Methods For Processing And Identifying Gamma Radiation Signals Of Nuclides In Nuclear Power Plants

Nuclear power plants contain a large amount of complex radioactive isotopes in their reactors,which can cause serious environmental pollution and even accidents if they leak,affecting the safe operation of the nuclear power plant.By continuously measuring the gamma rays emitted by all radioactive isotopes in the cooling water and environmental samples of the nuclear power plant in real-time,the nuclide gamma energy spectrum can be obtained and the types of nuclides contained in the sample can be analyzed.It is possible to determine whether a leak has occurred in the nuclear power plant environment based on the identification results,providing operational support for the plant personnel.This thesis focuses on the rapid and accurate identification of key radioactive nuclide signals that need to be monitored in the operation of a nuclear power plant under environmental radiation monitoring.Starting from the electrical signals collected by the scintillation detector,the nuclide gamma radiation signal processing and identification algorithm is designed.After corresponding experimental tests,certain research results have been obtained.At the same time,in order to realize the edge deployment of the identification algorithm,the quantization method of the relevant neural network has been studied,and the real-time high-precision nuclide gamma signal identification has been implemented in FPGA.The research results of the thesis mainly include the following aspects:(1)A set of algorithms for collecting nuclear gamma energy spectra is developed and a spectral acquisition platform is built.The baseline correction algorithm designed in the thesis achieves good correction results by introducing prior signals for calibration.The peak detection algorithm achieves superior peak detection results through multistep processing and dual-threshold detection.The spectrum calibration algorithm achieves a correction result that satisfied a PE allowable error of ±0.25B/e,and the correction result is perfect.Experimental results in complex radiation monitoring environments demonstrates that the nuclear gamma energy spectrum acquisition algorithm designed in the study achieves a high spectral resolution,significant formation of nuclear energy peaks,and clear detection of weak element peaks,indicating good precision and robustness.(2)A multi-label wavelet convolutional neural network-based method for multinuclide recognition is designed,integrating wavelet transform and convolutional neural network to effectively extract spectrum features at multiple resolutions and significantly improve the accuracy of nuclide recognition.In the test of measured spectrum signal dataset,the subset average accuracy of the proposed method reaches97.8%,which is about 10% higher than the Vgg-net and Res-net methods,and about 2%higher than the CNN method.The comparison results of different network models show that the recognition method proposed in the thesis has achieved a good performance in multi-nuclide recognition.(3)A multi-nuclide identification network based on convolutional neural network is designed based on Kria KV260 development board,and the network is quickly deployed in FPGA through Vitis AI integrated development environment.The test results show that the accuracy of the quantized model is 91.7%,and the speed is about27% faster than that of the CPU under the same test sample capacity,indicating that deploying the multi-nuclide identification network using FPGA in edge devices is feasible.