Comparative analysis of support vector machine and logistic regression for gear fault detection and classification

Abstract

Bevel gears play a crucial role in mechanical systems, particularly in power transmission applications where changes in shaft orientation are required. Early faults detection and diagnosis in gearboxes are essential for ensuring operational efficiency and preventing costly failures. This study evaluates and compares the effectiveness of Logistic Regression (LR) and Support Vector Machine (SVM) classification methods in identifying gear faults, focusing on data-driven condition monitoring rather than predictive maintenance. To enhance the performance of these models, hyperparameter tuning was performed using Grid Search Cross Validation. These techniques are essential for improving model performance, reducing overfitting, and increasing classification accuracy. Gear fault classification was carried out using vibration data from a test bench under different speeds, loads and measurement directions. While LR initially achieved higher accuracy (64.10%) compared to SVM (38.46%), hyperparameter tuning significantly improves SVM performance, allowing it to reach 92.31% accuracy compared to 82.05% for LR. These findings underscore the capability of the optimized SVM model to provide more sensitive and precise fault diagnosis, highlighting its suitability for robust data-driven diagnostics of gear conditions.