Aprendizagem com Pequenos Dados: Rumo a Previsões Robustas e Interpretáveis com Modelação Eficiente Conduzida por Dados do Material em Simulações de Conformação

The main goal of this project is to develop accurate material modelling methodologies supported by small data-based learning approaches. These methodologies encompass the development of efficient, robust and interpretable computational tools for identifying constitutive parameters and non-parametric constitutive modelling. Particularly, this proposal includes the following scientific original items: i. From Big Data towards Small Data – Development of improved strategies for data manipulation, extraction of features and exploitation of deep learning algorithms towards a small data paradigm; ii. Small data learning-based constitutive parameter identification – Development of parameter identification strategies for classical constitutive models supported by small data learning techniques, as an alternative to classical and inverse parameter identification strategies; iii. Small data learning-based constitutive modelling – Development of non-parametric constitutive modelling approaches supported by small data learning techniques, as an alternative to conventional constitutive models and big data-supported constitutive modelling; iv. Material modelling computational tools – data-driven constitutive model implementation in the FEA commercial software ABAQUS by means of a user routine; development of a Python software application to identify constitutive parameters of classical models, based on small data learning techniques. The benefits and impact of the project are: i. efficient and accurate material modelling for real case scenarios with limited full-field experimental data; ii. increasing the precision and reliability of FEA simulations by providing accurate input data, thus filling a gap of the FEA software market and answering the request of such software users; iii. reduction of the development lead-time of metallic parts and the provision of robust technological solutions without excessive use of materials and energy, significantly contributing to green manufacturing.