Publicação
Influence of data filtering and noise on the calibration of constitutive models using machine learning techniques
| Resumo: | This work focuses on predicting material parameters that describe the plastic behaviour of metallic sheets using the XGBoost machine learning algorithm, with a dual focus on the influence of data filtering and data noise. A dataset was populated with finite element simulation results of cruciform tensile tests, including strain field data during the test. Different noise levels were added to the strain-related features of the dataset; additionally, a feature importance study was carried out to identify and select the most relevant features of the dataset. A systematic analysis shows how feature noise and selection individually and simultaneously influence the predictive performance of machine learning models. The results show that feature selection will greatly accelerate model training, without losing its predictive performance. Also, adding noise to the features does not have significant impact on model performance, highlighting the robustness of the models. |
|---|---|
| Autores principais: | Prates, Pedro |
| Outros Autores: | Pinto, José; Marques, João; Henriques, João; Pereira, André; Andrade-Campos, António |
| Assunto: | Parameter identification Machine learning Feature analysis Noise Sheet metal forming |
| Ano: | 2024 |
| País: | Portugal |
| Tipo de documento: | artigo |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Aveiro |
| Idioma: | inglês |
| Origem: | RIA - Repositório Institucional da Universidade de Aveiro |
| Resumo: | This work focuses on predicting material parameters that describe the plastic behaviour of metallic sheets using the XGBoost machine learning algorithm, with a dual focus on the influence of data filtering and data noise. A dataset was populated with finite element simulation results of cruciform tensile tests, including strain field data during the test. Different noise levels were added to the strain-related features of the dataset; additionally, a feature importance study was carried out to identify and select the most relevant features of the dataset. A systematic analysis shows how feature noise and selection individually and simultaneously influence the predictive performance of machine learning models. The results show that feature selection will greatly accelerate model training, without losing its predictive performance. Also, adding noise to the features does not have significant impact on model performance, highlighting the robustness of the models. |
|---|
Atividades financiadas
Não foram encontrados projetos financiados.