A machine learning prediction–optimisation framework for dry turning of grey cast iron to enhance machinability and prolong tool life: A comparative study of hybrid DNN, SVM and decision tree

Abstract

This study addresses the dry turning of EN-GJL-250 grey cast iron and examines how insert coating, cutting parameters and machining duration affect tool wear, cutting force and surface roughness. The aim was to build an accurate prediction and optimisation framework. Three predictive models were compared: a decision tree (DT), a support vector machine (SVM) and a hybrid deep neural network (H-DNN–IGWO-CV) whose architecture was configured by an Improved Grey Wolf Optimizer (IGWO) within a cross-validation loop. The most accurate model was then coupled with the Superb Fairy-wren Optimisation Algorithm (SFOA) for multi-objective Pareto optimisation. The measurements show that increasing flank wear raises both cutting force and surface roughness ( r = 0.49 and r = 0.48 ), and that the TiN-coated insert reduces wear ( r = − 0.60 ) and roughness ( r = − 0.33 ) relative to the uncoated ceramic. Among the models, the H-DNN–IGWO-CV gave the lowest errors and a cross-validated R 2 of about 0.99, indicating that searching the network architecture improves predictive stability compared with fixed network designs.