Fault Detection in Turbines Using Machine Learning: A study of the capabilities of Various Classification Algorithms

This paper presents a comprehensive study of application of different machine learning techniques for the prediction of turbine faults in an early stage. As the need to ensure an optimal turbine operation through early stage fault detection grows, precise predictive models using machine learning techniques are also increasing rapidly. This study primarily focuses on development and evaluation of the predictive models capable of anticipating fault phases using different range of classification algorithms. It assesses the effectiveness of different classification methods including Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest, and Convolutional Neural Networks (CNNs) to predict distinct turbine phases. These algorithms are evaluated on experimented dataset encompassing various stages of faults. The results from the algorithm show that CNN was accurately predicting the particular turbine fault phases achieving 80% accuracy with a small length of time series. The deep learning capabilities of CNN capture intricate spatial patterns in turbine data enabling them to discern fault indicators eluded by traditional machine learning methods. The insights gained from this research helps to extend the application of machine learning approaches in the field of turbine fault detection, paving the way for more precise and dependable early failure detection systems.