Citation
Spectroscopic analysis for a data-fusion based radiation detection system

Material Information

Title:
Spectroscopic analysis for a data-fusion based radiation detection system
Creator:
Martin, Allan ( author )
Language:
English
Physical Description:
1 online resource (15 pages) : illustrations ;

Subjects

Subjects / Keywords:
Nuclear Engineering thesis, M.S
Dissertations, Academic -- Materials Science and Engineering -- UF
Genre:
Electronic Thesis or Dissertation.
government publication (state, provincial, terriorial, dependent) ( marcgt )
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )
Electronic Thesis or Dissertation

Notes

Abstract:
A data-fusion based radiation detection system capable of detecting and tracking possible radiological threats is under development at the University of Florida. As an expansion to the capabilities of this system, an algorithm was expanded to focus on particular isotopes of interest in the spectra. This was accomplished by only analyzing radiation counts from under the photopeaks of the gamma ray spectra during the tracking as determined by user defined regions of interest (ROI). This would allow the system to analyze specific sections of the radiation detector data stream, with the goal of increasing the speed and accuracy with which a source is tracked and also let it be identified. Results from cesium-137 and cobalt-60 tracking scenarios are presented and further applications of this method are discussed.
Bibliography:
Includes bibliographical references.
Statement of Responsibility:
by Allan Martin.

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
036254238 ( ALEPH )
1043832375 ( OCLC )
Classification:
LD1780 2018 ( lcc )

UFDC Membership

Aggregations:
University of Florida Institutional Repository

Downloads

This item is only available as the following downloads:


Full Text



PAGE 1

1 SPECTROSCOPIC ANALYSIS FOR A DATA FUSION BASED RADIATION DETECTION SYSTEM BY : ALLAN MARTIN ADVISOR : DR. ANDREAS ENQVIST A PROJECT IN LIEU OF THESIS PRESENTED TO HERBERT WERTHEIM COLLEGE OF ENGINEERING OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER S OF NUCLEA R ENGINEERING UNIVERSITY OF FLORIDA 2018

PAGE 2

2 Abstract A d ata fusion based radiation detection system capable of detecting and tracking possible radiological threats is under development at the University of Florida. As an expansion to the capabilities of this system an algorithm was expanded to focus on particular isotopes of interest in the spectra This was accomplished by only analyzing radiation counts from under the photopea ks of the gamma ray spectra during the tracking as determined by user defined region s of interest (ROI) This wou ld allow the system to analyze specific section s of the radiation detect or data stream, with the goal of increasing the speed and accuracy with which a source is tracked and also let it be identified Results from c esium 137 and c obalt 60 tracking scenarios are presented and further applications of this method are discussed.

PAGE 3

3 Table of Contents i Abstract ................................ ................................ ................................ ................................ ........ 2 1. Introduction ................................ ................................ ................................ ................................ 4 2. Methods ................................ ................................ ................................ ................................ ....... 5 2.1 Source Movement Set up ................................ ................................ .............................. 5 2.2 Radiation Detector System Set up ................................ ................................ ................ 5 2.3 Energy Calibration for Regions of Interest ................................ ................................ ... 6 2.4 Data Acquisition ................................ ................................ ................................ ........... 6 2.5 Algorithm for Pulse Processing ................................ ................................ .................... 6 2.5.1 Net Counts ................................ ................................ ................................ ..... 6 2.5.2 Distance Prediction ................................ ................................ ........................ 7 3. Results and Discussion ................................ ................................ ................................ ............... 8 3.2 Energy Calibrated Regions of Interest ................................ ................................ .......... 8 3.2 Pulse Indexing by Regions of Interest ................................ ................................ .......... 8 3.3 Linear Distance Prediction ................................ ................................ .......................... 10 4. Conclusion and Future Work ................................ ................................ ................................ .... 14 5. References ................................ ................................ ................................ ................................ 1 5

PAGE 4

4 1. Introduction Radioactive materials can be found in applications ranging from civilian uses in medicine or agriculture, to government uses in nuc lear warheads or fuel in power plants. The security of this material is paramount to public safety, especially at facilities where special nuclear material (SNM) is used. SNM is of the utmost security importance as it can be used in nuclear weapons [1] but a ny modest amount of radioactive material could be used ill icitly in a radiation dispersal device (RDD) to inflict great harm and sow panic among the global population. Public spaces like airports, commerce hubs like ports or mailing facilities, or governme nt sites handling SNM all pose unique problems of their own for a radiation monitoring system but a few universal issues exists. Namely, the need to reasonably preserve the flow of people and goods respect situations like custo ms inspections) and provide a timely and accurate triggering of an alarm Current systems like portal monitors and distributed sensor networks address the main issue s of triggering an alarm and preserv ing the flow of people or goods as seen by their wide implementation, particularly portal monitors [2][3] These systems have limited capabilities bey e d. However, a system being developed at the University of Florida [4] has shown it has the ability to not only detect the presence of radioactive material, but also to track it This added capability creates fewer disturbances for commerce and the public by enabling a more accurate and targeted response to the possible threat This system is based on the concept of data fusion, where data streams from multiple radiation detectors and vision sensors are combined to leverage the unique information available to each sensor modality. The correlation between a distance data stream and a radiation count rate data stream comes from the concept of the radiation count rate being inversely dependent on the square of the distance to the detector Tracking scenarios and rough examples of the data from each sensor are shown in Figure 1. Th e project discussed in this paper sought to expand on the use of the inverse dep endence of the radiation count rate to the square of the distance by adding in an energy component to aid in tracking individual radiation signals in multisource environments or by limiting alarms to certain radionuclides. Many radionuclides emit gamma rays of specific energies as they decay which can be used to identify them. Two gamma ray emitting nuclides of interest to national security and the public health are cesium 137 and c obalt 60 which were used as the primary radiation sources for this project Their characteristic energies are 662 keV (Cs 137 ), 1173 keV (Co 60) and 1333 keV (Co 60) gamma rays The sodium i odide detector chosen for this project is widely used by other systems in this field Figure 1 Tracking scenarios and the principle of the overlapping distance/time domain between the radiological and visual sensor

PAGE 5

5 The goal of this project was to demonstrate the ability to divide a radiation data stream from a moving multisource environment into separate radiation count vectors containing only pulses with energies unique to in this case around Cs 137 gamma ray and around Co 60 rays This must be done while preserving timing information for each pulse in our detector The radiation count data would then be used in an algorithm to calculate a predicted distance from the detector to each source as a function of time. 2. Methods 2.1 Source Movement Set up The data fusion tracking system being developed required human subjects to carry radia tion sources in backpacks and walk throughout the scene being monitored. To investigate the use of only photopeak counts in this project, another method of supplying source movements on a smaller scale was developed. Normal movement is often confined to the x y plane but this project moved the sources in only the x z plane where motion was a combination of upward/downward and forward/backward movement for ease of analysis. An electric motor supplied the movement while a wooden meter stick allowed multiple sources to be attached to it and moved by one motor The motor was a 12 Volt direct c urrent electric motor from Uxcell with a period of 12 revolutions per minute It was powered by a n AC to DC adapter connected by alligator clips to the positive and negative terminals of the motor. The wooden meter stick had a hole drilled through the 50 cm mark so a bolt could be put through it, which was th en threaded into a coupling attached to the rotating shaft of the motor. Then the motor was attached to a mounting plate, which was screwed into a polyethylene block and taped down fo r stability as shown in Figure 2 Small gamma check sources were used to supply the radiation counts in this project and were affixed to the meter stick as shown in Figure 2. 2.2 Radiation Detector System Set up The main sources used in this project cesium 137 source calib rated on June obalt 60 source calibrated on January 2014. The detector chosen was a 2 in ch by 2 inch thallium a ctivated sodium i odide d etector which is sensitive to gamma rays and is Figure 2. (a) Motor and Meter Stick Set up (b) Source Mounting on Me ter Stick. (a) (b )

PAGE 6

6 commonly used in gamma spectroscopy and coupled with an Ortec 226 photomultiplier tube The de Struck SIS3316 DT 16 channel digitizer with a bank size of 500 pulses and a window size of 300 points per pulse A CAEN GECO2020 high voltage power supply was used to power the detector at 1330 volts which was selected to calibrate the detector response for a Cs 137 662 keV photopeak to approximately 0.3 volts. 2.3 Energy Calibration for Regions of Interest Before data with the motor spinning was be taken, the detector was calibrated by measuring the Cs 137 and Co 60 sources together and the n plotting the resulting spectrum of counts vs output voltage. The range of voltages encompassing each full photopeak were used to define the Region (s) of Interest (ROI) corresponding ( s ) and will be discussed further in Section s 2.5.1 and 3.1 This step could have been skipped if the detector had already been calibrated. 2.4 D ata A cquisition A series of measurements were taken with no source, just a Cs 137 source, just a Co 60 source, and both a Cs 137 and Co 60 source affixed to opposite ends of the meter stick. The higher activity source was always positioned closest to the detector w hen the measurement was started at time correspo nding to zero, the stronger source was 10 cm from the front face of the detector as shown in Figure 3. The motor was plugged in and allowed to rotate around once so that the data acquisition program could be started as close to the time when the stronger source was at the 10 cm distance from the detector The data acquisition program record ed pu lses for ap proximately 10 minutes for each measurement including background. 2.5 Algorithm for Pulse Processing 2.5.1 Net Counts A significant portion of the project time was spent developing an algorithm in MATLAB to first read the single file containing gamma ray total puls e integral data from each measurement The pulse integral data points were in the form of output voltages from the detector, proportional to the incident energies of the gamma rays. A file containing timing data corresponding to each pulse was also read into MATLAB. The algorithm operated as a set of two nested for loops which iterated through the pulse integral file and outputted to a single matrix The total measurement time was divided up into a time vector according to a user defined time step The first 2 values in the time vector would read 0 and Figure 3 Source Position in Relation to Detector.

PAGE 7

7 0.5 if the time step was 0.5 second The outer loop created a logical index that went through the pulse int egral file and determined whether a pulse was recorded during a particular time interval if yes then it added one to the number of pulses, or counts, in th at ROI within that particular time interval If that pulse was not within the first ROI, the algorithm would keep checking each successive ROI defined by the user until it found a match or discarded it if none matched It would continue this iteration for each time interval until the end of the time vector was reached. Each column in the matrix was dedicated to a specific ROI and each row to the sum of pulses that were recorded in the corresponding time interval. For example a time vector of dimensions 600x1 would with a matrix of dimensions 600x2 for two ROIs The midpoint time between two time steps was also saved into a vector for plotting as the x coordinate when using the net counts or predicted distance because those quantities did not occur instantaneously at that time step but rather over a time interval centered on the midpoint time. The background measurement data file was processed immediately after the data from whatever source configuration was used. It was processed in a similar fashion to the source data in that it was indexed into a matrix according to a time vector from its measurement time and the same ROIs however it differed in that it calculated the average number of counts per time interval for each ROI over the entire measurement time. These average background counts per time interval were then subtracted from every sum of counts value for the same ROI, givi ng the net counts per time interval for each ROI. If the net counts were negative, they were reassigned a value of zero. 2.5.2 Distance Prediction The algorithm was designed to not only separate the detector output by energy, thus identifying isotopes/gamma rays of interest, but also to aid in the tracking of those sources. It accomplished this by predicting the linear distance from a source to the detector at a given time step. Because radiation count rates theoretically vary according to the inverse squared law counts=1/distance 2 this was used as the basis for determin ing a distance ratio to re late how far away from the detector the source was compared to other data points The distance ratio for each net count value was calculated using Equation 1 Equation 1 is the inverse squared law solved for distance and a ratio of the net counts (N) at that time step divided by the maximum net counts (max(N)) If the net counts were less than or equal to zero, the distance ratio was reassigned the same value as the largest d istance ratio for that ROI. The largest value of the distance ratio should correspond to the furthest possible distance a source could be and fewest counts thereby this value was assigned a liner distance of 1 10 cm, and the smallest value of the distance ratio (should be 1) was assigned a linea r distance of 10 cm. Then, a line ar regression was performed using the two maximum and minimum distance ratios for each source to fit to The linear distance equations fit to eac h ROI change d wheneve r the time step was changed as the equation depend on the distance ratio for its components. Plugging the distance ratios into the linear distance equation gave a matrix of linear Equation 1

PAGE 8

8 predicted distances which was plotted to compare against an equation for the exact linear distance of a rotating object over time or net counts to verify 3. Results and Discussion 3.1 Energy Calibrated Regions of Interest The first step of operating any spectroscopic capable radiation detection system is calibrating the detector so that the energy of the output pulses can be determined. The plot of counts (background subtracted) over output voltage shown in Figure 4 was generated and then used to calibrate the NaI (Tl) detector. The 2 sources measured were Cs 137 and Co 60 which are associated with a 662 keV gamma ray from Cs 137 (Ba 137m) and 1173 and 1333 keV gamma rays from Co 60 Those 3 gamma ray photo peaks were identified in Figure 4 and the output voltages of pulses contributing to the photopeak s were used as the range of the Region s of Interest (ROI) The ROI chosen by hand for Cs 137 encompasses voltages from 0.26 volts to 0.33 volts and is centered at 0.29 which contains the maximum counts. The Co 60 ROI includes outputs from 0.42 volts to 0.48 volts which encompass both gamma rays. The chosen ROI for Co 60 includes both photopeaks because the Cs 137 source at the time of the measurement was roughly 14.6 times higher than the Co 60 source and thus the Cs 137 had roughly 14.6 times as many counts in its photopeak as did the Co 60 photopeaks. If these measurements were to be repeated, sources with more comparable activities would have been o btained. This ROI selection method is viable for any number of gamma sources once the detector system has been calibrated 3.2 Pulse Indexing by Region s of Interest Once the system was calibrated and all measurements were completed, the analysis started. Fig ures 5 and 6 compare the net counts per time interval when the time step was changed from 1 second to a 0.5 second. Both sources were attached to opposite ends of the meter stick for the measurement shown so the peak of one source corresponds to the valle y of the other source. Figure 4 Cs 137 and Co 60 ROI from Energy Spectrum

PAGE 9

9 The osci llation s of the counts show a period of roughly 5 seconds which is consistent with the physical rotation of the 12 RPM motor A smaller time step appears to create sharper, more defined peaks in the Co 60 counts per 0.5 second plot, showing more of the features. If the motor had a faster speed, then a shorter time step would be necessary to ensure that there were enough points to show the oscillation properly. These plots show that the algorithm correctly indexed the counts by ROI and time. This was confirmed by determining an equation (2a and 2b) to describe the exact linear distance of the source in terms of the rotati on of the meter stick and motor period, and plotting it against the net counts in Figure 7 The motor period was 5 second s and the length of the meter stick was 100 cm. Using the length as the amplitude and assuming steady rotation, the following equation s were derived: Equation s 2a, 2b. Figure 6. Multisource Net Counts per 0.5 Second Figure 5. Multisource Net Counts per 1 Second

PAGE 10

10 The slight offset or misalignment between the peak s of the net counts and the local minima of the exact distance is due to the time component of each value The exact distance is computed for an instantaneous moment in time whereas the net counts are compiled over a time interval as wide as the time step. Were an extremely short time step used the features might line up better however the trade off would come from fewer counts in each time step, thus increasing the measurement uncertainty These plots show t hat it was possible to track the movement of a source in terms of towards or away from a detector by comparing net counts generated by this method and a single detector 3.3 Linear Distance Prediction The original tracking system developed by our research group is equipped to utilize multiple different detectors such as NaI and in different locations to create a 3 dimensional web of overlapping distance predictions. T aken together these offer a predicted location and motion track over time. The original system has only just started measuring multiple sources and the method presented in Section 2 .5.2 may offer a more novel solution. The linear fit s to the Cs and Co source s maximum and minimum distance ratios were plotted vs. elapsed time with 1 second and 0.5 second time steps. Figures 8 and 9 illustrate the notable skewness recorded in the net c ou nt data for Cs 137 Predicting a linear distance from the Cs 137 data proved difficult due to the wide range of net count values and how many low net count values there are This was especially skewed at low time steps (Figure 9 ) Figure 7 Co 60 Net Counts and Exact Linear Distance from Detector [cm] per 0.5 Second

PAGE 11

11 E xample s of the Cs 137 distance prediction s compared to the Co 60 prediction s at a 1 second and a 0.5 second time step are shown in Figure s 10 and 11 The peak predicted distances for Cs 137 in Figure 1 0 which should have been around 110 cm, did not often exceed 45 cm; and in Figure 10 the peak distances did not often exceed 80 cm. Whereas the Co 60 predicted dist ances usually followed the range of exact distance values from 0 to 110 cm An exception to this was caused by net counts for Co 60 over the first 140 seconds that stayed strangely high and caused the distance prediction to under predict for that entire time range thereby increasing the values of the mean and median percent error s and mean sq uared errors, all of w hich were consistently greater than Cs 137 as shown later in Table 1 Figure 9. Histogram of Cs 137 Net Counts per 0.5 Second Figure 8. Histogram of Cs 137 Net Counts per 1 Second

PAGE 12

12 Figure 10 Cs 137 and Co 60 Predicted D i stance Comparison, per 0.5 Second Figure 11. C s 137 Predicted and Exact Distance from Detector, per 1 Second

PAGE 13

13 Due to the Co distance more accurately than Cs 137 when judged v isually in plots like Figures 11 and 12 The shape of the Cs 137 predicted distance curve appeared closer to the shape of the exact curve th an Co 60 but did not cover the full range of possible source distances like Co 60 did In cases of a weak source, net counts of zero correspond ed to the fu rthest allowable distance of the device, in this set up, 110 cm. The short 0.5 second time interval also led to too many bins having zero net counts (negative before the algorithm adjusted them). This is the reason why the predicted distance Co 60 distance in Figure 10 seems to flatten out with most predictions showing 110 cm. S tatistical comparison of the predicted distances at varying time step s to the exact distances yielded the results shown in Table 1. A visual comparison did not support the claim of a poor fit as evidenced by high percent error results This appeared to be true when viewing the plots in small intervals unless the whole elapsed time was seen As mentioned before, the Co 60 predicted dista nce appears to fit the exact very well if all but the first 140 seconds are viewed. The Co 60 measurements were also impacted by low overall net counts giving poor counting statistics. The Cs 137 error values for the 0.5 second time step do not appear to follow the same pattern as the other time steps but a reason for that is not clear at this time Another factor leading to deviations from the exact distance is the stochastic nature of radioactive decay T hat adds a degree of randomness to the number of counts per time step that causes them to not always follow the expected relationships like the inverse squared law. Figur e 12. Co 60 Predicted and Exact Distance from Detector, per 1 Second Table 1. Statistical Analysis

PAGE 14

14 One problem that this method did not address yet was how to code the algorithm so that count data distributed like Figures 8 and 9 would not drastically skew the distance ratio and linear distance prediction. Despite those problems, the mean of the absolute mean predicted distance difference values was right around t he 20 cm average error that the original tracking algorithm pre sented. Analyzing the data using the shorter time step of 0.5 seconds gave a more accurate prediction in terms of mean difference using the Co 60 ROI, but the longer 1 second time step was more effective at predicting the distance of Cs 137 when the net counts values were less skewed than over 0.5 seconds Other values of the time step like 0.25 seconds could be investigated further to optimize this factor in the future. 4. Conclusion and Future Work The proposed method and algorithm for indexing gamma ray data from a spectroscopic capable radiation detector by time and output voltage was investigated with a series of small scale experiments These experiments were designed to replicate radiation source movements in an area monitored by a data fusion based radiation detection and tracking system. The ability for the user to input specific voltage regions (proportional to incident gamma ray energies) that they want to monitor allows this method to be applied to multisource environments wher e certain radionuclides are of greater importance than others. The successful separation of counts from a single detector into energy specific regions of interest, ROIs, was clearly shown in plots comparing counts over time from the photopeak of Cs Ba ray and the combined photopeaks of Co ray emissions. A high percentage of time intervals with low Cs 137 net counts skewed the distance ratio part of the algorithm which resulted in under prediction of the distance and large percent errors from the exact distance. Calculated R Square and adjusted R Square predicted distance curve with the exact distance were less than negative one leading us to conclude that a nother prediction algorithm or method of fitting to the data would fit the exact distance more closely Some of this discrepancy could be from the fac t that the exact distance is a calculated quantity assuming constant motor speed at 12 RPM while the predi cted distance has u ncertainty from radioactive decay, actual motor rotation speed, and attenuation in air. Overall, the algorithm captured the distance trend as displayed by the similar shapes and close visual agreement of the curves. This method and algorithm were shown to have promise in a data fusion radiation tracking system and should be investigated more to achieve better prediction statistics. This method would be useful in scenarios where there are some radionuclides of greater importance than others, f or example medical radiation treatments leave patients with a lar ge radiation signature that would not be indicative of a threat to national security so the system would not need to include those radionuclides or raise the alarm threshold. Future work coul d investigate shielding situations that would potentially disguise the characteristic energy emissions and lower counts.

PAGE 15

15 5. References [1] B.D. Geelhood, J.H. Ely, R.R. Hansen, R.T. Kouzes, J.E. Schweppe, and R.A. Warner, [2] R. C. Runkle, T. M. Mercier, K. K. Anderson, and D. K. Carlson, "Point Source Detection and Characterization for V 6, pp. 3020 3025, 2005. IEEE Computer Magazine, vol. 37, no. 8, pp. 57 59, 2004. [4] K Sta Data Fusion for a Vision Aided Radiological Detection System: Sensor dependence and Source Tracking Proceeding s of Advancements in Nuclear Instrumentation Measurement Methods and their Applications (ANIMMA), 5 th Edition Meeting, June 2017.