To what extent is the implementation of tree-based models effective and feasible in predictive maintenance under Industry 4.0?

Abstract

The technology available in Industry 4.0, combined with the constantly evolving machine learning techniques have allowed managers and stakeholders of the man ufacturing, transportation, and logistics sectors to predict the likelihood of their machines failing at an unprecedented reliability. Although Machine learning modelshave become highly complex in recent years, tree-based models starting from Decision Trees to Random Forests and boosting models, continue to exist in academia due to their feasibility, interpretation and even effectiveness. The purpose of this study is to grasp the relevance of such models in the real-world and whether they are worth investing into. To guide the process of building tree-based models, the Cross-Industry Standard Process for Data-Mining (CRISP-DM) will be utilized in order to understand the business as well as the data, prepare the data, create and evaluate the models, and build a deployment strategy in an iterative and flexible manner. A synthetic dataset that simulated the sensor data of a milling machine was used for the research, and the results through multiple evaluation techniques,indicated that the boosting model, XGBoost, outperformed the Random Forest,Decision Tree, and Logistic Regression models. Though the models from other re search outperformed XGBoost, however, XGBoost along with Random Forests are advised to still be taken under consideration due to their feasibility to be produced, trained, and interpreted, while still generating good results.

Date of Award	26 Jan 2023
Original language	English
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Pedro Afonso Fernandes (Supervisor)