Machine Learning Based Fault Diagnosis in Variable Speed Synchronous Generators

In this study, the performance of five different machine learning algorithms, decision tree, random forest, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and logistic regression, was investigated for fault diagnosis in variable-speed synchronous generators. The dataset, consisting of real-world experimental data, includes both healthy and faulty generator operating states. Pre-processing steps such as normalization, Z-Score standardization, and feature selection were applied to the data, and the effects of these processes on classification performance were evaluated. According to the findings, the decision tree algorithm achieved the highest performance with an accurate rate of 99.43% and Matthews Correlation Coefficient (MCC) value of 0.975. While the random forest algorithm yielded similar results, the KNN, SVM, and logistic regression algorithms achieved lower accuracy values. It was determined that the pre-processing steps did not provide a significant increase in model performance, and the dataset was already balanced in terms of scale. The results revealed that the decision tree algorithm is the most suitable and reliable method for fault detection in variable-speed synchronous generators. This study demonstrates that machine learning-based approaches can be used effectively in early fault diagnosis in generators.