Exploring Deep Learning Methods to Forecast Mechanical Behavior of FSW Aluminum Sheets

This work aimed to develop effective data-driven approaches to forecast the stress-strain curves of Al6061-T6 aluminum alloy base material and welded using Friction Stir Welding (FSW) technique under different temperature conditions. Accurate forecasting of the material's behavior is undoubtedly essential to predict the mechanical piece's life span under different working conditions and get relevant information, such as strain softening and material characteristics. Importantly, two deep learning models were investigated, namely long short-term memory (LSTM) and gated recurrent unit (GRU). This choice is motivated by the capacity of LSTM and GRU to learn temporal dependencies from time-series data. In addition, these deep learning-driven methods promise forecasting results but require no assumptions on the data distributions. According to the existing literature, this is the first study introducing the LSTM and GRU models to forecast the stress-strain curves effectively. Experiments have been conducted using Al6061-T6 aluminum alloy and FSW joining process under different temperature levels: 25, 100, 200, and 300 degrees C. Forecasting results demonstrated LSTM and GRU models' promising capacity to capture the future trends of stress-strain curves under different temperature conditions. In terms of efficiency and accuracy, the GRU-driven forecasting approach converges faster and exhibits better performance than the LSTM approach.