Multi-q Pattern Classification of Polarization Curves
Ricardo Fabbri,Ivan N. Bastos,Francisco D. Moura Neto,Francisco J. P. Lopes,Wesley N. Goncalves,Odemir M. Bruno
DOI: https://doi.org/10.1016/j.physa.2013.09.048
2013-05-10
Abstract:Several experimental measurements are expressed in the form of one-dimensional profiles, for which there is a scarcity of methodologies able to classify the pertinence of a given result to a specific group. The polarization curves that evaluate the corrosion kinetics of electrodes in corrosive media are an application where the behavior is chiefly analyzed from profiles. Polarization curves are indeed a classic method to determine the global kinetics of metallic electrodes, but the strong nonlinearity from different metals and alloys can overlap and the discrimination becomes a challenging problem. Moreover, even finding a typical curve from replicated tests requires subjective judgement. In this paper we used the so-called multi-q approach based on the Tsallis statistics in a classification engine to separate multiple polarization curve profiles of two stainless steels. We collected 48 experimental polarization curves in aqueous chloride medium of two stainless steel types, with different resistance against localized corrosion. Multi-q pattern analysis was then carried out on a wide potential range, from cathodic up to anodic regions. An excellent classification rate was obtained, at a success rate of 90%, 80%, and 83% for low (cathodic), high (anodic), and both potential ranges, respectively, using only 2% of the original profile data. These results show the potential of the proposed approach towards efficient, robust, systematic and automatic classification of highly non-linear profile curves.
Computational Engineering, Finance, and Science,Computer Vision and Pattern Recognition