矿样中钛铁EDXRF分析的基体效应和神经网络校正研究

Matrix effect and ANN correcting technique in EDXRF analysis of Ti and Fe in core samples

用能量色散X荧光(EDXRF)分析仪测量钛铁矿样时,存在基体效应影响分析准确性的问题。本文用EDXRF测得钛、铁元素特征峰,通过类Gauss函数对实测谱进行拟合分解,初步分析了矿样中Ti-Fe间的影响特征。针对各元素计数率与含量的非线性关系,采用先分类后预测的方法,首先用SOFM自适应神经网络对矿样分类,样本总数80组,对铁精矿、钛精矿的识别率为100%;然后用RBF神经网络进行钛铁含量预测,与化学分析结果对比,其中65.4%的样品相对误差在1%以内,其余均在3%以内,小于工业生产仪器分析相对误差5%的要求,表明基于先分类后预测的神经网络校正技术在矿样元素含量分析中有着一定的实用价值。

Matrix effect always affects EDXRF （Energy Dispersive X-ray Fluorescence） analysis of mineral samples In this work, characteristic X-ray peaks of Ti and Fe in different ore samples were fitted by Gaussian fitting function and analyzed in an attempt to solve nonlinear relation between the counting rate and elemental contents, Classification and forecast methods were used in the analysis. Core samples （80 groups） were classified by the SOFM self-adapt neural net and the identification rate of concentrated iron ore and concentrated titanium ore samples is 100%. The RBF net was used to forecast the contents of Ti and Fe. Comparing the forecast results with chemical analysis results, relative errors of 65.4% samples were less than ±1%, and the others were less than ±3%, whereas less than 5% is demanded by the industry production. The results show that it is an effective method to use ANN technique to classify and forecast element content in core samples.