|
|
Effect of magnetic field on laser-induced plasma characteristics
Vrábel, Jakub ; Pořízka, Pavel (referee) ; Prochazka, David (advisor)
This work is based on topic of laser induced breakdown spectroscopy „LIBS“ and its application to chemical analysis of materials. Paper presents a brief theoretical basis of LIBS with focus on effect of magnetic field to plasmas and modeling spectras for calib-ration-free LIBS analysis. Considerable part of work is dedicated to practical demon-stration of modeling spectras, using Matlab and interpretation of experimental results. The final output is automatized method for evaluation of spectras, which could leads to calibration-free LIBS analysis. Also there is an effort to increase sensitivity of method.
|
|
|
Interconnection of Restricted Boltzmann machine method with statistical physics and its implementation in the processing of spectroscopic data
Vrábel, Jakub ; Hrdlička, Aleš (referee) ; Pořízka, Pavel (advisor)
Práca sa zaoberá spojeniami medzi štatistickou fyzikou a strojovým učením s dôrazom na základné princípy a ich dôsledky. Ďalej sa venuje obecným vlastnostiam spektroskopických dát a ich zohľadnení pri pokročilom spracovaní dát. Začiatok práce je venovaný odvodeniu partičnej sumy štatistického systému a štúdiu Isingovho modelu pomocou "mean field" prístupu. Následne, popri základnom úvode do strojového učenia, je ukázaná ekvivalencia medzi Isingovým modelom a Hopfieldovou sieťou - modelom strojového učenia. Na konci teoretickej časti je z Hopfieldovej siete odvodený model Restricted Boltzmann Machine (RBM). Vhodnosť použitia RBM na spracovanie spektroskopických dát je diskutovaná a preukázaná na znížení dimenzie týchto dát. Výsledky sú porovnané s bežne používanou Metódou Hlavných Komponent (PCA), spolu so zhodnotením prístupu a možnosťami ďalšieho zlepšovania.
|
|
|
Correction of laser-induced breakdown spectroscopy calibration models for changing ablation energies
Dvořák, Tomáš ; Vrábel, Jakub (referee) ; Képeš, Erik (advisor)
This bachelor thesis investigates transfer learning as a potential method for correcting calibration models based on multilayer perceptron neural networks in laser-induced breakdown spectroscopy due to changes in ablation energies. MLP calibration models for four elements (chromium, nickel, molybdenum and manganese) and six different ablation energies (40, 50, 60, 70, 80 and 90 mJ) were designed and trained. Models initially trained at an ablation energy of 50 mJ were then used to predict concentrations from spectra measured at the different ablation energies. Subsequent differences in the measurements were resolved by applying transfer learning. The results suggest that transfer learning could serve as a valid method for correcting inaccuracies due to differences in ablation energy, achieving comparable results to models trained from scratch in a fraction of the time and at significantly lower computational cost. However, this study could not clearly demonstrate consistent improvements in neural network performance using transfer learning in the context of LIBS. Likely contributing factors are the lack of optimization of the neural networks used, the limited complexity of the experimental dataset, or a combination of both. This research builds on and extends the literature and offers a deeper understanding of the capabilities and limitations of transfer learning in the context of LIBS.
|
|
|
Artificial neural networks for prediction of laser-induced plasma parameters from emission spectra
Grünwald, Martin ; Képeš, Erik (referee) ; Vrábel, Jakub (advisor)
This bachelor thesis deals with predictions of plasma parameters using an artificial neural network and construction of datasets designed for training artificial neural networks on spectroscopic data. The basic theory of laser-induced breakdown spectroscopy (LIBS), basic aspects of plasma physics and an introduction to artificial neural network theory are all included in this thesis. Furthermore, the thesis focuses on the proceduce of generation, augmentation and preprocessing of a dataset. Along with that, the method for generation of various spectroscopic data and subsequent artificial neural network training is described. The outcome of the thesis is a series of scripts made for executing individual steps and also a proof of concept for plasma parameter prediction from LIBS spectra using an artificial neural network model.
|
|
|
Interconnection of Restricted Boltzmann machine method with statistical physics and its implementation in the processing of spectroscopic data
Vrábel, Jakub ; Hrdlička, Aleš (referee) ; Pořízka, Pavel (advisor)
Práca sa zaoberá spojeniami medzi štatistickou fyzikou a strojovým učením s dôrazom na základné princípy a ich dôsledky. Ďalej sa venuje obecným vlastnostiam spektroskopických dát a ich zohľadnení pri pokročilom spracovaní dát. Začiatok práce je venovaný odvodeniu partičnej sumy štatistického systému a štúdiu Isingovho modelu pomocou "mean field" prístupu. Následne, popri základnom úvode do strojového učenia, je ukázaná ekvivalencia medzi Isingovým modelom a Hopfieldovou sieťou - modelom strojového učenia. Na konci teoretickej časti je z Hopfieldovej siete odvodený model Restricted Boltzmann Machine (RBM). Vhodnosť použitia RBM na spracovanie spektroskopických dát je diskutovaná a preukázaná na znížení dimenzie týchto dát. Výsledky sú porovnané s bežne používanou Metódou Hlavných Komponent (PCA), spolu so zhodnotením prístupu a možnosťami ďalšieho zlepšovania.
|
|
|
Effect of magnetic field on laser-induced plasma characteristics
Vrábel, Jakub ; Pořízka, Pavel (referee) ; Prochazka, David (advisor)
This work is based on topic of laser induced breakdown spectroscopy „LIBS“ and its application to chemical analysis of materials. Paper presents a brief theoretical basis of LIBS with focus on effect of magnetic field to plasmas and modeling spectras for calib-ration-free LIBS analysis. Considerable part of work is dedicated to practical demon-stration of modeling spectras, using Matlab and interpretation of experimental results. The final output is automatized method for evaluation of spectras, which could leads to calibration-free LIBS analysis. Also there is an effort to increase sensitivity of method.
|
guest ::
RSS feed.