Comparative Analysis of RSM, ANN and ANFIS Techniques in Optimization of Process Parameters in Laser Assisted Turning of NITINOL Shape Memory Alloy

Abstract

The rationale behind the rising demand for orthogonal cutting in the processing of hard-to-cut materials is attributed to its advantages over discrete turning processes, which are achieved through the simultaneous application of two or more machining techniques. The utilisation of laser technology in machining processes is considered to be a sophisticated method for the processing of materials that are difficult to cut. The present study involved the utilisation of a Nd:YAG laser source to preheat a Nitinol shape memory alloy (SMA) work piece, which was subsequently subjected to machining using a laser-assisted Computer numerical control (CNC) turning centre. The objective of this investigation was to evaluate the properties of laser assisted machining (LAM) under varying machining conditions. The present study focuses on investigating the impact of process parameters on cutting force and surface roughness in the context of laser-assisted turning (LAT) of Nitinol SMA. To achieve this objective, statistical techniques such as the Response Surface Method (RSM), Adaptive Neurofuzzy Inference System (ANFIS), and Artificial Neural Networks (ANN) with Back Propagation (BP) algorithmbased numerical modelling are employed. The study was conducted utilising the Central Composite Design (CCD) methodology. The study examined the impact of process parameters, specifically cutting speed, feed, cutting depth, and laser power, on the response variables of cutting force (Fz) and surface roughness (SR). The results of the ANOVA analysis indicate that the cutting speed is the primary factor that significantly affects both Fz and SR, accounting for 31.39% and 60.36% of the variance, respectively. The feed rate is the second most influential factor, contributing 26.71% to Fz and 9.078% to SR. The optimisation of LAT parameters was ultimately achieved through the utilisation of a multi-objective desirability function, which effectively reduced both the SR and Fz simultaneously. The evaluation of the ANN model (4-24-2) involved its modelling and expected capabilities, which were compared to those of the RSM model using statistical measures such as root mean square error (RMSE) and absolute standard deviation. The outcomes of Fz and SR anticipated by RSM, ANFIS, and ANN exhibited a high degree of agreement with the empirical findings.