the stress field around free edges where transverse shear and axial stress are dominating. However, proper numerical models to detect such fields tend to be cumbersome due to the necessity of using highly refined 3D meshes. An alternative approach is based on refined structural models to build 1D or 2D finite elements and relax aspect ratio constraints. The use of Neural Networks (NN) in structural problems is increasing due to their flexibility and ability in dealing with nonlinearities. This paper proposes a novel use of NN in combination with Node Dependent Kinematics structural models in which each node of the finite element discretization can assume a different structural theory. NDK is useful to augment the resolution of the model locally. NN is trained using different distributions of nodes to predict failure indexes. The output is the optimal refinement of the model concerning the structural model's order and where to use it. The use of the trained NN is helpful to localize the most critical areas to be modelled and set the proper structural model to use.
2022
978-84-123222-8-6
File in questo prodotto:
File Dimensione Formato  
PPIC_WCCM2022.pdf

non disponibili

Tipologia: 2. Post-print / Author's Accepted Manuscript
Licenza: Non Pubblico - Accesso privato/ristretto
Dimensione 72.33 kB
Formato Adobe PDF
72.33 kB Adobe PDF   Visualizza/Apri   Richiedi una copia
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11583/2970559