ALBERT

Description: We have built three multivariate analysis mathematical models based on principal component analysis (PCA), partial least squares (PLS), and artificial neural networks (ANNs) to detect sulfate minerals in geological samples from laser Raman spectral data. We have critically assessed the potential of the models to automatically detect and quantify the abundance of selected Ca-, Fe-, Na-, and Mg-sulfates in binary mixtures. Samples were analyzed using a laboratory version of the Raman laser spectrometer (RLS) instrument onboard the European Space Agency 2018 ExoMars mission. Our results show that PCA and PLS, can be used to quantify to some extent the abundance of mineral phases. PCA separated hydrated from dehydrated mixtures and classified mixtures depending on the phase abundances. PLS provided relatively good calibration curves for these mixtures. Upon spectral pre-processing, ANNs provided the most precise qualitative and quantitative results. The detection of mineral phases was 100% accurate for pure samples, as was for binary mixtures where the abundance of mineral phases was 〉10%. The outputs of the ANN were proportional to the phase abundance of the mixture, thus demonstrating the ability of ANNs to quantify the abundance of different phases without the need for calibration. Taken together, our findings demonstrate that multivariate analysis provides critical qualitative and quantitative information about the studied sulfate minerals.

Description: The Raman Laser Spectrometer (RLS) is part of the payload of the 2018 ExoMars rover. The Sample Preparation and Distribution System (SPDS) of the rover will crush samples acquired from down to two meters depth under the Martian surface, and provide them to the RLS instrument in the form of flattened powdered samples. The RLS instrument will acquire a minimum of 20 points on the flattened surface of the samples. To be able to obtain the maximum scientific return from the instrument once on Mars, a simulator of the SPDS system has been built to perform a series of experiments in a representative scenario. The crushing process implies the loss of rock structure and texture and, hence, the geological context of the samples. However, qualitative analysis with the RLS simulator on powdered natural samples and rocks showed that the RLS is capable of detecting carbonaceous material occurring in trace amounts in one of the rock samples (a silicified volcanic sand), more easily than with the same analysis on bulk. Furthermore, it is shown that minor phases in carbonate cements that cannot be detected by Raman in the bulk sample can be detected in the powder, thus allowing the identification of all the carbonate phases present in the cement crust. In order to quantify the detection threshold of the instrument, further analysis on controlled samples were performed. The results with the RLS SPDS simulator showed that the instrument can reach detection thresholds down to 1 % on powdered samples. Furthermore, analysis of controlled mixtures showed that performing a very simple intensity-based statistical analysis of the spectra can provide semi-quantification of the abundance of the mineral species with quite linear calibration curves.