LAUSR

Abstract We have addressed the issue of improper and unreliable analysis of materials characterization data by developing an artificial intelligence based methodology that can reliably and more efficiently analyze experimental results from extended X-ray absorption fine structure (EXAFS) measurements. Such methods help address growing reproducibility problems that are slowing research progress, discouraging the quest for research excellence, and inhibiting effective technology transfer and manufacturing innovation. We have developed a machine learning system for automated analysis of EXAFS spectroscopy measurements and demonstrated its effectiveness on measurements collected at powerful, third generation synchrotron radiation facilities. Specifically, the system uses a genetic algorithm to efficiently find sets of structural parameters that lead to high quality fits of the experimental spectra. A human analyst suggests a set of chemical compounds potentially present in the sample, used as theoretical standards. The algorithm then searches the large multidimensional space of combinations of these materials to determine the set of structural paths using the theoretical standards that best reproduces the experimental data. The algorithm further calculates a goodness of fit value from the suggested standards that can be used to identify the chemical moieties present in the measured sample.