Magnesium alloy is an emerging smart metal used in various industries like automotive and aerospace industry, due to their lightweight and excellent strength-to-weight ratio. Formability, a critical factor in manufacturing processes, determines the alloy’s ability to undergo deformation without fracture or defects. Fuel economy and environmental conservatives are the key desirable factors in selection of magnesium alloy sheets. Magnesium alloy sheets have low formability at room temperature due to their hexagonal closed-packed microstructures. As the magnesium’s formability at room temperature is considerably low, stretch forming tests are conducted at moderate temperatures. For this purpose, commercially available AZ31B magnesium alloy sheet of 1.1mm thickness has been used and tested at room temperature, 25 degree to within medium temperatures range and at a higher strain rate of 0.01/s. The main objective of an experimental study to predict the formability of magnesium alloy sheets is to gather data through controlled tests and measurements. This data and Forming Limit Diagram (FLD) can be used to analyse the formability of material, it defines failure criteria. On the other hand, using a neural network to predict formability involves training the network on the collected experimental data. Once trained, the neural network can predict the formability of new magnesium alloy sheets based on their characteristics, offering a faster and potentially more accurate prediction method compared to traditional models. This work explores into the realm of regression modelling utilizing neural networks, a powerful subset of machine learning techniques. It begins with a discussion on the setup of machine learning models, emphasizing the crucial steps involved in data preprocessing, model selection, and evaluation.
Read full abstract