Influence Character Of Elements In EDXRF And Research Of Artificial Neural Net Correction Technique

Energy dispersive X-ray fluorescence analysis technology is the most advanced one of non-destructive assays. It is one such technique as it offers non-destructive, fast and reliable analysis for a wide variety of material, and the application field of EDXRF is widely.X ray analysis method, in theory, equipment and application, has achieved a high level, and remarkable achievement has been obtained since the disvovery of X ray. From original photo-register method to subsequent X-ray spectrom analysis system; from original radioisotope sources to subsequent X-ray tube and synchrotron radiation accelerator; as well as the application of normal temperature semiconductor detecotor and improvement of computer technique, which have made a quick development of X-ray fluorescence analysis thechnique in the last 50 years. In 1950s, some national research institutions have developed in some respect of EDXRF and accumulated abundant of experience, however, more fundermental research and techniqueexploitation should be taken as compared with other countries in this field.Matrix effect is the key question to affect the accuracy of EDXRF analysis, esspecially the absorption-enhancement effect existed among elements, which can bring great interference to fluorescence count rate caused non-linerary relation between fluorescence count rate intensity and element content. So, to get the influence extent is very essencial to improve the analysis precision. Meanwhile, to get approprate correction method for matrix effect and to decrease the affect of matrix effect are the most important research work in X-ray analysis field to improve the analysis accuracy.According to the key questions above to be solved, the emphasis is placed on two respects of research work, one is the research of affect character among some kinds of elements analyzed by EDXRF technique, the other is the research of esteblishment of matrix effect correction model in XRF based on artificial neural net. All research work are based on the sufficiency study and research of international EDXRF analysis techninology, and also linked with the National Natural Science Fundation of China (NSFC No.40574059) : Reseach of Auto-Classified and Non-Linary Dynamic Model Based on EDXRF.1.Calculate of binary influence coefficientIt can be date back to the year 1966 when foreign science researchers proposed the concept and calculate method of influence coefficient among elements for how to get the inffluence degree. The method was named as Lachance-Trail,which combined the analyte element content, other elements content and mass attenuation coefficient to obtain the influence coefficient among different elements. The research paper adopts the right algorithm above to obtain influence coefficient of target elements. The research work is combined experiment and mathmatic method to analyze the influence character among Ti,V,Fe,Ni,Cu,Zn these six elements by using Lachance-Trail algorithm and get the value of binary influence coefficient a_ij,they are,The absolute value of a_ij can reflect the degree size of influence among elements, and it changes along with the change of element content, so a_ij can not be considered as a constant value when taking element content analysis. But while the content of element is changed in 10% range and 5% relative error in analysis result, the value of a_ij can be considered as a constant.2. Equation of element characteristic X-ray fluorescence normalized counting rate (R_k) and content (W_k) in samplesNine kinds of binary element samples were prepared in the experiment, Ti-V,Ti-Fe,V-Fe,Fe-Ni,Fe-Cu,Fe-Zn,Ni-Cu,Ni-Zn,Cu-Zn, as well as one ternary element sample of Fe/Ni/Zn. Each sample was measured by a EDXRF analyze system which employs electronic cooled Si(Li) semiconductor detector, and the X-ray fluorescence count rate of each element was obtained. The intensity of element characteristic X-ray was achieved by Gauss fitting method, and the curve of R_k-W_k is also showed in this paper. It is can be seen from the curve that the relation between R_k and W_k is not linear but shows a nonlinear of hyperbola shape. So, an exponent fitting method was presented in this research work. Exponent least-square fitting method was adapted to fit the R_k-W_k curve in each sample, the correlation coefficients are better than 0.999 which is shown a good fitting result. At the same time, Monte Carlo simulation is applied in the element influence character research work, and the result is very similar to the experiment. So, further application of Monte Carlo can be used to replace of experiment work, which can save much time and force amount.3. Auto-classification of EDXRF analyte by using SOFM net As X-ray fluorescence analysis method is a relative analysis method depend on standard samples, so it is very essential to classify samples with different types (contains different elements and different content) to reduce the influence degree of matrix effect to element content. In practice research work, auto-classified technique is adopted which can classify samples with same matrix effectively and useful to correct the matrix effect.A model of Self-Organizing Feature Map (SOFM) net was established in this paper for XRF technology. The net can pick-up the characteristic value of samples to get the cluster center and achieve the classification of amount of samples. The model can classify and recognize complex samples effectively; it also has some excellence of having self-organizing structure, fast study rate, great mapping and popularize faculty. The established artificial neural net model has some function of memorizing new type, associating existed type, study unknown type, and finally achieve effective auto-classification of complex samples, thus, the influence probability of complex samples to matrix effect and measurement results is decreased.4. Dynamic nonlinear forecast of EDXRF measurement data by using RBF netOn the basis of influence character among elements and auto-classification technique research, one dynamic nonlinear forecast technique is used to establish a dynamic nonlinear data forecast model for EDXRF analysis. The model can forecast the nonlinear relation between element content and fluorescence count rate dynamically. The model not only has arbitrary precision approximation of the functional capacity but also has approximation with optimal functional characteristics and fast convergence rate, as well as self-correcting. The application of this model can improve the analysis accuracy of EDXRF. The application result is shown that, the combination of this model and auto-classification technique can decrease the influence of matrix greatly.This task has developed a fundamental research work, and also made some new concept in method theory and technique. The final result is achieved the purpose to decrease the matrix effect influence of complex samples analyzed by EDXRF, and establish an integrated approach to correct matrix effect.