Multivariate Modelling of Effectiveness of Lubrication of Ti-6al-4v Titanium Alloy Sheet using Vegetable Oil-Based Lubricants

Abstract

Abstract The article presents the results of modelling the friction phenomenon using artificial neural networks and analysis of variance. The test material was composed of strip specimens made of 0.5-mm-thick alpha-beta Grade 5 (Ti-6Al-4V) titanium alloy sheet. A special tribotester was used in the tests to simulate the friction conditions between the punch and the sheet metal in the sheet metal forming process. A test called the strip drawing test has been conducted in conditions in which the sheet surface is lubricated with six environmentally friendly oils (palm, coconut, olive, sunflower, soybean and linseed). Based on the results of the strip drawing test, a regression model and an artificial neural network model were built to determine the complex interactions between the process parameters and the friction coefficient. A multilayer perceptron with one hidden layer and eight neurons in this layer showed the best fit to the training data. The network training was conducted using three algorithms, i.e. Levenberg-Marquardt, back propagation and quasi-Newton. Taking into consideration both the coefficient of determination R2 (0.962) and S.D. ratio (0.272), the best regression characteristics were presented by the network trained using the Levenberg-Marquardt algorithm. From the response surfaces of the quadratic regression model it was found that an increase in the density of lubricant at a specific pressure causes a reduction in the coefficient of friction. Low density and high kinematic viscosity of the oil leads to a high coefficient of friction.