Análise de sinais eletroquímicos por mapas de difusão

Abstract

This thesis proposes an innovative use in the study of electrochemical signals by the technique of nonlinear dimensionality reduction known as Diffusion Maps. Diffusion maps can discover patterns in datasets, grouping them and considering different scales for classification. The dimensionality reduction was initially applied to digital images, showing the capacity of organizing objects and revealing the potentiality of this technique to treat data. In addition to providing a general bibliographic review of the method

and explaining its operation in detail, the work comprehensively analyzes the qualita- tive behavior of the technique in terms of modeling parameterization, bringing to the

debate issues that are still poorly understood about the method. Next, an application for classifying electrochemical signal profiles from polarization curves and electrochemical impedance spectroscopy is presented, mainly used to evaluate the corrosion kinetics of electrodes in corrosive environments. The analysis used the profiles of experimental tests with dozens of replicates. Correct classification is a challenge, as the strong non-linearity of the profiles, in addition to overlap, becomes a difficult task. Furthermore, one must imagine that outliers are present and these may have different origins, such as material variations, experimental errors and artifacts, and the classification procedure must be sensitive to detect them from the dataset. The results obtained are considered promising, reaching rates of 81, 88 and 79% for the low and high electrochemical potential ranges and for the entire studied range. These results are compared with classical classification methods in terms of efficiency achieved. This work also shows how Diffusion Maps can be useful in the search for outliers of electrochemical signals.After debugging the experimental data with the technique, the classifier reaches 94% in the high potential range, when the best result is obtained. Finally, the thesis also presents a critical analysis of the effect of the scale parameter that measures the similarity between the mapping data obtained with the diffusion maps using simulated data. This parameterization refers to structural connectivity and how data can be connected.