Study Of A Method For Monitoring Surface Deposition Of First Mirror Materials Based On Spectral Data

In tokamak fusion devices,optical diagnostics are one of the most important means of obtaining and monitoring plasma parameters.The harsh neutron environment and nuclear-related operation of future fusion reactor devices require that all optical diagnostic systems must use a large number of mirrors to form a labyrinth of optical paths,and the mirror at the front end of the device directly faces the plasma,which is called the first mirror.Due to the strong interaction between plasma and material of the device,impurity deposits will occur on the surface of the first mirror,which will reduce the optical performance of the first mirror,thereby affecting the accuracy and validity of the optical diagnosis signal,and even causing the optical diagnosis to fail to operate normally.Using the sputtering characteristics of radio frequency plasma on the material,the first mirror surface deposition layer can be actively removed and the reflectivity of the first mirror can be restored,and in a closed tokamak in situ,how to monitor the plasma cleaning process to ensure the effect of radio frequency plasma cleaning the impurity deposition layer on the surface of the first mirror,so as to avoid under-cleaning or over-cleaning,is essential for the reliable operation of optical diagnostic systems.In this paper,under laboratory conditions,the surface of the first mirror of Mo was cleaned by radio frequency plasma,and the plasma spectrum in the cleaning process was collected by spectrometer,and the plasma spectrum before and after cleaning was measured by spectrophotometer to analyze the characteristics of plasma spectral data during the cleaning process and the change of the reflectance of the first mirror before and after cleaning.On this basis,the one-dimensional convolutional neural network technology is used to establish the first mirror surface impurity deposition state recognition model,complete the code writing of the first mirror surface deposition recognition program,and accurately identify the presence and absence of Al₂O₃ impurity deposition on the surface of the first mirror of Mo,and the main research results of this paper are as follows:（1）Determine the spectral data acquisition method and system structure involved in the first mirror cleaning,build the first mirror surface impurity deposition cleaning monitoring system under laboratory conditions,and obtain the first mirror and impurity deposition spectral signal by spectrometer during the process of impurity deposition on the surface of the first mirror by radio frequency plasma sputtering cleaning,when the air pressure is 3 Pa and the self-bias pressure is-400 V,the characteristic peak of the first mirror of Mo at the wavelength of 313 nm is successfully obtained;When the air pressure is 2 Pa and the self-bias pressure is-300 V,the characteristic peak of Al₂O₃deposited impurities at a wavelength of 394 nm is obtained.（2）Based on the one-dimensional convolutional neural network,establish the first mirror surface impurity deposition state recognition model,based on the reflectivity data before and after the first mirror cleaning of Mo,establish the training dataset,and expand the reflectance dataset to meet the requirements of the convolutional neural network training data,consider the noise immunity performance of the model,superimpose Gaussian noise on the dataset,through the analysis of the optimizer and batch size in the model training process,determine that the model optimizer is Adam and the batch size is 16,and the first mirror surface deposition state is successfully identified.Moreover,the recognition accuracy on the noise dataset is 98.5%,which is higher than that of traditional recognition algorithms.（3）Based on the one-dimensional convolutional neural network first mirror surface impurity deposition recognition model,combined with the Flask framework,the relevant code of the first mirror surface deposition recognition program is written,and the state of the first mirror surface impurity deposition is identified in real time by entering the spectral reflectance data,and the input data is visualized and displayed,and finally real-time monitoring of the first mirror cleaning process is realized.