Advanced Machine Learning Approaches for Predicting Machining Performance in Orthogonal Cutting Process

Abstract

We investigated the orthogonal cutting process by using machine learning models to predict its performance. This study used the AZ91 magnesium alloy as the workpiece material, and machining was performed under the Minimum Quantity Lubrication (MQL) technique. The input parameters were the feed, cutting speed and MQL flow rate. Additionally, the outputs were flank tool wear, the chip contact length, peak distance, valley distance, pitch distance, chip segmentation ratio, compression ratio and shear angle. Studies on machine learning (ML) models being employed to evaluate the performance of the MQL-assisted orthogonal machining of AZ91 are very rarely found in the literature. This study explored machine learning (ML) as a data-driven alternative, evaluating decision tree regression, Bayesian Optimization, Random Forest Regression and XGBoost for predicting machinability. A comprehensive dataset of the cutting parameters and outcomes was utilized to train and validate these models, aiming to enhance the accuracy of the predictive analysis. The performance of each model was evaluated based on error metrics such as the mean squared error (MSE) and R-squared values. Among these models, XGBoost demonstrated a superior predictive accuracy, outperforming the other methods in terms of its precision and generalizability. These findings suggest that XGBoost provides a more robust solution for modeling the complexities of the orthogonal cutting process, offering valuable insights into process optimization. The analysis supports that the XGBoost model is the most accurate, with a 34.1% reduction in the mean squared error and a 17.1% reduction in the mean absolute error over these values for the Decision Tree. It also outperforms the Random Forest Regression model, achieving a 19.8% decrease in the mean squared error and a 7.1% decrease in the mean absolute error.