Evaluation of surface roughness in the turning of mild steel under different cutting conditions using backpropagation neural network

Abstract

This paper exhibits a model of feedforward backpropagation neural network system for estimating surface roughness in the turning operation. The workpiece of mild steel (carbon content 0.2%; hardness125 BHN) has been taken for turning operation under different cutting conditions with highspeed steel (HSS) tool (carbon content 0.75%; vanadium content 1.1%, molybdenum content 0.65%, chromium content 4.3%, tungsten content 18%, cobalt content 5%, hardness 290 BHN). Experiments have been executed on lathe machine HMT LB20. In the neural network model, the speed, feed and depth of cut have been considered as process parameters and surface roughness was taken as a response parameter. The neural network was developed based on initial experimental data. The developed neural network model during testing and validation was found to be within acceptable limits. The estimated maximum error was expected to be 10.77%. Error below 20% was considered reasonable, taking into account the fact that there is an intrinsic irregularity in metal cutting procedure.