IEEE 2017 NSS/MIC/RTSD ControlCenter

Online Program Overview Session: N-26

To search for a specific ID please enter the hash sign followed by the ID number (e.g. #123).

Data Analytics

Session chair: Caroline Shenton-Taylor University of Surrey, UK; Michael C. Wright Oak Ridge National Laboratory, USA
 
Shortcut: N-26
Date: Wednesday, October 25, 2017, 13:40
Room: Regency V
Session type: NSS Session

Contents

1:40 pm N-26-1 Download

Robust Detection of Radiation Threat by Simultaneous Estimation of Source Intensity and Background (#1957)

E. Lei1, K. Miller1, A. Dubrawski1, K. Nelson2, S. E. Labov2

1 Carnegie Mellon University, Auton Lab, Pittsburgh, Pennsylvania, United States of America
2 Lawrence Livermore National Laboratory, Nuclear and Chemical Sciences Division, Livermore, California, United States of America

Content

Detecting, localizing, and characterizing potentially dangerous sources of radiation is an important problem in national security, medical, and industrial operations. A common approach to this problem is to build statistical models that detect anomalous, possibly threatening radiation. Existing methods commonly assume that a comprehensive training set of background radiation measurements is available and that the characteristics of this background agree with those observed in a live run. By relaxing these assumptions, we investigate a scenario in which a discrepancy may exist between background radiation in training and test sets or in which there is no training set at all. This scenario corresponds to many practical use cases. We propose a new method that is robust to these uncertainties by adapting to the local distribution of background radiation in real-time. In particular, a Kalman filter is employed to simultaneously estimate source intensity and background spectrum. The estimated background spectrum is then passed as input to an existing method called Gaussian-Poisson (GP) MAP Estimation, a Bayesian method that computes a likelihood ratio between the hypotheses of source and no source by modeling radiation as Poisson variates under a Gaussian prior. While the original GP method exhibits state-of-the-art performance in conventional settings, our modification with the Kalman filter is more robust when the training set is uninformative. We conducted experiments on an authentic radiation dataset collected in a noisy urban environment by the RadMAP project. The background spectra in the training set were shifted to render them less informative. By simulating an Americium-241 source at randomly selected locations, we tested roadside detection in a single pass of a sensor. The Kalman filter method had significantly better performance than the GP method while requiring no training set. Our method could advance the state-of-the-art in many practical situations.

Keywords: radiation, threat, source, detection, classification, training, kalman filter, gaussian, poisson
1:58 pm N-26-2

Computational Techniques for Optimizing Performance of a LiF:ZnS(Ag) Neutron Detector using Recorded Waveforms (#1264)

K. Pritchard1, 2, A. Osovizky1, 4, N. Maliszewskyj1, E. Binkley1, M. Jackson3, P. Tsai1, C. Hurlbut3, N. Hadad1, C. Majkrzak1, J. Ziegler1, J. Gelmann1

1 National Institute of Standards and Technology, NIST Center for Neutron Research, Gaithersburg, Maryland, United States of America
2 University of Maryland Baltimore County, CSEE, Baltimore, Maryland, United States of America
3 Eljen Technology, Sweetwater, Texas, United States of America
4 Rotem Industry Ltd., Mishor Yamin, Israel

Content

Data analysis and pattern recognition techniques optimized the performance of an ultra-thin LiF:ZnS(Ag) scintillating neutron detector. Using datasets of recorded waveforms, Pulse Shape Discrimination algorithms were designed to maximize neutron sensitivity with very high selectivity against gamma radiation and noise. A high purity dataset of neutron waveforms, 99.9%, was recorded using a neutron reactor cold source at the NIST Center for Neutron Research. Likewise, a high purity gamma dataset, >99.5%, was recorded using a Cs-137 gamma source.

We analyzed the energy and decay time of the datasets using the Two-Window Charge Comparison algorithm. With a special statistical technique, we determined the two ideal integration windows for our detector are from 0ns to 100ns and from 101ns to 2300ns following the pulse rising edge. We accurately computed our neutron detector’s Neutron Sensitivity vs. Gamma Selectivity from the waveform datasets, and we experimentally confirmed the accuracy of this sensitivity/selectivity curve. Gathering the neutron and gamma datasets and computing our detector performance was a relatively quick procedure. We used this procedure to compare a series of detector configurations, and we optimized our silicon photomultiplier (SiPM) bias voltage for neutron sensitivity. The statistical techniques presented have potential uses throughout the radiation detection field.

Keywords: neutron detector, LiF:ZnS(Ag), CANDOR, support vector machine, SVM, radiation detector, silicon photomultiplier, SiPM, pulse shape, charge comparison algorithm
2:16 pm N-26-3

Rapid Kalman filter stabilisation technique for single- and multi-detector systems (#2887)

M. J. Neuer1, 2, C. Henke1, E. Jacobs1

1 innoRIID GmbH, Research & Development, Grevenbroich, North Rhine-Westphalia, Germany
2 VDEh Betriebsforschungsinstitut, Quality and Information Technology, Düsseldorf, North Rhine-Westphalia, Germany

Content

Most spectroscopic applications, especially homeland security relevant equipment relies on well-calibrated spectral data. A Kalman filter is applied to the known problem of stabilization, which means online calibration of a scintillation detector to a predefined reference peak. The method uses stochastic assumptions on the measurement uncertainties and an iterative update with new measurements. Introducing prior knowledge about the stochastic and continuously resupplying new spectral data to the filter enhances the predicted gain correction value and allows a rapid determination of this value.

In multi-detector setups, the spectral alignment of data from all detectors is mandatory for performing data fusion or any further evaluations. The described Kalman technique also allows to perform this rapid cross-detector synchronization online, leading to matched up spectra at acquisition time. 

For a 3’’x1’’ sodium iodide detector and a 2’’x1’’ LaBr3 detector, the proposed stabilisation is tested and convergence times towards reliable gain correction values are determined. In the same setup, the cross-detector synchronisation is tested, yielding an alignment of NaI and LaBr3 within very short times.

Keywords: Kalman filter, Stabilization, Scintillation Detectors, Data fusion
2:34 pm N-26-4

Application of Multivariate Data Analysis Techniques for the Portable Isotopic Neutron Spectroscopy system (#1694)

D. Lee1, J. Wharton1, B. Bucher1, A. Caffrey1, K. Krebs1, E. Seabury1

1 Idaho National Laboratory, Global Security & International Safeguards, Idaho Falls, Idaho, United States of America

Content

The Portable Isotopic Neutron Spectroscopy (PINS) is a commercialized system developed by Idaho National Laboratory (INL) to examine chemical compounds non-destructively, utilizing Prompt Gamma Neutron Activation Analysis (PGNAA) techniques. The PINS systems have been successfully deployed around the world to identify chemical munitions and containers. The PINS system takes advantage of a high-resolution gamma-ray spectrum from a high-purity germanium (HPGe) detector, and gamma-ray peak analysis provides input to its chemical identification logic with the probabilistic decision tree (PDT). Gamma-ray peak analyses, however, require knowledge of pre-selected gamma-ray peaks whose energies, intensities, and origins are well studied, and some peaks might be excluded from the decision making processes due to lack of current nuclear data. In contrast, Multivariate Analysis (MVA) treats a whole gamma-ray spectrum as a collection of multiple variables or a pattern. The effectiveness of chemical identification algorithm is determined by the availability of a wide range of data to train an algorithm to identify chemical fills with accuracy. INL has a collection of gamma-ray spectra of various chemical-fills from the field-deployed PINS systems over the years, and it was envisaged to combine such a database with the principle of multivariate statistics. Therefore, in parallel with the current decision tree algorithm, an MVA-based chemical identification algorithm was developed as an independent verification of the current results. The Principal Component Analysis (PCA) method was adopted to build an MVA-based identification algorithm, using the collected field data to refine and validate the algorithm. A benchmarking study of the MVA-based algorithm’s performance was conducted alongside the current algorithm, with the results presented and discussed in this study.

Keywords: Prompt Gamma Neutron Activation Analysis, HPGe detector, gamma-ray spectroscopy, Multivariate Data Analysis, Principal Component Analysis, Portable Isotopic Neutron Spectroscopy, Chemical Warfare Agents, Chemical Identification Algorithm, Mahalanobis Distance
2:52 pm N-26-5

Facility On/Off Inference by Fusing Multiple Effluence Measurements (#1782)

C. Ramirez1, N. Rao1

1 Oak Ridge National Laboratory, Computational Science and Engineering Division, Oak Ridge, Tennessee, United States of America

Content

Inferring the operational status of a reactor facility using measurements from an independent in-situ monitoring system is critical to the assessment of its compliance to agreements. In particular, such a monitoring system could assist in identifying activities beyond the agreed upon ones, for instance, longer operational periods. In this paper, we consider the problem of inferring the on/off status of a reactor facility by using effluence measurements of three gases, namely, Ar-41, Cs-138, and Xe-138, which are collected on the facility’s stack. We first present classifiers based on thresholding the measurements of individual effluence types, and then present methods that combine their outputs or measurements. We develop sample-based implementations of four fusers based on a simple majority rule, Chow's recognition function, physics-based radiation counts model, and correlation-coefficient method. We apply the latter three fusers to pairs and all three gas effluence types. Our results show that: (i) these gas effluence measurements are effective in inferring the on/off status of a reactor facility, for example, best fusers achieve 97% detection at 1% false alarm rate, and (ii) the performance depends on the data and classification method, and in particular, fusers that combine three effluence types based on physics-based models and correlation-coefficients outperform the majority rule and Chow's fusers as well as individual and pairs of effluence types, thereby illustrating the importance of the fuser choice.

This work has been carried out at Oak Ridge National Laboratory managed by UTBattelle, LLC for U.S. Department of Energy under Contract No. DE-AC05-00OR22725. This research was supported in part by an appointment to the Higher Education Research Experiences Program at Oak Ridge National Laboratory.

Keywords: reactor facility, analytics, on/off inference, gas effluence, classifier, fuser
3:10 pm N-26-6

Background Spectrum Estimation from Panoramic Images (#3177)

C. Kaffine1, B. Pires1, A. Laddha1, D. Bayani1, K. Miller1, M. Hebert1, A. Dubrawski1

1 Carnegie Mellon University, Robotics Institute, Pittsburgh, Pennsylvania, United States of America

Content

A key challenge for radiation threat detection is the wide variability of the background radiation spectra. Current methods often utilize a statistical model of background to reduce the impact of this variance. In contrast, this work presents a method for estimating the background spectrum explicitly and incorporating this prediction in standard threat detection methods. We aim to predict background from panoramic images to impart knowledge of the surrounding environment to the threat detection model. The approach relies on a Convolutional Neural Network to regress the spectrum directly from the panoramic image. We show that this estimate can be used to improve spectral anomaly detection methods.

Keywords: Threat Detection, Background Estimation