Evaluating machine learning models for predictive maintenance in industrial setting : A comparative study
Irfan, Mueed (2024)
2024
Master's Programme in Computing Sciences and Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2025-05-30
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-2024112610522
https://urn.fi/URN:NBN:fi:tuni-2024112610522
Tiivistelmä
The industrial sector generates vast volumes of data from various sources, necessitating advanced methods for managing and deriving insights. This thesis investigates machine learning (ML) approaches for predictive maintenance, focusing on preventing equipment failures through data-driven predictions. Utilizing a dataset from Valmet Automation Oyj, comprising sensor data from heavy industrial machinery, this study addresses challenges such as significant class imbalances and data preprocessing.
A comparative analysis of ten machine learning algorithms, ranging from decision trees to neural networks, is conducted to evaluate their performance in predictive maintenance. Metrics including accuracy, ROC-AUC, and classification reports are used for assessment. Additionally, methods like Synthetic Minority Over-sampling Technique (SMOTE) are employed to mitigate class imbalance, enhancing the detection of minority-class failure events.
The research proposes a robust framework integrating preprocessing techniques, feature engineering, and advanced classification algorithms. Practical applications of this framework are validated using real-world data, emphasizing the balance between model complexity and interpretability. The findings aim to improve predictive maintenance strategies, minimize operational disruptions, and optimize resource allocation in industrial settings.
A comparative analysis of ten machine learning algorithms, ranging from decision trees to neural networks, is conducted to evaluate their performance in predictive maintenance. Metrics including accuracy, ROC-AUC, and classification reports are used for assessment. Additionally, methods like Synthetic Minority Over-sampling Technique (SMOTE) are employed to mitigate class imbalance, enhancing the detection of minority-class failure events.
The research proposes a robust framework integrating preprocessing techniques, feature engineering, and advanced classification algorithms. Practical applications of this framework are validated using real-world data, emphasizing the balance between model complexity and interpretability. The findings aim to improve predictive maintenance strategies, minimize operational disruptions, and optimize resource allocation in industrial settings.