Satellite battery fault detection using Naive Bayesian classifier

Control ground station performs satellite commanding and onboard subsystems health monitoring in addition to satellite tracking (orbit determination) and attitude determination based on received telemetry. The traditional approach for satellite health monitoring is to check sensor values whether being within permissible ranges or not. This paper proposes a supervised Naive Bayesian classifier to build data driven models that detect power supply system anomaly. Battery flight test data has been used as the abnormal class, representing possible failures, an approach to overcome the problem of unavailability of labeled abnormal data in a supervised classification. Data used to build the model is a three months observation of battery's capacitance, voltage, temperature and pressure. Data has been subjected to principal component analysis before Naive Bayesian classifier model building for visualization, labeling training data and to increase the variables independence as a restriction imposed by the naive Bayesian classifier. The built Naive Bayesian classifier model is validated using real faulty data in terms of accuracy, recall, precision, F1-score and ROC curve.