|
|
Using Laser-Induced Breakdown Spectroscopy (LIBS) for Material Analysis
Pořízka, Pavel ; Hrdlička, Aleš (oponent) ; Pína,, Ladislav (oponent) ; Kaiser, Jozef (vedoucí práce)
This doctoral thesis is focused on further development of the Laser-Induced Breakdown Spectroscopy (LIBS) device for in-situ and in real-time classification and quantification of samples. The major part of work, namely the whole experimental part for this thesis, was conducted at the Federal Institute for Material Research and Testing (BAM) in Berlin, Germany where a simple LIBS system was constructed. In parallel to experimental work, the literature was surveyed with the aim to give a thorough view on the usage of chemometrics in the LIBS community. The application of chemometric algorithms on LIBS data is generally recommended when more complex data sets are obtained. The research was primarily aimed on the LIBS capability of quantitative analysis and classifying the igneous rocks. Variety of samples was measured employing a simple LIBS system. The sample set was compiled from certified reference materials as well as from real samples collected directly at copper mines in Iran. The samples from Iran were classified in-situ by an experienced geologist and the copper content was estimated at the University of Clausthal, Germany. Even though the certified reference materials were analysed, the resulting calibration curve was highly nonlinear. For each individual rock type the relevant part of the calibration curve was observed under different trend. This separation of the calibration curve was assigned to the so-called matrix effect, which strongly affects the LIBS measurement. In other words, when different matrices with complex composition are analysed at once, the quantitative analysis employing the univariate calibration curves may not be reliable. Moreover, the normalization of such calibration curves using the intensity of selected matrix element lines did not let to a significant improvement in their linearity. It is generally not possible to pick up one line, which could perform the linearization independently on the complex data matrices. Chemometric algorithms, such as principal component regression (PCR) and partial least squares regression (PLSR), were used for multivariate calibration. PCR and PLSR may compensate for the matrix effect only to a certain extent. Furthermore, samples were successfully classified based on their spectral fingerprint (i.e. composition of matrix elements) employing principal component analysis (PCA) and Kohonen’s selfs-organizing maps. On the basis of theory and results, a solution for the reliable classification and quantification of unknown samples is proposed. The whole study should contribute to the processing of the analytical data measured by the in-situ stand-off LIBS device which is currently being constructed at Brno University of Technology in Brno, Czech Republic. However, LIBS can fulfil its potential as the versatile and irreplaceable technique for in-situ classification and quantitative analysis only when utilized with chemometric algorithms and data libraries. For those purposes, a fragment of the data library has already been established and tested for the application of LIBS to the mining industry.
|
host ::
RSS.