主成分分析结合极限学习机辅助激光诱导击穿光谱用于铝合金分类识别

随着我国社会发展,废旧产品的数量迅速增长,废旧铝随之大量产生。铝是优良的再生资源,传统分选技术不能将废旧铝按各自的成分牌号进行精细分类,导致很多优质铝资源被降级使用,造成巨大的浪费。研究了主成分分析(PCA)结合极限学习机(ELM)算法辅助激光诱导击穿光谱(LIBS)技术在铝合金分类识别方面的应用。选用2种系列的4个牌号铝合金作为实验样品,通过LIBS技术激发实验样品获得420组光谱数据。对原始光谱数据进行了预处理,并选取样品铝合金中5种主要差异元素(Mg、Mn、Cu、Fe和Si)的21条特征谱线构成了420×21的光谱数据矩阵,通过主成分分析对光谱数据进一步降维,使得模型输入变量从21个降至8个。选取120组光谱数据作为训练集,建立了基于极限学习机的铝合金分类模型,余下300组数据作为测试集。研究发现在主要非铝元素(Mg、Mn、Cu、Fe和Si)含量差异只有0.0021%~3.68%的情况下,PCA-ELM分类模型的平均识别准确率达到98.01%,标准差为0.82%,建模时间为0.081s。结果表明,PCA-ELM分类模型有着很高的效率及稳定性,将其与LIBS技术结合可以适用于工业快速分类领域,为精细分类行业提供了一种参考方法。

Application of laser-induced breakdown spectroscopy assisted by principal component analysis and extreme learning machine in the classification recognition of aluminum alloy

The quantity of waste products rapidly increases with the development of society in our country. A large number of waste aluminum is also generated. Aluminum is a kind of good renewable resources. However, the traditional separation technology cannot realize the sophisticated classification of waste aluminum according to the respective brand grades. As a result, many high-quality aluminum resources are used after downgrading, leading to a huge waste. The application of laser-induced breakdown spectroscopy (LIBS) assisted by principal component analysis (PCA) and extreme learning machine (ELM) algorithm in the classification recognition of aluminum alloy was investigated. Four brands of aluminum alloy in two series were selected as the experimental samples. Total 420 groups of spectral data were obtained by excitation of sample using LIBS technology. The original spectral data were pretreated. Then 21 spectral lines of five elements (Mg, Mn, Cu, Fe and Si) with main difference in aluminum alloy were selected to constitute 420×21 spectral data matrix. The spectral data were further treated by dimensionality reduction using PCA method. The input variable of model was decreased from 21 to 8. Then 120 groups of spectral data were selected as the training set to establish classification model of aluminum alloy based on ELM. The rest 300 groups of data were selected as the testing set. Under the condition that the content difference of main non-aluminum elements (Mg, Mn, Cu, Fe and Si) was only 0.0021%-3.68%, it was found that the average recognition accuracy rate of classification model based on PCA-ELM was up to 98.01%. The standard deviation was 0.82%. The modeling time was 0.081s. The results showed that the classification model based on PCA-ELM exhibited very high efficiency and stability. Its combination with LIBS technique could be applied for the rapid classification fields in industry. The proposed study provided a reference for the sophisticated classification industry.