LIBS feat. AI

This approach promises accurate and efficient interpretation of complex data not only in spectral but also in quantitative and qualitative analysis.

Due to the distinct spectral characteristics that every element possesses, training the ANN on vast libraries of LIBS spectra from known materials allows the network to associate specific spectral patterns with different elements, even in complex mixtures. This newfound ability allows for rapid identification of the elements present in a sample. This is particularly valuable for analyzing materials with numerous components.

To ensure the proper functioning of ANN, besides the training phase, a multi-step procedure consisting of several other steps needs to be followed. After selecting the appropriate and informative input parameters, we need to pre-process the data and extract relevant information. Finally, adjusting the design of the ANN architecture is necessary, which determines how the network processes the information.



The study from N. Herreyre et al. (2023) has already showcased the remarkable capabilities of Artificial Neural Networks for analyzing spectral data from micro-LIBS. Their study was focused on developing a simple ANN algorithm to identify the composition of archaeological mortar. This research highlighted a key advantage of ANNs, which led to the realization that even a basic network structure can achieve fast, automatic, robust, and efficient processing of a dataset.

Over 1300 reference spectra of various well-known materials were used to train the ANN.

Inspired by these remarkable results, we launched our activities to develop a secure classification method for 7 different species of lithium-bearing mineral samples. After collecting the representative big data files using our M-Trace LIBS device, we have implemented some of the ANN learning techniques with impressive 97 to 99 % classification reliability!
Following the fact, that our LIBS instruments can produce hundreds to thousands of spectral data in seconds and the ability of AI/ANN to analyze and use these Big Data shows the path AtomTrace will follow.
With the integration of AI we will change the simple elemental analysis into user-friendly material classification and categorization.