Quantitative analysis of 16-17th century archaeological glass vessels using PLS regression of EPXMA and mu-XRF data

In this paper we report on the application of partial least squares regress ion (PLS) to electron probe X-ray microanalysis (EPXMA) and micro X-ray flu orescence (mu -XRF) analysis of 16-17th century archaeological glass sample s for the determination of Na2O SiO2, K2O, CaO, MnO and Fe2O3. Traditional quantitative processing of the spectral data acquired by these analytical t echniques is time-consuming and requires a skilled operator. The applicatio n of PLS improves both the speed and simplicity of the analysis. A total of 180 glass samples were analyzed by EPXMA and a subset of 53 samples by mu -XRF. These data sets serve as a basis to investigate the performance of PL S when applied to these kinds of analytical techniques. In the case of EPXM A a small training set of 25 carefully selected samples is used to build th e PLS model, while for mu -XRF all 53 samples are used. Results show that t he PLS method is able to quantify the major constituents of glass (SiO2, K2 O and CaO) with a relative error smaller than 3% in the case of EPXMA data and 5% in the case of mu -XRF. The EPXMA and mu -XRF data sets are also use d to investigate how the matrix effect present in EPXMA and mu -XRF analysi s affects the predictive ability of PLS. The effect of different preprocess ing techniques is studied as well. Copyright (C) 2000 John Wiley & Sons, Lt d.