Electrochemical impedance spectroscopy (EIS) is one of the most powerful electrochemical characterization techniques used in electrochemistry research. One of the most challenging steps associated with EIS is data interpretation, which typically uses non-unique equivalent circuit models. The distribution of relaxation time (DRT) method is an alternative approach to equivalent circuits. However, obtaining DRT from EIS data involves solving an illposed inverse problem. We tackle this issue by formulating the DRT problem in a Bayesian statistical framework. The inherent flexibility of the Bayesian framework allows adding prior knowledge about the DRT. We propose two prior models. First, regularization penalty coefficients are treated as multivariate random vectors. Such formulation leads to a timescale-dependent regularization giving better DRT recovery than the conventional timescale independent approach. The second prior is proposed for the weights of the fitting residuals. We show by treating them as random vectors that anomalies present in the EIS data can be detected, leading to DRT recovery that is robust with respect to experimental noise. We further extend this approach to the context of Gaussian processes, showing that Bayesian statistics can be used to univocally determine hyperparameters.
Mechanical Engineering
OPEN TO: Public