Falcon: Fault Analysis and Learning-Based Classification for Operational Normality in Power Plant Condensers

Condensers maintain the vacuum that a steam turbine needs to deliver its rated output, so any deterioration in their condition directly reduces plant efficiency and availability.In practice, condenser fault diagnosis still depends on periodic inspection and operator judgement, which is slow, subjective and likely to miss faults at an early stage.This paper presents a machine learning framework that classifies the operating state of a power plant condenser as normal or faulty using routinely logged operating parameters.The raw records are cleaned through missing-value handling, outlier treatment and normalization.Additional physically meaningful features are then derived from the measured parameters, and correlation analysis retains the features most strongly associated with fault occurrence while discarding redundant ones.The selected features are used to train an Extreme Gradient Boosting (XGBoost) classifier, which is evaluated using accuracy, precision, recall, F1-score and the confusion matrix.Experiments on the AI4I 2020 operational fault dataset show that XGBoost achieves 99.15% accuracy and an F1-score of 86.40% on the fault class, clearly outperforming SVM, KNN and Naive Bayes, which supports early fault detection and predictive maintenance in power plant operation.