Machine Learning Benchmarks for the Classification of Equivalent Circuit Models from Electrochemical Impedance Spectra

Abstract

Analysis of electrochemical impedance spectroscopy (EIS) data for electrochemical systems often consists of defining an equivalent circuit model (ECM) using expert knowledge and then optimizing the model parameters to deconvolute various resistance, capacitive, inductive, or diffusion responses. For small data sets, this procedure can be conducted manually; however, it is not feasible to manually define a proper ECM for extensive data sets with a wide range of EIS responses. Automatic identification of an ECM would substantially accelerate the analysis of large sets of EIS data. We showcase machine learning methods to classify the ECMs of 9300 impedance spectra provided by QuantumScape for the BatteryDEV hackathon. The best-performing approach is a gradient-boosted tree model utilizing a library to automatically generate features, followed by a random forest model using the raw spectral data. A convolutional neural network using boolean images of Nyquist representations is presented as an alternative, although it achieves a lower accuracy. We publish the data and open source the associated code. The approaches described in this article can serve as benchmarks for further studies. A key remaining challenge is the identifiability of the labels, underlined by the model performances and the comparison of misclassified spectra.

Is Unsupervised Dimensionality Reduction Sufficient to Decode the Complexities of Electrochemical Impedance Spectra?

“…To address the growing challenges in processing EIS data, multiple research groups have independently explored the use of supervised machine learning (ML) to automate the identification of appropriate equivalent circuits. [12][13][14][15][16][17] This trend is not exclusive to EIS but is also observed in the broader field of electrochemical sciences, including the analysis of voltammetric data. [18][19][20][21][22] A plethora of methodologies have been spotlighted, including decision trees, [12] random forests, [12,14] the naive Bayes classifier, [12] AdaBoost, [12] support vector machines, [15,16] and various neural network (NN) architectures.…”

“…[12][13][14][15][16][17] This trend is not exclusive to EIS but is also observed in the broader field of electrochemical sciences, including the analysis of voltammetric data. [18][19][20][21][22] A plethora of methodologies have been spotlighted, including decision trees, [12] random forests, [12,14] the naive Bayes classifier, [12] AdaBoost, [12] support vector machines, [15,16] and various neural network (NN) architectures. [13,14,[18][19][20] While these strategies have achieved commendable success -with some classifier algorithms nearing a staggering 99 % accuracy [11,17,19] their drawback remains the indispensable need for labeled data.…”

“…[18][19][20][21][22] A plethora of methodologies have been spotlighted, including decision trees, [12] random forests, [12,14] the naive Bayes classifier, [12] AdaBoost, [12] support vector machines, [15,16] and various neural network (NN) architectures. [13,14,[18][19][20] While these strategies have achieved commendable success -with some classifier algorithms nearing a staggering 99 % accuracy [11,17,19] their drawback remains the indispensable need for labeled data. The term "labeled data" refers to the process of matching specific equivalent circuits with their corresponding EIS spectra.…”

“…An alternative approach highlights the use of synthetic EIS data, where the dataset is generated numerically from predefined equivalent circuits. [13,14,23] While this method appears to circumvent the previously mentioned challenge, it still necessitates prior knowledge regarding the specific equivalent circuits to be modeled. In this study, we endeavor to surmount the mentioned challenges associated with the need for labeled data by employing unsupervised ML methodologies.…”

Introducing Hybrid Modeling with Time-Series-Transformers: A Comparative Study of Series and Parallel Approach in Batch Crystallization

“…By integrating real-time data, advanced analytics, and ML/AI algorithms, digital twins enable continuous monitoring, predictive analysis, and optimization of chemical processes in real-time. , A majority of the existing digital twins utilize data-driven black-box models, which are majorly based on different ML architectures as they provide high computational efficiency. Specifically, deep, recurrent, or convolution neural networks (DNN, RNN, and CNN) are used to mimic the process dynamics of complex chemical systems. ,− For instance, various RNN and DNN models have been utilized in the literature to mimic the crystallization of pharmaceutical and food products by Wu and colleagues. − Similarly, Kwon and colleagues have DNN-based models for continuous crystallization of quantum dot (QD) systems.. , Similarly, other DNN-based models have also been demonstrated for modeling and control of thin-film deposition for different substrates. − Furthermore, Braatz and colleagues have demonstrated a plethora of impressive different ML models for the prediction of battery life, developing optimization or control frameworks using these ML models. − …”