Anomaly Detection based on Information-Theoretic Measures and Particle Filtering Algorithms
Marcos E. Orchard, Benjamín E. Olivares, Matías A. Cerda, Jorge F. Silva · Annual Conference of the PHM Society · 2012
This paper presents an anomaly detection module that uses information-theoretic measures to generate a fault indicator from a particle-filtering-based estimate of the posterior state pdf of a dynamic system. The selected measure allows isolating events where the particle filtering algorithm is unable to track the process measurements using a predetermined state transition model, which translates into either a sudden or a steady increment in the differential entropy of the state pdf estimate (evidence of an anomaly on the system). Anomaly detection is carried out by setting a threshold for the entropy value. Actual data illustrating aging of an energy storage device (specifically battery state- of-health (SOH) measurements [A-h]) are used to test and validate the proposed framework.