A unified meta based machine learning model for sustainable manufacturing using characterization and regression of machinng data

Sangeetha, Elango (2023) A unified meta based machine learning model for sustainable manufacturing using characterization and regression of machinng data. PhD thesis, UTAR.

Preview

PDF Download (2014Kb) | Preview

Abstract

Polyoxymethylene(POM), Polytetrafluoroethylene(PTFE), Polyether ether ketone(PEEK) and multiwall carbon nano tubes reinforced PEEK(PEEK/MWCNT) are significant polymeric materials used in industrial applications and house-hold items and hence they were considered for this research. Considering its potentiality, this research was attempted to investigate and develop a unified meta based machine learning model for turning of different polymeric materials. Developed meta based model has two main parts: classifier and regressor. Classifier is the one initially taking the experimental data and classify them into reasonable and more accurate groups. Regressor use the output from the classifier and predict the surface finish. For classification, XGBoost algorithm and Logistic regression algorithm were investigated. k-fold cross validation method was adopted to apply all data patterns in learning of the model. Grid searching method was used to tune thehyper parameters for each algorithm. It was found from these results that Logistic Regression model is the better to be used as classifier. Once classifier model was confirmed, the output of the classifier was added to the database as a new feature. Now, with four independent features (including output of classifier), Support vector regressor and XGBoost algorithm were used to complete meta based model for each material. Further, a unified model (a model for all four polymeric materials) was developed using the same procedure as discussed above. It used an additional input feature known as material number. Interestingly observed that XGB model is the best model working great in both classification and regression. It resulted almost 100% accuracy in training and 98.86% in testing. After confirming the best model, a group of predicted results was generated from the prediction model and validated experimentally. Finally, user interface (API) was developed for the unified meta based model which industry can use in its production line for achieving high productivity.

Actions (login required)