Sammendrag
Turbines, the backbone of numerous industries for power generation, assume a critical role in ensuring seamless operations and a continuous power supply to the system. Their uninterrupted working conditions and performance are crucial to providing continuous power to the system. The impact of turbine tripping due to various factors can lead to substantial economic losses, operational downtime, and decreased production efficiencies. Therefore, early identification of potential turbine failures becomes paramount in optimizing resource allocation and reducing costly disruptions. In this context, the present thesis endeavors to comprehensively analyze and preprocess real-world data, aiming to develop a robust predictive model for turbine failure. This study embarks on a two-fold approach to achieve its objectives effectively. Initially, the raw sensor data is transformed, enabling exploration and experimentation with diverse methodologies to attain precise predictions. Subsequently, machine learning techniques, including Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), Logistic Regression, Support Vector Machine (SVM), and K-Nearest Neighbor (KNN), are employed to forecast turbine failure within a sixty-minute timeframe before it occurs. The implemented methods demonstrate promising results, with the models achieving better accuracy in comparing actual and predicted turbine failures within the 60-minute window. The implications of this research are profound, particularly in enhancing proactive maintenance practices and resource management strategies within industrial settings. By adapting to the predictive capabilities of these models, industries can implement well-timed maintenance actions and prepare contingency plans, thereby effectively lessening potential losses resulting from turbine failures. The findings thus signify a major advancement in the domain of predictive analysis for turbine failure detection, imparting long-term benefits to industrial processes and overall operational resilience.