Speaker
Description
High-Z materials are used as plasma facing components due to excellent material properties in fusion devices. However, impurities are inevitably induced in plasma discharge by plasma-wall interactions. The presence of these impurities can lead to increased energy loss and degradation of plasma confinement. Therefore, the study of impurity behavior based on impurity spectral diagnosis is of crucial importance in fusion research. At present, the newly developed four fast-time-response Extreme Ultraviolet (EUV) Spectrometers on EAST have been operating [1-3]. Capability of accurate wavelength measurement is requirement for line identification of EUV spectra from high-Z impurity especially tungsten. In this work, an integrated impurity data analysis and wavelength calibration method is developed based on the complete impurity spectra database established on EAST [4, 5]. This approach incorporates algorithms for direct peak detection, Gaussian peak fitting, and Lorentzian peak fitting, replacing traditional manual methods to enhance efficiency. Integration of sub-pixel interpolation has improved peak detection accuracy. In addition, the methodology uses a model trained from the impurity spectral database with linear regression to predict the peaks corresponding to strong spectral lines, followed by wavelength calibration using polynomial approximation, polynomial fitting, and piecewise polynomial approximation methods for theoretical calculation values, cubic polynomial fitting values, and algorithm reference values. The results of the wavelength calibration are evaluated using Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and pixel point uncertainty, with cross-validation between the three types of wavelength values for iteration. As shown in Fig. 1, the absolute value of the wavelength uncertainty is within 0.03 Å. After determining the wavelength calibration results based on strong spectral lines, the spectral drift correction for weaker spectral lines is performed using the Particle Swarm Optimization (PSO) algorithm, and the aforementioned evaluation and iteration process is repeated. This work has not only significantly enhanced the efficiency and accuracy of data analysis but also markedly reduced reliance on manual operations, offering a practical and viable solution for automating and enhancing the intelligence of data processing.
