A framework for prediction of extrusion responses using machine learning algorithm

The objective of the present study is to develop a Customized Automated Machine Learning (CAML) framework to support machine learning models in manufacturing processes. The proposed CAML framework features a user-friendly interface that enables users to perform key tasks and predict outcomes based on user-defined process parameters. In the present study, machine learning-based prediction of output responses is employed to estimate the process parameters during the cold extrusion process. The input responses are Die angle, Ram speed and Coefficient of Friction while the output responses are Extrusion Force, Damage Factor, Displacement of work piece and Extrusion Time. The framework utilized various machine learning algorithms, including Linear Regression, Ridge, Lasso, Elastic Net, Polynomial Regression, Gaussian Process Regressor, XGB Regressor, LGBM Regressor, Random Forest, Gradient Boosting Regressor, AdaBoost Regressor, Bagging Regressor, Extra-Trees Regressor, KNN Regressor, Nu SVR, Support Vector Regression (SVR), Kernel Ridge, RANSAC Regressor, Huber Regressor, LarsCV, Orthogonal Matching Pursuit, and Bayesian Ridge. The evaluation was performed by analyzing the aggregate R² score and aggregate Root Mean Absolute Error (RMSE). Additionally, for optimization, a machine learning-based algorithm, Multimodal Optimization NSGA-2 (Non-dominated Sorting Genetic Algorithm 2), is employed to predict extrusion process parameters and enhance the efficiency of the actual extrusion operation. This approach bridges the gap between simulation results and real-world production systems. For Extrusion Force, Displacement of work piece and Extrusion Time. accuracies are 99.38%, 99.80%, and 99.25% respectively, with error percentages below 1% for GBR. However, the Damage Factor, with smaller values 0.017, 0.014), showed higher error (8.24%), where XGBR proved more consistent. The training of all machine learning models (MLMs) took 17.89645 s based on R2 and 18.9131 s based on RMSE.