Electrochemical impedance spectroscopy (EIS) is a world-wide characterization technique used on many fields such as electrochemistry, biology, and medicine. Although EIS can provide meaningful insights on the system under study its interpretation is challenging. Several approaches based on physical models and equivalent circuits have been developed, however, the formers are usually too specific, and the latter are not unique. Recently, the distribution of relaxation of time (DRT) has risen as an effective methodology to address these issues. The conventional deconvolution of the DRT presented two main limitations: 1) the deconvolution of the DRT was performed assuming that EIS data were exclusively a function of the frequency. 2) diffusive systems could not be properly addressed since D...[ Read more ]

Electrochemical impedance spectroscopy (EIS) is a world-wide characterization technique used on many fields such as electrochemistry, biology, and medicine. Although EIS can provide meaningful insights on the system under study its interpretation is challenging. Several approaches based on physical models and equivalent circuits have been developed, however, the formers are usually too specific, and the latter are not unique. Recently, the distribution of relaxation of time (DRT) has risen as an effective methodology to address these issues. The conventional deconvolution of the DRT presented two main limitations: 1) the deconvolution of the DRT was performed assuming that EIS data were exclusively a function of the frequency. 2) diffusive systems could not be properly addressed since DRT cannot be properly deconvolved from unbounded impedance. These two issues are addressed using the promising deep-neural-network (DNN) as a tool for DNN deconvolution. Leveraging a DNN system, the deep-DRT model can analyse EIS data including the experimental conditions performing both regression and prediction without the need of any regularization or specific spacing. The use of DNN is still at the pioneering stage, and the DNN-DRT provide a guidance to reduce the cumbersome training time, a formal DNN error analysis, and addresses impedance models whose DRT has negative peaks. The models included in this thesis will help to unravel insights from EIS data probed from electrochemical systems by promoting the use of DNN systems.