Abstract
The service life of investment casting products is measured through its mechanical properties like ultimate tensile strength, yield strength, percentage elongation, hardness etc. These mechanical properties are procured through destructive testing which is time consuming and leads to material wastage. In the past, some machine learning models are utilized to predict the mechanical properties using the chemical composition and process parameters of the investment casting process. This industrial data contains a large number of input variables, which are complex to model and results in low prediction accuracy. In this proposed paper, two feature selection technique named least absolute shrinkage and selection operator (LASSO) and variable selection using random forests (VSURF) are implemented to select significant features from a total of 25 independent variables which are utilized for predicting the mechanical properties for the investment casting process. The efficacy of selected features is also evaluated by several machine learning models, including random forest (RF), K-nearest neighbor (KNN) algorithm and extreme gradient boosting (XGBOOST). The results show that the VSURF can extract a smaller subset of critical variables compared to LASSO, which helps to enhance the prediction accuracy and interpretation of the machine learning models; XGBOOST has the best capability to predict mechanical properties with the highest accuracy.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have