Development of A Hybrid Neural Network System For Prediction and Optimization of Process In Cryogenic Machining of 316 Series Stainless Steel

Abstract

The high cutting temperature developed during machining at high cutting velocity and feed rate affects the ability to achieve high productivity and quality. It also causes dimensional deviation, premature failure of cutting tools, impairs the surface integrity of the product by inducing tensile residual stresses, and induces surface and subsurface micro cracks in addition to rapid oxidation and corrosion. Unlike conventional coolants which generally cause environmental and health problems to the machine operators, Cryogenic machining using LN2 is an environmentally safe coolant which can achieve desirable control of cutting temperature and improve the machining performance. Many researchers have tried different cryogenic cooling methods such as cryogenic pre-cooling the workpiece, indirect cryogenic cooling or cryogenic tool back cooling and cryogenic jet cooling by micro-nozzles on the cutting tool edges or faces, tool–chip and tool–work interfaces. In the present research work, cryogenic cooling system was developed for supplying LN2 at tool-chip interface during milling process. The machining study was conducted on SS316 of work material under dry, wet and cryogenic machining environments with the following work – tool combination i.e. SS316 steel Physical Vapour Deposition - TiAlN coated. The performance of the milling study involves three different cooling approaches. They were: (i) Dry machining (ii) Wet machining (iii) Cryogenic machining. In cryogenic environments, the LN2 was supplied at the tool – chip interface under constant pressure of three bar, using nozzle. The experimental results of cutting temperature, cutting force, surface roughness under cryogenic cooling were compared with those of dry and wet machining. With artificial neural network, prediction of responses of milling process are carried out using 4 different error back propagation algorithms such as (Gradient Descent, Scaled Conjugate Gradient Descent, Levenberg Marquart and Bayesian regularization or Bayesian Neural Network) models. Later, predicted results were compared between the conventional and non-conventional techniques and best suitable back propagation was identified for the current study. The validity of the models was established. The artificial neural network model formulated for cutting temperature cutting force, surface roughness and tool wear are found to predict the corresponding responses quite accurately. Support vector regression and machine learning techniques were applied for prediction using Regression- Epsilon Method by using various kernel functions (Linear, Polynomial, Sigmoid, and Radial Basis Function). The best kernel function suitable was identified. Later on, incorporation of support vector machine to optimization (Particle Swarm Optimization was introduced in order to build the novel hybrid model).