DETECTING ANOMALOUS SENSOR DATA IN WAYSIDE DIAGNOSTICS USING ENHANCED LBP-KURTOGRAMS

Abstract

This paper examines three different methods in comparison for discovering abnormal sensor data retrieved by acoustic and mono-axial accelerometer sensors that are employed in the environment of different train sets and passes to achieve a cost friendly wayside diagnosis in Prague metros. Proposed methodology, Local Binary Patterns (LBP) on resized Kurtogram images is superior to compared methods up to 75.8% Fisher Linear Discriminant Analysis (FLDA) for anomaly detection in the sensor data. Results may count to be promising even if combined acoustic and vibration sensor data related Kurtograms are used for individual train sets. Proposed method is considered to be the first step in order to achieve an efficient diagnosis framework in wayside vehicle diagnosis.

Keywords

• Wayside diagnosis
• local binary patterns
• wavelet packet energy
• detecting abnormal data