Assessment on electrical discharge machining of ultrasonication assisted stir-casted AA8081-B4C-Gr hybrid composites and prediction using Levenberg-Marquardt technique

In this investigation, electrical discharge machining (EDM) of hybrid aluminium composite (HAC) comprising AA8081, boron carbide (B 4 C) (10 wt. %) and graphite (Gr) (5 wt.%) is carried out. The ultrasonic cavitation-aided stir casting procedure is adopted to cast the AMMC. Experiments were designed using response surface methodology, considering discharge current (DI), pulse off-time (Toff), and pulse on-time (Ton) to optimize the outputs over cut (OC), tool wear rate (TWR), and material removal rate (MRR). Microscopic images reveal the uniform spreading of reinforcements, and the machined layer shows few globules with minimal cracks. Cavitation promotes heterogeneous nucleation, leading to a fine and equiaxed microstructure and reduced porosity. Increased Ton rises MRR and TWR owing to enhanced thermal energy transfer and prolonged spark duration. An intense thermal load and spark energy develop gradual electrode erosion. Increasing the Toff significantly lowers the MRR and TWR. Rising the DI considerably increases the MRR and TWR. Higher Ton, DI, and Toff increases the OC. Desirability analysis finds the optimal machining condition as Ton of 3 μs, Toff of 8.63 μs, and DI of 20 A, providing a higher MRR (0.25 g/min), lower OC (0.278 mm), and TWR (0.058 g/min) with a desirability value of 0.776. A Levenberg-Marquardt Neural Network model (3-7-3 architecture) predicts the outputs more precisely than the regression models, with R 2 values of 0.9992, 0.992, 0.9693, and 0.9946 for training, validation, testing, and overall, respectively, with error values lower than those of the experimental datasets. • Ultrasonic assisted stir casting procedure homogenizes the distribution of reinforcements (boron carbide and graphite) in aluminum matrix. • Electrical Discharge Machining (EDM) on hybrid aluminium composite (AA8081-B 4 C-Gr) was performed. • Optimal pulse on-time (3μs), pulse off-time (8.63 μs) and discharge current (20 A) produced lower overcut (0.278 mm) and TWR (0.058 g/min) rate with higher material removal (0.25 g/min). • A Levenberg-Marquardt neural network model predicts the outputs most precisely than the regression models.