Experimental investigation and machine learning modeling using LSTM and special relativity search of friction stir processed AA2024/Al2O3 nanocomposites

Abstract

In this study, the friction stir technique is proposed to process aluminum nanocomposites reinforced with alumina nanoparticles. The effects of different processing parameters, including spindle speed (900–1800 rpm), feed (10–20 mm/min), and number of passes (1–3) on the mechanical and dynamic properties of the processed samples were investigated. The investigated properties were ultimate tensile strength, yield strength, natural frequency, and damping ratio. An advanced machine learning approach composed of a long short-term memory model optimized by a special relativity search algorithm was developed to predict the properties of the processed samples and different processing conditions. The adequacy of the developed model was tested and compared with three other machine learning models; the predicted properties were in good agreement with the measured properties. The developed model outperformed other tested models and was found to be a powerful prediction tool for predicting processing conditions to obtain high-quality nanocomposite samples. The model succeeded in predicting the ultimate tensile strength, yield strength, natural frequency, and damping ratio with good R2 of 0.912, 0.952, 0.951, and 0.987, respectively. The obtained results showed that the samples' damping ratio and loss factor increase with the number of passes, while the natural frequency, shear modulus, and complex modulus decrease with the number of passes. Thus, friction stir processing can be used to improve the damping properties of materials.