Simulation of surface generated during abrasive flow finishing of Al/SiCp-MMC using neural networks

Abstract

Abrasive flow machining (AFM) is a multivariable finishing process which finds its use in difficult to finish surfaces on difficult to finish materials. Near accurate prediction of generated surface by this process could be very useful for the practicing engineers. Conventionally, regression models are used for such prediction. This paper presents the use of artificial neural networks (ANN) for modeling and simulation of response characteristics during AFM process in finishing of Al/SiCp metal matrix composites (MMCs) components. A generalized back-propagation neural network with five inputs, four outputs, and one hidden layer is designed. Based upon the experimental data of the effects of AFM process parameters, e.g., abrasive mesh size, number of finishing cycles, extrusion pressure, percentage of abrasive concentration, and media viscosity grade, on performance characteristics, e.g., arithmetic mean value of surface roughness (Ra, micrometers), maximum peak–valley surface roughness height (Rt, micrometers), improvement in Ra (i.e., ΔRa), and improvement in Rt (i.e., ΔRt), the networks are trained for finishing of Al/SiCp-MMC cylindrical components. ANN models are compared with multivariable regression analysis models, and their prediction accuracy is experimentally validated.