X-RAY FLUORESCENCE ANALYSIS BASED ON KNOWLEDGE SYSTEM

摘要

A system based on knowledge is proposed to analyze a sample without the online supervision of a human expert in x-ray fluorescence analysis. Evolving factor analysis, pattern recognition and neural networks are applied to the system. For Cr, Fe and Ni in alloys, the neural network model combined with the fundamental parameters decreased the prediction errors by 1.1 to 2.8 times as compared with the theoretical coefficients calculated by the classical Lachance model, the hyperbolic function and the COLA algorithm. In most cases, the predictability of the neural network algorithm is better than that of the fundamental parameter method.