Research On The Neutron Spectrum Unfolding Method Based On Trust Region Algorithm

It is necessary to know the accurate neutron spectrum in many fields, such as nuclear physics research, reactor techniques, accelerator operation, space exploration, radiological protection and so on. Unfolding neutron spectrum with response functions and pulse height distributions is a key technique in the filed of neutron detection, and it is of great value to both scientific research and practical application. In this dissertation, we carried out the related research using liquid scintillation detector, and proposed a novel unfolding method based on Trust Region (TR) algorithm.In this paper, the basic principle of neutron spectrum unfolding with liquid scintillation detector was introduced at first. Then the ill-condition property of the response matrix generated by 05S Monte Carlo program was checked up. Meanwhile, the error analysis of the ill-posed unfolding problem was done under two conditions: one situation for only the pulse height distributions contaminated and the other for both the response functions and the pulse height distributions contaminated, which was used as the theoretical basis of the simulated validation and the experimental validation in the below. Singular Value Decomposition (SVD) method was used to explain the primary cause of the ill-poseness of the response matrix, and then the existing unfolding methods which were effective for the ill-posed unfolding problem were illustrated.To alleviate the serious ill-poseness of the response matrix, several preconditioners were tested, including Jacobi, row (column) balancing, Gauss-Seidel, Successive Over Relaxation (SOR), Symmetric Successive Over Relaxation (SSOR). By comparing the condition number of the processed matrix of each preconditioner and the results of numerical tests, the row balancing was chosen as the best preconditioner to combine with the Trust Region algorithm. Then Least Squares (LS) method, Truncated Singular Value Decomposition (TSVD) method, Tikhonov regularization method and Trust Region algorithm were utilized for the simulated validation. The single-peak, multi-peak and 1/E peak simulated neutron spectrum under different Signal Noise Ratio (SNR), containing∞,140dB, 100dB,60dB, were unfolded during the validation process, and the results showed that Trust Region algorithm owned a higher accuracy and a better robustness.In the experimental validation, a hardware platform including several neutron sources, liquid scintillation detectors, photomultiplier tubes, electronics circuits and computers was build, and a hardware circuit for measurement was designed with ORTEC standard nuclear electronic modules. Then the D-T monoenergetic neutron source, the 241Am-Be radioisotope neutron source and the 252Cf spontaneous fission neutron source were measured in the experiment. The recoil proton distributions were obtained after n/y discrimination, background subtraction, energy calibration and the change from recoil electron distributions to recoil proton distributions. After that, the three neutron sources were unfolded by Trust Region algorithm and Tikhonov regularization method, and the efficiency of the two methods was proved compared with the reference spectrum. What is more, Trust Region algorithm could get more accurate results and therefore the effectiveness of this method was demonstrated.