Comparative Analysis of Machine Learning Models for Predictive Maintenance: A Case Study in Manufacturing Company
Fragment książki (Rozdział monografii pokonferencyjnej)
MNiSW
20
Poziom I
| Status: | |
| Autorzy: | Antosz Katarzyna, Kulisz Monika |
| Dyscypliny: | |
| Aby zobaczyć szczegóły należy się zalogować. | |
| Wersja dokumentu: | Drukowana | Elektroniczna |
| Język: | angielski |
| Strony: | 585 - 598 |
| Bazy: | EI Compendex | INSPEC | Norwegian Register for Scientific Journals and Series | SCImago | WTI AG zbMATH |
| Efekt badań statutowych | NIE |
| Materiał konferencyjny: | TAK |
| Nazwa konferencji: | International Congress and Workshop on Industrial AI and eMaintenance 2025 |
| Skrócona nazwa konferencji: | IAI 2025 |
| URL serii konferencji: | LINK |
| Termin konferencji: | 13 maja 2025 do 15 maja 2025 |
| Miasto konferencji: | Luleå |
| Państwo konferencji: | SZWECJA |
| Publikacja OA: | NIE |
| Abstrakty: | angielski |
| This study explores the role of machine availability in improving production efficiency in manufacturing and investigates different machine learning models for predictive maintenance. Frequent machine downtime poses a significant challenge to operational reliability and customer satisfaction, requiring advanced predictive approaches. In this research, models including Support Vector Machines (SVM), K-Nearest Neighbours (KNN), Kernel Logistic Regression and Kernel Naive Bayes were developed using data from a case study in the flooring industry. The dataset included a number of operational and maintenance parameters such as maintenance schedules, staffing levels and equipment usage. The results showed significant differences in model performance. Both SVM and KNN achieved perfect test accuracies of 100%, demonstrating their robustness in predictive tasks. Kernel Logistic Regression followed closely with an accuracy of 99.1%, demonstrating competitive performance. In contrast, Kernel Naive Bayes had a lower accuracy of 86.7%, reflecting its limitations in dealing with complex patterns in the dataset. Error rates were minimal for SVM and KNN (0%), but slightly higher for Kernel Logistic Regression and Kernel Naive Bayes. In addition, the F1 scores for SVM and KNN reached the highest levels, further confirming their reliability in predictive maintenance scenarios. In conclusion, the study highlights the superiority of SVM and KNN models in predictive maintenance applications and demonstrates the transformative potential of predictive analytics in minimising downtime and improving overall equipment effectiveness. These findings provide valuable insights for improving industrial operations through machine learning-driven maintenance strategies. |
