An Integrative Framework for Real-Time Anomaly Detection in Source Search Using Fuzzy Inference, DDDAS, and Matrix Profile

Abstract

Securing metropolitan areas against nuclear threats has been a central objective of nuclear security since 9/11. A key defense strategy is source search, where radiation spectrometers are used to identify anomalies—sources that may pose potential threats—within consecutively acquired spectra. This article presents a novel framework for real-time anomaly detection in gamma spectra collected at short intervals with a mobile detector. The framework integrates three components: dynamic data-driven application systems (DDDASs), matrix profile (MP), and fuzzy inference. This combination enables efficient spectral processing and timely decision-making regarding the presence of anomalies. At its core, DDDAS maintains and continuously updates an MP-based model of background radiation as new spectra are acquired. Anomaly detection is then carried out by a fuzzy inference system that uses four features derived from the stored MP model, the MP itself, and the Kolmogorov–Smirnov (KS) distance of the most recent spectra. Performance is evaluated by introducing source spectra as anomalies at random points in background datasets. Results show that the framework achieves high detection accuracy with very low false alarm rates, while delivering decisions in under one second, making it suitable for real-time applications. Furthermore, comparison with the conventional spectral ratio method based on the KS test and sequential Bayesian detection highlights a substantially lower false alarm rate—about an order of magnitude reduction. Notably, in several test scenarios, the framework yielded zero false alarms per 1000 spectra, underscoring its effectiveness and robustness for nuclear security operations.