Abstract
Radiation Portal Monitors (RPM) acquire sequential measurements of gamma activity that can be modeled as a sequence of Poisson distributed random numbers. These series can be analyzed with Statistical Process Control (SPC) methods to trigger alarms when the statistical distribution of the ambient dose during a transit changes. Some usual SPC methods have been compared and modified to narrow the time required to bound an alarm. Both the probabilities of detection and the Average Run Length (ARL) are used and fitted to curves to benchmark the SPC methods, which are ranked as: SR, EWMA, CUSUM, SPRT, FLR, and Shewhart.
Similar content being viewed by others
References
Ely JH, Kouzes RT, Geelhood BD et al (2004) Discrimination of naturally occurring radioactive material in plastic scintillator material. IEEE Trans Nucl Sci 51:1672–1676. https://doi.org/10.1109/TNS.2004.832286
Burr T, Hamada MS (2013) Moving neutron source detection in radiation portal monitoring. Technometrics 55:296–308. https://doi.org/10.1080/00401706.2013.775909
Runkle RC, Mercier TM, Anderson KK, Carlson DK (2005) Point source detection and characterization for vehicle radiation portal monitors. IEEE Trans Nucl Sci 52:3020–3025. https://doi.org/10.1109/TNS.2005.862910
Cibils RM (2016) Sensitivity enhancement in radiation portal monitoring using adaptive matched filtering. IEEE Trans Nucl Sci 63:1162–1168. https://doi.org/10.1109/TNS.2016.2524627
Kouzes RT, Ely JH, Evans J et al (2006) Naturally occurring radioactive materials in cargo at US borders. Packag Transp Storage Secur Radioact Mater 17:11–17. https://doi.org/10.1179/174651006X95556
ANSI (2007) American national standard for evaluation and performance of radiation detection portal monitors for use in homeland security
CEI-IEC (2006) Radiation protection instrumentation—installed radiation monitors for the detection of radioactive and special nuclear materials at national borders
IAEA (2006) Technical and functional specifications for border monitoring equipment
ISO (2004) Monitoring for inadvertent movement and illicit trafficking of radioactive material
Kenett R, Zacks S (1998) Modern industrial statistics: design and control of quality and reliability. Brooks/Cole Publishing, Pacific Grove
Montgomery DC (2009) Introduction to statistical quality control, 6th edn. Wiley, London
Frisén M (2007) Statistical surveillance, Optimality and methods. Int Stat Rev 71:403–434. https://doi.org/10.1111/j.1751-5823.2003.tb00205.x
Moustakides GV, Polunchenko AS, Tartakovsky AG (2009) Numerical comparison of CUSUM and Shiryaev–Roberts procedures for detecting changes in distributions. Commun Stat Theory Methods 38:3225–3239. https://doi.org/10.1080/03610920902947774
Ergashev BA (2004) Sequential detection of US business cycle turning points: performances of Shiryaev–Roberts, CUSUM and EWMA Procedures
Lazariv T, Schmid W, Zabolotska S (2013) On control charts for monitoring the variance of a time series. J Stat Plan Inference 143:1512–1526
DeVol TA, Gohres AA, Williams CL (2009) Application of classical versus bayesian statistical control charts to on-line radiological monitoring. J Radioanal Nucl Chem 282:933–938. https://doi.org/10.1007/s10967-009-0255-8
Acknowledgements
The authors would like to thank Prof. J. A. González for his contribution to this work and for his guidance during the experimental analysis, as well as the support provided by the Polytechnic University of Madrid (UPM), Indra, ENRESA, CSN and Spanish Customs. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the supporting entities.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
\( {{\upalpha '}} \) | Error type I (false positive) in a Bayesian scheme |
\( {\text{a}}_{\text{n}} \) | Bottom limit of the interval defined in ROIn |
A, B, C | Auxiliary variables to fit the response of \( \frac{{\mathop \sum \nolimits_{{\updelta_{\rm i} \ge \updelta^{*} }} \widetilde{\text{ARL}}_{{{{\updelta }}^{*} }} /{\text{ARL}}_{0} }}{{\mathop \sum \nolimits_{{\updelta_{\rm i} \ge \updelta^{*} }} i}} \) versus UCL, where \( {\text{A, B, C}} \ge 0 \) |
a, b, c | Auxiliary variables to fit some curves, where \( {\text{a, b, c}} \ge 0 \) |
ARL | Average Run Length. It is defined as the average number of samples from a Poisson distribution before the SPC method signals |
\( {\text{ARL}}_{ 0} \) | ARL of the in-control state in a quasi-infinite sequence being monitored (background case) |
\( \widetilde{\text{ARL}}_{ 0} \) | ARL of the in-control sub-sequence representing the time through the RPM being monitored (background case) |
\( {\text{ARL}}_{{{\updelta }}} \) | ARL of a shift of size \( {{\updelta }} \) in the state in a quasi-infinite sequence being monitored (source case) |
\( \widetilde{\text{ARL}}_{{{\updelta }}} \) | ARL a shift of size \( {{\updelta }} \) in the sub-sequence representing the time through the RPM being monitored (source case) |
\( \widetilde{\text{ARL}}_{{{{\updelta }}^{*} }} / {\text{ARL}}_{ 0} \) | Ratio between the ARL of a change that triggers a true alarm in a transit over the time needed to trigger a false alarm in an infinite sequence. Then, values can be compared between methods as the FAR is the same |
\( {{\upbeta '}} \) | Error type II (false negative) in a Bayesian scheme |
\( {\text{b}}_{\text{n}} \) | Top limit of the interval defined in ROIn |
\( {\text{C}}_{\text{in}} \) | Sequence of gross counts in the channel i and integrated between the channels defined in ROIn |
\( {\text{C}}_{\text{n}}^{{{\# }}} \) | Detection limit of the gross counts \( {\text{C}}_{\text{in}} \) being monitored in ROIn, where \( {{\updelta }} = {{\updelta }}^{ *} \) (minimum shift to trigger an alarm) |
CUSUM | Cumulated Sums |
d | Perpendicular distance between the radioactive source and the detector |
\( {{\updelta }} \) | Size of the shift introduced in the mean above the background level. The size \( {{\updelta }} \) of the shift is sampled every 0.25 units between 0 and 10 |
\( {{\updelta }}^{ *} \) | Period of time in the profile when the alarm is triggered (coverage factor). It is obtained as twice the normal inverse cumulative distribution of the detection probabilities: 0.505, 0.525, 0.6, 0.7, 0.775, 0.8, 0.85, 0.9, 0.9332, 0.95, 0.965, 0.975, 0.98, 0.9825, 0.985, 0.9875, 0.99, 0.995, 0.999, 0.9999 |
EWMA | Exponentially Weighted Moving Average |
\( {\text{f}}_{\text{in}} \) | Background measured at ith position in the window ROIn |
\( {\text{f}}_{\text{Rn}} \) | Background of reference integrated over the range defined in ROIn |
\( {{\mathcal{F}}} ( {\text{x)}} \) | Normal cumulative distribution |
FAR | False alarm rate as the ratio provided by the total amount of times that \( \widetilde{\text{ARL}}_{ 0} \) triggers an alarm with respect to the total number of assays |
FLR | Full Likelihood Ratio as the ratio between the probability of measuring the detection limit \( {\text{C}}_{\text{n}}^{{{\# }}} \) and the probability of measuring the background |
FNR | False Negative Rate as the complement probability of the TAR |
\( {{\uplambda}} \) | EWMA parameter to balance how much distant measurements are considered in the average (\( 0\le {{\uplambda}} \le 1 \)) |
\( {{\Lambda }}_{\text{in}} \) | Full likelihood ratio in ROIn to detect a source as a change in the gross counts sequence \( {\text{C}}_{\text{in}} \) |
\( {{\mathcal{G}}} ( {\text{x}}|\upmu ,\upsigma ) \) | Gaussian distribution of a process of mean \( {{\upmu}} \) and standard deviation \( {{\upsigma}} \) |
\( {\text{m}}_{\text{in}} \) | Net counts after background subtraction |
N | Total number of channel from a spectrum (\( 1\le {\text{a}}_{\text{n}} {\text{ < b}}_{\text{n}} \le {\text{N}} \)) |
\( {{\mathcal{N}}} ( 0 , 1 ) \) | Standardized normal distribution |
NORM | Naturally Occurring Radioactive Material |
\( {\text{N}}_{{{\upsigma}}} \) | Sigma level significance of the Shewhart SPC method |
\( p_{1} , p_{2} \) | Parameters to generalize the UCL according to the variable \( {\text{x}} \) being monitored. When \( {\rm x} = {\text{N}}_{{{{\upsigma}}_{\text{in}} }} \), then p1 = 0, and p2 = 1. Conversely, when \( {\rm x} = {\text{C}}_{\text{in}} \), then \( p_{1} = {\text{f}}_{\text{Rn}} \), and \( p_{2} = \sqrt {{\text{f}}_{\text{Rn}} } \) |
\( {{\mathcal{P}}} ( {\text{x}}|\upmu) \) | Poisson distribution of a process of mean \( {{\upmu}} \) |
PVT | Polyvinyl-toluene |
r | Distance sample-detector |
ROI | Region of interest as a range of channels \( [ {\text{a}}_{\text{n}} , {\text{b}}_{\text{n}} ] \) from the spectra where the integral is obtained |
RPM | Radiation Portal Monitor |
RSP | Radiation Sensor Panel |
\( {{\upsigma}}_{{{\text{f}}_{\text{Rn}} }} \) | Standard deviation of the background of reference in ROIn |
\( {\text{s}}_{\text{ij}} \) | Original spectra acquisition from the detector’s channel i at time j |
\( {\text{S}}_{{ 0 {\text{n}}}} \) | Initial condition of a recursive statistic from a SPC method. It can be set to a value different than zero to more steadily detect a shift at the beginning of the monitoring |
\( {\text{S}}_{\text{in}} \) | Recursive value of the statistic used in a SPC method. Its value is compared with the UCL to trigger an alarm when \( {\text{S}}_{\text{in}} \ge {\text{UCL}} \) |
SPC | Statistical Process Control |
SPRT | Wald’s Sequential Probability Ratio Test |
SR | Shiryaev–Roberts procedure |
t | Time spent during the sequence being monitored |
T | Time (fixed) of the length of the subsequence corresponding to a radioactive profile/subsequence (transit) |
TAR | True Alarm Rate as the ratio provided by the total amount of times that \( \widetilde{\text{ARL}}_{{{\updelta }}} \) triggers an alarm with respect to the total amount of essays |
TAR* | TAR probabilities when the FAR \( { = 0} . 0 1 \)% |
TNR | True Negative Rate as the complement probability of the FAR |
UCL | Upper Condition Limit of the SPC to trigger an alarm in the corresponding method |
v | Velocity of the radioactive source during the profile/subsequence. Its value is sampled from 0.5 to 6 m/s by 0.5 m/s steps to represent most of the monitoring scenarios |
Rights and permissions
About this article
Cite this article
Armenteros, J.C., Suárez, M.J. & Pujol, L. Dynamic numerical recipes to improve detection schemes for the interdiction of radioactive shipment. J Radioanal Nucl Chem 318, 865–875 (2018). https://doi.org/10.1007/s10967-018-5958-2
Received:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10967-018-5958-2