Experimental Investigations and Surface Characteristics Analysis of Titanium Alloy Using Machine Learning Techniques.

Milling difficult-to-machine metals like titanium alloys requires a precise surface finish at the nanoscale levels. This examines the surface properties of the Ti-6Al-4V alloy when milled in a tungsten carbide tool coated with TiAlN. With the latest technological improvements, more awareness is being developed to utilize the field of Machine learning. This research is focused on different Machine learning algorithms such as Neural networks (NN), Linear Regression analysis (LR), and Hyperplane based Support Vector Regression (SVR) are being employed to predict the accuracy of the cutting force and Surface finish (Ra) model during milling of Titanium alloy machining. The cutting force is simulated using the AdvantEdge software and compared the results with experimental values using a PVD-coated cemented tungsten carbide tool coated with TiAlN. Taguchi analysis was applied to the milling process in dry conditions using various cutting speeds (120-180 m/min), feed rates (0.05-0.1 mm/tooth), and depths of cut (0.5-1 mm) while applying TOPSIS multi-criterion decision making to get the optimum process parameters for cutting forces. According to the TOPSIS multi-criterion technique, cutting speed has the biggest contribution (64.17%) when compared to the feed rate (18.12 %). Milling cutting force is analyzed using the Deep learning model and compared the results with experiments and error percentile are calculated to obtain the most optimum model for the prediction. The productivity of the manufacturing product is improved by using simulation software to predict the cutting force and surface roughness and implementing intelligent machine learning tools. The polynomial regression model was used to predict the cutting force on milling Ti alloy and found that linear regression prediction was 220.21 whereas polynomial regression prediction was 222.98 at depth of cut 0.5 mm which is more accurate and the curve is fitted properly with most of the data points. [ABSTRACT FROM AUTHOR]