Detecting Anomalies in Temperature Sensors for Fault Diagnosis in Aviation Meteorology

This study focuses on anomaly detection in univari- ate time-series data for fault diagnosis in temperature sensors in aviation meteorology. The algorithms used in this research include Isolation Forest, One-Class SVM, One-Class SVM with Stochastic Gradient Descent (SGD), Recurrent Neural Network (RNN) with Autoencoder, Robust Covariance Estimation, and Local Outlier Factor. The analysis was conducted on temperature sensor data from the Automated Weather Observation System (AWOS) at the Juanda Meteorological Station in Sidoarjo. Among the various algorithms tested, the aim was to find a suitable algorithm for anomaly detection for sensor fault diagnosis. The test results showed that the RNN with Autoencoder algorithm was the most suitable because it had the lowest false positive rate. However, it is possible to use other algorithms such as Isolation Forest, One-Class SVM and Robust Covariance Estimation. Because it has a significant difference, namely for the number of anomaly points detected. Damage to the temperature sensor can be detected by the number of anomalies in the data. For recommendations on the anomaly detection process, it is best to use one-period data.