DAMAGE QUANTIFICATION ON COMPOSITE STRUCTURES USING NEURAL NETWORKS AND HYBRID DATA

• Autor(a) principal: de Paula Silva Ferreira, Leonardo
• Data de Publicação: 2022
• Outros Autores: Yano, Marcos Omori [UNESP], da Silva, Samuel [UNESP], Cimini, Carlos Alberto
• Tipo de documento: Artigo de conferência
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://hdl.handle.net/11449/247396
• topic: hybrid learning; Lamb wave simulations; neural networks; SHM

In the aeronautic industry, in addition to the aging of the current fleet of aircraft in operation, increasing cargo capacity and the use of composite materials have increased interest in developing Structural Health Monitoring systems (SHM) by aircraft manufacturers and airlines. On the other hand, the increase in the processing capacity of computers enabled the development of Artificial Intelligence systems. These systems can make decisions based on an incomplete data set and are particularly attractive in applications where human intelligence and critical thinking are needed. However, the performance of SHM based on Machine Learning is limited to only the knowledge used in the learning phase, not able to describe the structural behavior under conditions different from those used in the model training. This work proposes a hybrid learning methodology as an alternative to augment the amount of data available during the training phase. A finite element model is adjusted with limited experimental data and used to simulate new damage scenarios. Then, a multilayer neural network is trained with different experimental and numerical data combinations. The system’s performance is evaluated with experimental data that is not used during model training, and the model’s accuracy is compared using scenarios with and without.