Simultaneous quantification of Au and Ag composition from Au-Ag bi-metallic LIBS spectra combined with shallow neural network model for multi-output regression

摘要

The multi-output regression model has been used in this work to predict the simultaneous compositions of gold (Au) and silver (Ag) from the laser-induced breakdown spectroscopy (LIBS) spectra of an Ag-Au bimetallic target. Multi-output regression with shallow neural networks uses two nodes at the output layer to predict the composition of one element each. The LIBS spectra of the bimetallic target at three different compositions (Au20-Ag80, Au50-Ag50, and Au80-Ag20) in conjunction with aluminium, copper, and bronze LIBS spectra (substituting for Au0-Ag0 composition) were used to train the model. Principal component analysis (PCA) was performed for dimensionality reduction and the output of the PCA was fed to the regression model. The trained model, after evaluation on the test set, was used for predictions from unseen LIBS spectra of Au30-Ag70, which was not present in the original training set. We have demonstrated the success of this model with an error of 10%. For the first time, to the best of our knowledge, we used the time-resolved spectra for data augmentation to generalize the model and likewise to study any further improvement in the performance of the model. This facilitated in generalizing the model, which in turn improved the performance of the model. This method can further be developed to predict the composition of multiple elements simultaneously. The terminology used in the deep learning models is elucidated with simplified equations and illustrations.