Benchmarking of Machine Learning Algorithms for Temperature Sensor Fault Time Series Classification

Reliable sensors functioning is important for safe and efficient operation of a nuclear system. Accurate detection and classification of sensor anomalies are crucial for efficient maintenance and prevention of reactor shutdowns. In this proof-of-principle study, we benchmark performance of several machine learning (ML) algorithms in temperature sensor fault detection. The database consists of four time series, with one representing normal sensor operation and the remaining three consisting of various sensor faults. The ML models in this study include Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM) networks, Decision Tree (DT), and Extreme Gradient Boosting (XGBoost). Performance of the ML models is evaluated using Precision, Recall and F1-score metrics. The results of the benchmarking study show that XGBoost achieves the best performance.