Study On Rapid Identification Of Radionuclides Based On Sequential Bayesian Analysis

Radioactive materials play a very important role in national developments ranging from algriculture to industry. With the development of nuclear power and the deterioration in nuclear terrestism threats, it is indispensable to supervise and control the raodiactive materials effectively in order to prevent the proliferation and smuggling of radioactive materials, especially Sepcial Nuclear Materials (SNM). However, rapid detection and identification of radioactive materials is claimed to be one of the most difficult problems. Generally, radioactive materials can be exclusively identified by passively detecting and analyzing their characteristic gamma rays. In tranditional methods, energy spectrum analysis method, statistically analyzing the gamma ray energy distribution based on Gaussian hypothesis, has been well developed and established and this method incorporating with HPGe detector can identify radioactive material and precisely calculate its radioactivity. Due to the long counting time resulted from the measurement of a large number of gamma ray events for low uncertainty of results, the energy spectrum analysis method is not suitable to be applied in security checking circumstance including custom, seaport, airport and checking points on the borders. Typically, these radioactive materials waiting for inspection are deposited in shielding packages or cargo containers so that their characteristic gamma rays are easily attenuated and distorted.In2009, James Candy et al from the Lawrance Livermore National Laboratory first proposed a new method based on sequential Bayesian analysis to solve the difficulty mentioned above. In this method, three "fingerprint" features of radionclide, e.g. the decay half life, the characteristic gamma ray’ energy and emission probability, are jointly input into a model-based signal processor using the Bayes theorem and the sequential probability ratio tests theory to make a decision that whether the targeted radionuclide exists or not. According to the pulished literatures, the current limitations and deficiencies in this research field were comprehensively analyzed based on which three studies in this dissertation, e.g. numerical simulations, off-line experiments and on-line experiments, were systematically developed and a prototype of sequential Bayesian analysis system incorporating with a LaBr3(Ce) sintillator detector was also proposed to verify the feasibility of the sequential Bayesian analysis approach and to illustrate the detection performance (detection probability, mean dedection time and false alarm probability) of this method varying as function of radionuclide’s equivalent activity.With regard to study on the performance and energy calibration of the detector used, the energy resolution, detection efficiency and working stability of the KLB5075-type LaBr3(Ce) sintillator detector were tested, and based on the obtained parameters the detector’s response function of monoenergetic gamma rays in the0.1-2MeV energy range were calculated using MCNP5. Besides, the internal radioactivity in the sintillator derived from138La and227Ac was investigated. Thus, a self-calibration method was proposed using this internal radioactivity and its feasibility and effectiveness was furtherly clarified according to the proof-of-concept experimental results.With regard to the numerical simulations, the basic problem of interest was strictly defined and the procedure of signal processing was devided into seven functional modules, e.g. initialization, detector, data acquisition, event discrimination, parameter estimation, decision function update and existence judgement. A universal physical-based detection model, which contains an adjustable parameter of model detection rate, was constructed to simplify the source-shield-detector configuration in practical application. Based on the detection model, the sequential Bayesian analysis processor was developed with the details of algorithms in each functional module were minutely demonstrated using the Bayes theorem and the sequential probability ratio tests theory. Meanwhile, a radiation event sequence generator, which contains an adjustable parameter of tested detection rate resembling the model detection rate in processor, was developed using Monte Carlo sampling theory to simulate various radiation fields. Both the processor and the generator mentioned above were realized using Matlab7.0. The results of numerical simulation show that the processor’s response time for making a decision on137Cs and60Co is remarkably shorter (≤1s), preliminarily validating the feasibility of this method for rapid identification on radionuclides. The detection performance depends on the interval of the tested detection rate correlating to the model detection rate and at same time is influenced by Compton scattering effect from high-energy sources and cosmical gamma rays.With regard to the off-line experiments, for detection of the radionuclide in real world scenarios, the algorithm for the sequential Bayesian analysis was fully constructed and optimized so that it is adaptable to the background radiation. Because of the limited experimental devices and instruments that are available at the present stage, the off-line experiment method is introduced, in which event sequences can be generated using Monte Carlo sampling theory according to both the measured pulse-height spectrum under various conditions and the exponential distribution of event arriving interval. The off-line experiment method is proved to be both effective and statistically correct. The processor performance for the detection of137Cs,60Co,133Ba and152Eu as function of source placement distance was studied, and the Maximum Detection Distance (MDD) and the Background False-alarm Level Distance (BFLD) are introduced to quantitively evaluate the processor performance. The off-line experimental results show that the mean detection time of each radionuclide was only a few seconds, indirectly validating the feasibility of this method for rapid identification on radionuclides. The maximum detection distances of137Cs (7.676×103Bq),137Cs (1.679×104Bq),60Co (1.365×103Bq),152Eu (8.962×103Bq) and133Ba (7.147×104Bq) are approximately55cm,70cm,30cm,25cm, and157cm, respectively.With regard to the on-line experiments, for on-line experiments, a sequential Bayesian analysis system, consisting of a LaBr3(Ce) scintillator detector, a FPGA-based pulse analyzer, and control&analysis software, was developed with the energy resolution of3.2%at661keV, the time resolution of10"7s and the counting rate of at least2kHz. For more compatibility, expansibility and effectiveness, the software for data acquisition control and the sequential Bayesian analysis processor were programmed using Labwindows CVI8.5. The basic principles and properties of the FPGA-based pulse analyzer were demonstrated, and its performance was tested with a sampling rate of109/s and an analog-digital conversion precision of12bits. The results show that the pulse analyzer can directly or simultaneously analyze the voltage signal from the detector to obtain the energy and arrival time of events. In the on-line experiments, the system performance for the detection of137Cs,60Co,133Ba and152Eu as function of measurement distance was studied, and other than MDD and BFLD, the Equivalent Minmum Detectable Activity (EMDA) is also proposed to quantitively classify the system performance. The on-line experimental results agree well with the results from both off-line experiments and numerical simulations, which furtherly and directly validate the feasibility of the sequential Bayesian analysis approach for rapid identification on radionuclides. Generally, in regard to each of the target radionuclides, the detection probability decreases and the mean detection time increases as the measurement distance increases. Meanwhile, the false-alarm probability of the absent radionuclides decreases to the background level as the detection distance increases. The maximum detection distances of137Cs (1.588×105Bq),60Co (1.862×104Bq),133Ba (6.874×104Bq), and152Eu (9.723×104Bq) are approximately123cm,80cm,102cm, and80cm with the corresponding equivalent minmum detection activity of16.4Bq,4.54Bq,10.32Bq, and23.72Bq, respectively.In summary, the sequential Bayesian analysis approach for rapid detection and identification of radionuclides was systematically studied, and its feasibility was completely verified according to the numerical simulations, off-line experiments and on-line experiments. The physical ideas and mathematical algorithms within this method as well as the detection performance as function of measurement distance (equivalent radioactivity of radionuclide) were minutely illustrated. A prototype of the sequential Bayesian analysis system is developed, paving the way for pratical applications in real scenarios and providing technical backup for prevention of radioactive material proliferation and smuggling.