Wire EDM Parameter Optimization of AlSi10Mg Alloy

Abstract

This work attempts to optimize the multi-objective characteristics of wire electrical discharge machining (WEDM) in SLM-fabricated AlSi10Mg through a hybrid artificial neural network (ANN) coupled genetic algorithm approach. The SLM is unambiguously one of the most effective commercially viable successful additive manufacturing (AM) technologies that have the potential to replace many traditional methods of manufacturing. However, the highly intricate metallic support structures created in SLM are too strong to be eliminated by hands for which precision machining operations such as WEDM are widely employed for post-processing of SLM–AlSi10Mg. The Taguchi experimental design, considering the three most influencing factors, is performed to obtain micro-hardness and surface roughness results. The input factors for optimization are discharge current, discharge voltage, and pulse time-on in the WEDM process. The multi-objective optimization is performed using the ANN coupled GA approach where the ANN model has been generated first and the results of the best model are fed to GA for optimization. For this, five variants of three-layered, multi-perceptron models with feed-forward (BP) neural structures are also developed. The current model is supplemented with a Levenberg–Marquardt algorithm that uses logarithmic sigmoid (logsig) and linear (purelin) transfer functions. Finally, the response values from the best ANN model (3–10–2) are employed for multi-objective optimization using GA. The present study establishes the following optimized process parameters: 12 A discharge current, 42 V discharge voltage, and 12 µs time-on for maximized micro-hardness of 478 VHN and minimized surface roughness of 4.3 µm, both with greater than 98% confidence level. The present study reports briefly phase characterization such as the presence of Si particles, α-Al, and Mg2Si phases on the recast surface. The surface quality of the optimized specimen exhibits superior surface quality than its other experimental counterparts.