Parametric Analysis of the Process Performance of Surface Roughness Data of Machined Aluminium using PSO-ANN

The external appearance of a workpiece is an indicator of its quality. It relies heavily on visual inspection used in manufacturing to verify if a product is free from defects. Surface roughness (SR) on a machined part can be a defect that affects the aspect of a product. Machine setting is controllable parameters that are closely related to the surface texture. However, many studies have pointed out that these influencing parameters are nonlinear. In the present study, a computational approach based on Artificial Neural Network (ANN), trained by Particles Swarm Optimisation (PSO) has been used to predict the SR. The dataset considered in this study has been collected from the previous study that developed mathematical models employing least square approximation method (LSA) as well as response surface methodology (RSM) to determine the surface roughness of an experiment conducted on 30 samples of aluminum. The tool's speed of rotation, the feed rate, the radial depth cut, and the axial depth cut have been considered as inputs. The SR was considered as the output parameter. So as to achieve the PSO-ANN finest code, a parametric analysis was carried out taking into consideration the number of neurons (n) in the hidden layer., size of swarm population (N) and acceleration factors (c1 and c2). This analysis has reported the highest trained regression value (R2) of 0.9938 and a lower mean square error of 0.00044 corresponding to nine (9) neurons, a swarm population size of 200, and acceleration factors of 1.5 and 2.25. This study reveals the potential of the hybrid ANN-PSO for the prediction of surface roughness which could enhance the machining of aluminium potentially.