Data-Driven Material Models for Engineering Materials Subjected to Arbitrary Loading Paths: Influence of the Dimension of the Dataset

Engineering materials are subjected to complex stress states, mutable environmental conditions, and strain rates during their operating life. It is therefore paramount to develop methodologies capable of capturing their behaviour from experimental data, in order to predict their response under different thermo-mechanical sequences and histories. This is particularly relevant for materials that exhibit different strength in tension, compression, shear, and their combination, such as titanium alloys, magnesium alloys, composites, etc. The adoption of machine learning data-driven models obtained from arbitrary thermo-mechanical loading experiments provides an accurate and computationally efficient way to predict the response of engineering materials during loading sequences typical of real case scenarios. This study presents how neural networks with different structures can capture the response of materials measured during experiments carried out under arbitrary sequences of load. The effect of the data set size on the accuracy of the surrogate model is also assessed.