Electrochemical impedance spectroscopy (EIS) is an effective method for studying the electrochemical systems. The interpretation of EIS is the biggest challenge in this technology, which requires reasonable modeling. However, the modeling of EIS is of great subjectivity, meaning that there may be several models to fit the same set of data. In order to overcome the uncertainty and triviality of human analysis, this research uses machine learning to carry out EIS pattern recognition. Raw EIS data and their equivalent circuit models were collected from the literature, and the support vector machine (SVM) was used to analyze these data. As the result, we addresses the classification of EIS and recognizing their equivalent circuit models with accuracies of up to 78%. This study demonstrates the great potential of machine learning in electrochemical researches.
Electrochemical impedance spectroscopy (EIS) is to study the mechanism of electrode materials, solid electrolytes, conductive polymers and corrosion protection by measuring the change of impedance with sinusoidal frequency [1]. The fuzziness of interpretation is probably the biggest problem for EIS technology. Constructing equivalent circuit model is the most widely used method for EIS analysis [2]. In this method, the electrochemical system is regarded as an equivalent circuit, which is composed of basic components in series or parallel, such as resistance (R), capacitance (C) and constant phase element (Q).
The structure of these equivalent circuit and the value of each element can be measured and fitted. Consequently, the details of the electrochemical systems and the properties of the electrode processes can be analyzed by using the electrochemical meaning of these components [1,2].
The model selection should reflect the practical significance, and the consequence discussion are based on the assumption that there is an accurate model. At present, the common method is screening out several potential equivalent circuit models according to the different applications, and then using mathematical fitting to simulate the corresponding pattern. Nevertheless, when several trusted models are available, they should be sorted to find the most reasonable model. Therefore, many subjective factors and judgments involved in the process of analyzing the equivalent circuit of EIS. How to choose the suitable equivalent circuit model is the critical step in the EIS technology.
Machine learning is a kind of algorithm which automatically analyzes and obtains the rule from the data and uses the rule to predict the unknown data [3,4]. The most appropriate choice of a variety of possible models is a typical classification problem. Classification is a common task in machine learning. To this regard, machine learning has a great application prospect in dealing with EIS analysis. Among various machine learning technologies, we choose the support vector machines (SVM) to deal with this classification problem and recognize the most suitable of EIS results.
The mechanism of SVM is that data points are regarded as p-dimensional vectors, and these points can be separated by (p-1)-dimensional hyperplanes. There may be many hyperplanes that can categorize the data. SVM can find the best hyperplane which maximizes the distance to the nearest data point on each side. SVM has been applied to pattern recognition (pattern recognition) problems such as portrait recognition (face recognition), text classification (text categorization) and so on.
In this paper, a SVM was constructed to analyze the suitable model of EIS results. The raw EIS data were collected from published articles, which focus on the electrochemical energy storage applications (i.e. batteries and supercapacitors). Hundreds of EIS data and their equivalent circuit models were studied by SVM.
To conduct this research, the first step is to establish an EIS database, thus we extracted over 250 sets of EIS data and their equivalent circuit from published papers. There are two main reasons to extract data from the existing literature. First, there are a lot of related researches, suggesting that the total amount of data is large enough. Considering that the test instruments, test specifications and selected parameters used by each researcher are very different, it can provide enough universality for our data analysis process. On the other hand, in the published papers, the equivalent circuit models of EIS must have been carefully
During the process, the normalized EIS is used as the input data and the corresponding equivalent circuit as the output data. The data analysis system is established by machine learning algorithm. The whole process is shown in Figure 1. After extracting enough EIS data, the next step is to classify the equivalent circuit of EIS.
The EIS data we selected come from electrochemical energy storage devices, especially lithium-ion batteries and supercapacitors. 5 main equivalent circuit models are summarized as shown in Figure 2.
The reason why these five types is from the essence of EIS equivalent circuit. All these models consist of three components, which are impedance, constant phase element (CPE) and Warburg element (W). Impedance is a general term for the hindrance of resistance, inductance and capacitance. In the actual test, capacitors in EIS experiments often do not behave ideally. Instead they act like a constant phase element (CPE). Warburg element is used to describe the electrode behavior when the charge diffuses through a barrier layer.
At very low frequencies, charged ions can spread deep and even penetrate the diffusion layer to produce a finite thickness Warburg element, which, if the diffusion layer is thick or dense enough, will result in even at a low limit of frequency, forming an infinite thickness of Warburg elements. On the basi
This content is AI-processed based on open access ArXiv data.