| With the continuous development of phoswich detector technology,related research on particle discrimination is also deepening.In complex irradiation environments such as deep space,nuclear reactors,and high-energy research,identifying particle types not only achieves corresponding research goals but is also crucial for personnel health protection design.In these application scenarios and in monitoring atmospheric radioactive xenon isotopes,the design of β-γ particle discrimination is an important part.Since its inception,the phoswich detector has undergone decades of development,and many studies have used it in particle discrimination design.Research on particle discrimination phoswich detectors is developing towards structural optimization and algorithm optimization,constantly improving the efficiency and accuracy of particle discrimination.In existing research,the identification ability of the detector needs to be improved and further design and optimization of the algorithm are required.With the deepening of research on machine learning methods,some related application work has also emerged in this field.This paper carries out the development of a β-γ particle discrimination phoswich detector based on pulse shape discrimination technology.The structure of the detector is designed,and a pulse discrimination algorithm is constructed based on the pulse characteristics of the detector.The calculation results of the algorithm are discussed.To further improve the pulse classification effect,machine learning methods for particle discrimination are studied.The main work of this paper includes:(1)The development of a β-γ particle discrimination phoswich detector was carried out.This part first summarizes the issues that need to be considered during the development process and proposes a structural framework for the phoswich detector.Through the analysis of relevant performance parameters of scintillators,the combination of EJ-260/Cs I(Tl)scintillators was selected as the design scheme for the phoswich detector.Then,the GEANT4 program was used to construct a model of the detector for structural design.The energy deposition of 3.5 Me V electrons injected into EJ-260 was simulated to determine that the thickness of this scintillator is 20 mm.On this basis,a Cs I(Tl)crystal with a thickness of 50 mm was selected as the rear-end scintillator.The pulses formed by 1 Me V β-rays injected into the EJ-260 scintillator and1 Me V γ-rays injected into the Cs I(Tl)scintillator were simulated to verify that the detector can be distinguished by pulse shape discrimination algorithm.On this basis,the phoswich detector was assembled and tested using actual β-rays and γ-rays,and the corresponding average pulses were calculated.The results show that the actual pulses also have obvious differences in rise time and can be processed by rise time discrimination methods to achieve pulse classification effects.(2)Particle discrimination of the phoswich detector was achieved based on pulse shape discrimination algorithm.This part first introduces the pulse discrimination algorithm and selects the rise time discrimination algorithm as the scheme for implementing particle discrimination.After smoothing the pulse with a moving average filter,an algorithm for extracting the rise time parameters of each pulse was written using MATLAB.The pulse rise time data was analyzed to achieve pulse identification and ray spectrum drawing.The results show that the detector can distinguish pulses from different scintillators and draw corresponding particle spectra.Finally,the figure of merit value of the detector was calculated to be 6.35 and compared with the results in other literature.The results show that the phoswich detector structure built in this work can achieve particle discrimination effect and has a high figure of merit value.(3)In order to achieve better particle discrimination effect,this paper explores the application of machine learning methods in particle discrimination.This part summarizes the development history and principles of machine learning methods and selects Gaussian mixture model,support vector machine and convolutional neural network for data classification machine learning methods for testing.Three types of models were constructed in MATLAB program,and the pulses of EJ-260/BGO phoswich detector measured in the experiment and the training set data constructed from pulses collected separately by two scintillators were used.The GEANT4 program was used to simulate the pulses to demonstrate the reliability of the dataset.The training results show that all three models can discriminate detector pulses to a certain extent.Among them,the accuracy of data classification by Gaussian mixture model,support vector machine and convolutional neural network are 72.9%,85.5% and 96.1%,respectively.Limited by the type of dataset and the characteristics of the model itself,both Gaussian mixture model and support vector machine have certain defects in classification effect,while convolutional neural network has the highest classification accuracy.The innovative points of the paper are as follows:(1)Through the analysis of ray characteristics and scintillator performance,a phoswich detector structure with EJ-260/Cs I(Tl)combination was established.The geometric structure of the scintillator was optimized by using the GEANT4 model,and the distinguishability of its pulse output was verified.On this basis,the detector was assembled and preliminary tested.(2)The rise time discrimination method was used to process data for pulses of phoswich detector to achieve pulse type identification and spectrum drawing.(3)The application of machine learning in particle discrimination was studied.Based on the experimental data of EJ-260/BGO phoswich detector,three different machine learning models were constructed for testing. |
