With LIBS, a pulsed laser ionizes matter and the instrument collects and analyzes the spectrum emitted during electron recombination. I’ve used the technique for more than 30 years, beginning with environmental analysis of pollutants in atmosphere, water and soil, and latterly industrial diagnostics, forensic and bio-medical applications, as well as cultural heritage studies and conservation. It is an interesting technique suitable for a range of applications, particularly when robustness, reliability, speed of analysis and operational availability are important.
To get accurate, reproducible results from LIBS you need to plan your measurements thoroughly – and have access to a good instrument. But the procedure for extracting relevant information from the LIBS spectrum is just as important as the quality of the instrument being used to acquire it.
The classical approach to quantitative analysis of materials with LIBS uses calibration curves obtained from several reference samples. Unfortunately, LIBS signals are extremely sensitive to any change in laser-sample coupling produced by laser irradiance fluctuations or by matrix effects. Consequently, LIBS is placed at the lower end of the analytical figures of merit, behind more established techniques, such as X-ray fluorescence (XRF) or inductive coupled plasma-optical emission spectroscopy (ICP-OES), for example. Several years ago – to address the above lower ranking of LIBS – my group developed a calibration-free (CF) approach using our knowledge of the chemical/physical processes in the laser-induced plasma. And it proved very effective for correcting all the effects that would prevent its use. For example, CF-LIBS analysis overcomes the matrix effect enabling in principle very accurate measurements. However, the CF-LIBS algorithm must be applied to a single spectrum, eventually averaging the elemental concentrations when taking several measurements from the same sample. Many people do the reverse and (for practical reasons) apply the CF-LIBS method to the average spectrum instead, which is conceptually incorrect. An alternative approach to LIBS quantitative analysis uses artificial neural networks (ANN), which is much quicker than the CF method. In the ANN method, the inputs (LIBS spectra intensities) are combined (non-linearly) to produce outputs (the corresponding elemental concentrations). During the training stage, you optimize the coefficients of the non-linear combination from the inputs to find the best correspondence between inputs and outputs using a set of test samples. Then you validate the reliability of the results using a different set of (known) samples. Training the ANN is similar to constructing a (multidimensional and nonlinear) calibration curve, which eventually becomes a surface in a multidimensional parameter space. ANNs are extremely fast and flexible, they operate on single spectra but suffer similar problems to other methods using calibration curves. In particular, the experimental conditions for acquiring calibration spectra must be consistent throughout the entire calibration process and must be constant during acquisition of LIBS spectra from unknown samples. In addition, the ANN approach is very sensitive to laser-sample coupling variations and matrix effects. Most recently, I’ve demonstrated how to embed the basic equations used in CF-LIBS within an ANN algorithm, thus combining the advantages of the two methods: the speed of an ANN and the precision of CF analysis. Therefore, I have finally succeeded in teaching plasma physics to a computer – and it passed its final examinations with good grades!
Newsletters
Receive the latest analytical science news, personalities, education, and career development – weekly to your inbox.
