Artificial neural network prediction and optimization of hardness and ductility for Al-Zn-Mg based hybrid composite

This study evaluates the hardness and ductility of an Al–5.6Zn–2.5Mg (Al7075) based hybrid metal matrix composite reinforced with silicon dioxide (SiO₂) and red mud particles having average particle sizes of 200 μm and 340 μm, respectively. The composites were fabricated using a liquid metallurgical technique. Experiments were conducted based on a standard L16 orthogonal array, and mechanical properties were evaluated using a hardness tester and a universal testing machine in accordance with ASTM E18 and ASTM E8 standards. A data-driven approach was developed using a feedforward artificial neural network (ANN) with two input parameters (SiO₂ and red mud content), a single hidden layer comprising 10 neurons, and two output responses (hardness and ductility), trained using the Levenberg–Marquardt algorithm. In addition, response surface analysis, residual error evaluation, and Particle Swarm Optimization (PSO) were employed to further analyze and optimize the experimental results. The ANN model demonstrated good predictive capability, with average percentage errors of 4.53% for hardness and 1.41% for ductility. The correlation coefficients (R) for training, testing, and validation datasets were 0.9999, 0.9918, and 0.8958, respectively. Optimization results indicated that the maximum hardness was achieved at 3.9 wt% SiO₂ and 10 wt% red mud, while optimum ductility was obtained at 1.5 wt% SiO₂ and 2.5 wt% red mud.