Central AsianResearch Index

Fault Diagnosis Using Limited Vibration Dataset with Hyperparameter Optimization

Fault diagnosis in mechanical systems plays a critical role in predictive maintenance and operational safety. This study presents a machine learning-based approach for identifying common mechanical faults using a limited vibration dataset. An experimental setup was developed to simulate and record vibrations from various conditions, including healthy operation, bearing faults, gear defects, and screw looseness. Vibration data were collected in three dimensions and transformed into the frequency domain. Several classification algorithms, including Naïve Bayes, Logistic Regression, Decision Table, Random Tree, and Random Forest, were trained to detect these faults. Additionally, Auto-WEKA was employed to optimize both algorithm selection and hyperparameters. Among all tested methods, the Random Forest classifier with hyperparameter optimization achieved superior performance with a classification accuracy of 97.1% and an estimated error rate of 0.85%. The study highlights that even with a relatively small dataset, high diagnostic performance is achievable through proper algorithm selection and hyperparameter tuning. The findings are particularly relevant for applications in industrial environments where collecting large labelled datasets is costly or impractical.

Сведения

Опубликовано
2026 · № 2, С. 46
Язык
Английский
ISSN
2706-977X
Издатель
Karaganda Technical University named after Abylkas Saginov
Аффилиации
Sultan Qaboos University