人工神经网络直读光谱法测定钢中酸溶铝  

Application of Artificial Neural Networks to the Determination of Acid-Solved Aluminum in Steel by Direct Reading Spectrography

作　　者：孙晓波[1] 李井会[1] 董林[1] 高礼让[1] 宋立伟[2] 

机构地区：[1]鞍山钢铁学院化工学院,辽宁鞍山114002 [2]鞍山钢铁(集团)公司,辽宁鞍山114002

出　　处：《光谱学与光谱分析》2003年第6期1177-1179,共3页Spectroscopy and Spectral Analysis

摘　　要：用火花直读光谱法测定钢中不同状态的铝是目前研究的热点之一。本文确立了直读光谱法的分析条件,将人工神经网络用于直读光谱法,用直读光谱的全铝光谱强度值作为网络的输入,标样中的酸溶铝值作为期望值,建立神经网络模型,用神经网络直读光谱法测定中低合金钢中酸溶铝含量,分析结果的准确度与化学分析法基本一致。神经网络使用改进的BP算法,避免了学习过程可能产生的麻痹现象,提出了目标向量的简单变换方法。该法用于钢中酸溶铝的直接测定,获得满意结果。Analysts are always concerned with the analysis of aluminum state in steel. In this paper, the analytical conditions of aluminum by direct-reading spectrography were established, and artificial neural network (ANN) has been applied to sparkle direct-reading spectrography. The intensities of sparkle direct-reading spectrography were used as the input and the acid-solved aluminum values of standard samples as expected values in the ANN model, and the acid-solved aluminium in alloy steel was determined by ANN-sparkle direct-reading spectrography. The determined results obtained by ANN-sparkle direct-reading spectrography and those by chemical analysis are approximately identical. The paralysis in the procedure of training on the ANN has been avoided with the improved back propagation algorithms. The simple linear transformation method for target vectors about networks has been put forward. A method has been developed for direct determination of acid-solved aluminum in steel with satisfactory results.

关 键 词：合金钢 酸溶铝 含量测定 直读光谱法 人工神经网络 

分 类 号：TG142.33[一般工业技术—材料科学与工程] TG115.33[金属学及工艺—金属材料]