Battery cell life diagnostic device and its operating method

Abstract

A battery cell diagnosis device according to an embodiment of the present invention can include a collection unit that collects time-series data of a battery cell according to operating conditions, a conversion unit that converts the time-series data into an image corresponding to a spatial trajectory of a predetermined dimension, an extraction unit that extracts a feature value from the image, and a calculation unit that calculates the life of the battery cell based on the feature value.

Description

[Technical Field]

[0001] The present invention claims the benefit of priority based on Korean Patent Application No. 10-2022-0067847 filed on June 2, 2022, and all contents disclosed in the document of said Korean patent application are incorporated as a part of the present specification.

[0002] Embodiments disclosed in this document relate to a battery cell life diagnosis apparatus and an operating method thereof. [Background Art]

[0003] In recent years, research and development on secondary batteries has been actively carried out. Secondary batteries are batteries that can be charged and discharged, and can include all of conventional Ni / Cd batteries, Ni / MH batteries, and the like, as well as all recent lithium ion batteries. Lithium ion batteries have the advantage of much higher energy density compared to conventional Ni / Cd batteries, Ni / MH batteries, and the like. In addition, since lithium ion batteries can be manufactured to be small and lightweight, they are used as power sources for mobile devices. In recent years, their application range has been expanded to power sources for electric vehicles, and they are attracting attention as next-generation energy storage media. [Prior art document] [Patent] [Patent Document 1] Japanese Unexamined Patent Publication No. 2020-106316 [Patent Document 2] Japanese Unexamined Patent Publication No. 2017-026616 [Patent Document 3] International Publication No. 2019 / 207399 [Patent Document 4] Japanese Unexamined Patent Publication No. 2013-217897

[0004] Batteries exhibit a tendency to degrade as charging and discharging are repeated. For example, as charging and discharging are repeated, the capacity and resistance of a battery degrade, and the remaining life can be shortened. Further, the degree of degradation and remaining life of a battery can change depending on usage conditions.

[0005] If the remaining life of a battery decreases sharply, safety problems may occur in the use of the battery. Therefore, there is a need for a method of diagnosing battery life in advance and preventing dangers caused by a sudden drop in battery performance. [Summary of Invention] [Problems that the invention aims to solve]

[0006] One objective of the embodiments disclosed in this document is to provide an apparatus and method for operating the apparatus that converts battery cell operating characteristic data, which is collected chronologically in response to operating conditions, into an image, so that the lifespan of the battery cell can be effectively calculated.

[0007] One objective of the embodiments disclosed in this document is to provide an apparatus for diagnosing the lifespan of a battery cell based on an image converted from time-series data, and a method for operating the same.

[0008] The technical problems of the embodiments disclosed in this document are not limited to those mentioned above, and other technical problems not mentioned can be clearly understood by those skilled in the art from the following description. [Means for solving the problem]

[0009] A battery cell diagnostic device according to one embodiment disclosed herein may include: an acquisition unit for collecting time-series data of a battery cell based on operating conditions; a conversion unit for converting the time-series data into an image corresponding to a spatial trajectory of a predetermined dimension; an extraction unit for extracting feature values ​​from the image; and a calculation unit for calculating the lifespan of the battery cell based on the feature values.

[0010] In one embodiment, the image may be an image corresponding to a trajectory in two-dimensional space. In one embodiment, the time-series data may correspond to data that reflects changes in the operating characteristics of the battery cell over time.

[0011] In one embodiment, the extraction unit can extract the feature values ​​via a first neural network. In one embodiment, the first neural network may be a convolutional neural network including convolutional layers and pooling layers.

[0012] In one embodiment, the calculation unit can calculate the lifetime via a second neural network. In one embodiment, the second neural network may be a deep neural network including multiple hidden layers.

[0013] In one embodiment, the operating conditions correspond to the feature values, and the calculation unit can generate a prediction function for the operating conditions via a third neural network.

[0014] In one embodiment, the third neural network may be a deep neural network including multiple hidden layers. In one embodiment, the calculation unit can calculate the lifetime based on the prediction function and the feature values.

[0015] In one embodiment, the conversion unit can convert the time-series data into a spatial trajectory and use the spatial trajectory to convert the time-series data into an image.

[0016] In one embodiment, the conversion unit can represent the distance between points located on the spatial trajectory using a distance matrix, and convert the region corresponding to the distance matrix into an image based on the distance values ​​of the distance matrix.

[0017] An operating method for a battery cell diagnostic device according to one embodiment disclosed herein may include the steps of: collecting time-series data of a battery cell based on operating conditions; converting the time-series data into an image corresponding to a spatial trajectory of a predetermined dimension; extracting feature values ​​from the image; and calculating the lifespan of the battery cell based on the feature values.

[0018] In one embodiment, the step of converting said time-series data into an image corresponding to a spatial trajectory of a predetermined dimension may comprise the step of converting said time-series data into a spatial trajectory, and the step of converting said time-series data into said image using said spatial trajectory.

[0019] In one embodiment, the step of converting said time-series data into said image using said spatial trajectory may comprise the step of representing distances between points located on said spatial trajectory with a distance matrix, and the step of visualizing a region corresponding to said distance matrix based on distance values of said distance matrix.

[0020] In one embodiment, the step of extracting feature values from said image may extract said feature values via a first neural network. In one embodiment, the step of calculating said lifetime calculates said lifetime via a second neural network, and said second neural network may be a deep neural network comprising multiple hidden layers.

[0021] In one embodiment, said operating condition corresponds to said feature value, and the step of calculating said lifetime may comprise the step of generating a prediction function for said operating condition via a third neural network, the step of concatenating said prediction function and said feature value, and the step of predicting said lifetime based on said prediction function and said feature value. Effects of the Invention

[0022] The battery cell life diagnosis device and the operating method thereof according to an embodiment disclosed in this document can calculate the state of a battery cell. The battery cell diagnosis device and the operating method thereof according to an embodiment disclosed in this document can calculate the state of a battery cell by converting time-series data of the operating characteristics of the battery cell into an image of a predetermined dimension.

[0023] A battery cell diagnostic device and its operating method according to one embodiment disclosed herein can calculate the remaining lifespan of a battery cell by analyzing image data via a neural network.

[0024] A battery cell diagnostic device and its operating method according to one embodiment disclosed herein can improve the accuracy of battery cell life calculation based on an image corresponding to the operating characteristics of the battery cell and the operating conditions of the battery cell. In addition, this document can provide various effects that can be understood directly or indirectly. [Brief explanation of the drawing]

[0025] [Figure 1] This is a block diagram showing a battery cell diagnostic device according to one embodiment disclosed in this document. [Figure 2] This document describes a method for processing time-series data of a battery cell diagnostic device according to one embodiment disclosed in this document. [Figure 3a] This document is intended to explain the image conversion method of a battery cell diagnostic device according to one embodiment disclosed in this document. [Figure 3b] This document is intended to explain the image conversion method of a battery cell diagnostic device according to one embodiment disclosed in this document. [Figure 4] This document describes a method for extracting characteristic values ​​from a battery cell diagnostic device according to one embodiment disclosed in this document. [Figure 5] This document shows a method for calculating the State of Health (SOH) of a battery cell diagnostic device according to one embodiment disclosed in this document. [Figure 6] This document shows a method for calculating the State of Health (SOH) of a battery cell diagnostic device according to another embodiment disclosed in this document. [Figure 7] This is a flowchart showing the operation method of a battery cell diagnostic device according to one embodiment disclosed in this document. [Figure 8] This is a flowchart showing the operation method of a battery cell diagnostic device according to another embodiment disclosed in this document. [Figure 9]This graph compares the lifespan prediction accuracy based on the operating method of the diagnostic device according to one embodiment disclosed in this document. [Figure 10] This is a block diagram showing the hardware configuration of a computing system for performing the operation method of a battery cell diagnostic device according to one embodiment disclosed in this document. [Modes for carrying out the invention]

[0026] The embodiments disclosed in this document will be described in detail below with reference to illustrative drawings. It should be noted that, when assigning reference numerals to components in each drawing, the same reference numerals will be used for the same components whenever possible when they appear in other drawings. Furthermore, when describing the embodiments disclosed in this document, if a specific description of a related known configuration or function is deemed to hinder understanding of the embodiments disclosed in this document, such detailed description will be omitted.

[0027] In describing the components of the embodiments disclosed herein, terms such as First, Second, A, B, (a), (b), etc., may be used. Such terms are merely for distinguishing a component from other components and do not limit the nature, order, or procedure of that component. Furthermore, unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as those generally understood by a person of ordinary skill in the art to which the embodiments disclosed herein belong. Terms as defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and not as an ideal or overly formal meaning unless explicitly defined in this application.

[0028] Figure 1 is a block diagram showing a battery cell diagnostic device according to one embodiment disclosed in this document. Referring to Figure 1, the battery cell diagnostic device 10 according to one embodiment disclosed in this document may include a collection unit 100, a conversion unit 200, an extraction unit 300, and a calculation unit 400.

[0029] The data collection unit 100 can acquire the operating conditions of the battery cells and time-series data of the battery cells corresponding to those operating conditions. State of Health (SOH) is an indicator of the performance degradation of a battery cell and can be expressed as a percentage of the battery cell's capacity or resistance that has decreased relative to its initial capacity due to aging. Changes in SOH can occur due to factors such as increased resistance caused by polarization of chemical substances contained in the battery cell. When the performance of a battery cell falls below a predetermined critical point, replacement of the battery cell may be required. Generally, a level of 80% of the initial capacity can be considered the end of the battery cell's lifespan.

[0030] The operating conditions may be those that are kept constant during charge-discharge experiments to measure the performance index of the battery cell, in order to acquire the operating characteristics. Furthermore, the operating characteristics of the battery cell are the result values ​​obtained during the charge-discharge experiment under the aforementioned operating conditions, and may be values ​​that change over time. The time-series data may be data obtained by measuring the operating characteristics of the battery cell over a predetermined time interval. Therefore, the time-series data can show the change in the operating characteristics of the battery cell over time.

[0031] The performance index of a battery cell can, for example, refer to the aforementioned SOH or SOC (state of charge, remaining capacity). A charge-discharge experiment can refer to an experiment that maintains operating conditions and detects the operating characteristics of a battery cell. Operating conditions can refer to the experimental environment that affects the output of the battery cell.

[0032] When the operating temperature and charge / discharge current of the battery cell are used as operating conditions, the operating characteristics of the battery cell may also be the charge / discharge voltage. In this case, the time-series data may be data showing the change in the charge / discharge voltage of the battery cell over time.

[0033] The characteristics of a battery cell may deteriorate as the number of charge-discharge cycles increases. The data collection unit 100 can acquire time-series data of the battery cell for each charge-discharge cycle. According to this embodiment, the number of charge-discharge cycles can be an operating condition for the battery cell.

[0034] The operating conditions can be changed according to the time-series data to be acquired, and according to other embodiments, the operating characteristics can be the impedance, internal resistance, or capacitance of the battery cell.

[0035] According to one embodiment, the data collection unit 100 may include a measuring device that monitors the temperature, charge / discharge current, and charge / discharge voltage of the battery cell so that time-series data can be obtained based on operating conditions. The measuring device may also monitor the impedance, internal resistance, and capacity of the battery cell.

[0036] Furthermore, according to another embodiment, the collection unit 100 can match the operating conditions of the battery cell with the time-series data of the battery cell corresponding to the operating conditions and receive it from an external source.

[0037] The time-series data may be data sampled or normalized at predetermined time intervals by the acquisition unit 100. For example, when the operating temperature and charge / discharge current of the battery cell are kept constant, the acquisition unit 100 collects raw data for the charge / discharge voltage of the battery cell over time. By dividing this raw data at predetermined time intervals, multiple time-series data can be obtained.

[0038] The conversion unit 200 can convert time-series data of a battery cell into an image corresponding to a spatial trajectory of a predetermined dimension. The conversion of time-series data can be performed using the RP (recurrence plot) method. The RP method may be a conversion technique used to convert time-series data into an image.

[0039] The RP method may also be a technique that converts time-series data into a spatial trajectory of a predetermined dimension, represents the distance between points located on the spatial trajectory with a distance matrix, and converts the time-series data into an image. In this case, the spatial trajectory may represent the movement trajectory between data points included in the time-series data.

[0040] The transformation unit 200 can visualize time-series data into an image corresponding to a spatial trajectory of a predetermined dimension, based on the distance values ​​of the distance matrix. According to one embodiment, the conversion unit 200 can visualize time-series data as an image by representing the region corresponding to the distance matrix in two dimensions when the distance value of the distance matrix is ​​greater than or equal to a threshold.

[0041] According to another embodiment, the color of the region corresponding to the distance matrix can be represented in two dimensions such that it changes according to the distance value of the distance matrix, thereby visualizing an image that includes red, green, and blue channels. The conversion unit 200 converts time-series data into an image using the RP method, thereby clearly representing features that are difficult to grasp from time-series data alone.

[0042] The extraction unit 300 can extract feature values ​​from an image via a first neural network. According to this embodiment, the extraction unit 300 can generate multiple feature maps from the image. The extraction unit 300 can process the generated feature maps to extract feature values ​​and transmit the extracted feature values ​​to the calculation unit 400.

[0043] The first neural network may be a Convolutional Neural Network algorithm. A Convolutional Neural Network is a deep learning algorithm that includes a Convolutional Layer and a Pooling Layer. The Convolutional Layer may include a filter (Convolutional Kernel) that extracts features from an image, and an Activation Function that converts the filtered input data into non-linear values. The image processed by the Convolutional Layer can be output as multiple feature maps. A pooling layer can remove noise unrelated to the features from the feature map.

[0044] Convolutional neural networks can have different architectures depending on how the convolutional and pooling layers are connected. Examples of convolutional neural networks used to extract features from images include AlexNet, which uses superimposed convolutional layers; GoogLeNet, which includes an inception module; and ResNet, which includes a residual block.

[0045] The first neural network used in the extraction unit 300 can utilize a previously trained model. According to one embodiment, the extraction unit 300 can use a GoogLeNet or ResNet model trained on a visual database (e.g., the imageNet project) to extract feature values ​​by transfer learning.

[0046] In another embodiment, the extraction unit 300 can optimize the first neural network by adjusting the weights of the convolutional kernels used in the convolutional layer of the first neural network, or by adjusting the structure of the learned convolutional neural network.

[0047] The first neural network used in the extraction unit 300 may vary depending on the amount of images converted from the time-series data. In other words, when the amount of images converted from the time-series data is small, a previously trained model can be used for feature extraction through transfer learning, and when the amount of images converted from the time-series data is large, the structure and weights of a previously trained model can be modified and utilized.

[0048] According to the embodiment, the extraction unit 300 can generate feature values ​​by flattening and concatenating the multiple feature maps that have been generated. The number of feature maps may vary depending on the number of filters included in the first neural network. The feature maps may be referred to as layers.

[0049] The arithmetic unit 400 can predict or calculate the State of Health (SOH) of the battery cell via a second neural network. According to the embodiment, the second neural network may be a deep neural network algorithm. The arithmetic unit 400 can predict or calculate the SOH of the battery cell based on feature values ​​and operating conditions extracted from the image.

[0050] Deep neural networks are deep learning algorithms that include multiple hidden layers. By using multiple hidden layers, they can sophisticatedly process multiple complex datasets.

[0051] The second neural network included in the calculation unit 400 can improve accuracy compared to directly learning time-series data and calculating the SOH of a battery cell by learning based on feature values ​​extracted from converting time-series data into an image.

[0052] When time-series data is trained without image conversion, the characteristics of the data may not be properly reflected, leading to biased predicted values ​​or excessive reflection of local regional trends, which can reduce the calculation accuracy of the calculation unit 400.

[0053] For example, in the case of time-series data that reflects the change in battery cell capacity over time, noise may be generated in the time-series data due to phenomena such as capacity regeneration, where the capacity increases irregularly as residual reaction products are removed, and this noise may affect the prediction of State of Health (SOH).

[0054] When time-series data is converted into an image and SOH is predicted based on feature values ​​extracted from the image, the prediction accuracy of SOH by the calculation unit 400 can be increased by ensuring that the image recursively reflects changes in the time-series data and facilitates feature extraction.

[0055] Deep neural networks can have different structures depending on how their hidden layers are connected. Examples of deep neural networks include Recurrent Neural Networks (RNNs) with a circular hidden layer structure, Long Short-Term Memory (LSTMs) which utilize the vanishing gradient phenomenon and add layers to store previous states to improve long-term dependencies.

[0056] The feature values ​​input to the arithmetic unit 400 are derived from time-series data, and therefore can be analyzed by the second neural network as if they were sequential data. Sequential data can refer to data that changes over time.

[0057] In another embodiment, the calculation unit 400 can generate a prediction function using operating conditions corresponding to feature values. The calculation unit 400 can concatenate feature values ​​and operating conditions corresponding to those feature values.

[0058] The prediction function is a function that allows the second neural network to reflect changes in the State of Health (SOH) due to operating conditions. By learning by linking the prediction function and feature values, the second neural network can reflect the trend of SOH changes due to operating conditions. Therefore, when the second neural network learns by linking the prediction function and feature values, the SOH prediction accuracy of the calculation unit 400 can be improved.

[0059] The extraction unit 300 extracts feature values ​​from an image obtained by converting time-series data, so any operating conditions can correspond to each feature value. The collection unit 100 can transmit the operating conditions corresponding to the feature values ​​to the calculation unit 400. The calculation unit 400 can generate a prediction function via a third neural network.

[0060] According to the embodiment, the third neural network may be a deep neural network, such as the second neural network. Learning using a second and third neural network involves learning with information of different dimensions (feature values ​​and operating conditions), and can be called multimodal learning.

[0061] Figure 2 shows a method for processing time-series data of a battery cell diagnostic device according to one embodiment disclosed in this document. Figure 2 shows the time-series data under arbitrary operating conditions (X) as the first graph (G1).

[0062] The collection unit 100 is set to a preset temperature (t s ), preset charge / discharge current (i s ), and the number of preset charge / discharge cycles (C s When the operating condition (X) is set to ), the charge / discharge voltage (V) of the battery cell can be collected according to the time (T). In other words, the first graph (G1) can show the charge / discharge voltage (V) of the battery cell according to the time (T).

[0063] The data collection unit 100 can collect multiple time-series data by dividing the charge / discharge voltage (V) corresponding to the time (T) of the battery cell into preset time intervals. Figure 2 shows an example of when the data collection unit 100 collects time-series data (D) divided from the first time (T1) to the second time (T2).

[0064] The collection unit 100 can collect time-series data by dividing it into arbitrary time intervals, and according to other embodiments, it can include time regions in which multiple time-series data overlap each other.

[0065] Figures 3a and 3b illustrate the image conversion method of a battery cell diagnostic device according to one embodiment disclosed in this document. The conversion unit 200 can convert time-series data (D) into an image (I) using the RP method.

[0066] To illustrate the RP method, an example graph (GX) with a time axis (T) and a voltage axis (V) is shown. Below, Figure 3a illustrates in detail the process by which the example graph (GX) is transformed into an example image (IX).

[0067] The example graph (GX) can be converted into an example spatial trajectory (PX), which is a two-dimensional spatial trajectory. Points (S1, S2, S3) on the example spatial trajectory can represent the change in voltage axis (V) value due to a change in the time axis (T) value of the example graph (GX).

[0068] In other words, S1 may be a coordinate representation (0, 1) of the change in the voltage axis (V) value from 0 to 1 while the time axis (T) value changes from 1 to 2. Similarly, S2 may be a coordinate representation (1, 2) of the change in the voltage axis (V) value from 1 to 2 while the time axis (T) value changes from 2 to 3.

[0069] A spatial trajectory can represent the movement trajectory of data contained in time-series data. Therefore, points located on a spatial trajectory can recursively represent the relationships between adjacent data contained in the time-series data.

[0070] The distance matrix may, for example, be a matrix composed of points located on a spatial trajectory. For example, the distance matrix may take the form of (S1, S2) or (S1, S3).

[0071] The distance values ​​in a distance matrix can represent the distance between points on a spatial trajectory. According to one embodiment, time-series data can be visualized as an image by representing the region corresponding to the distance matrix in a way that changes color according to the distance values ​​in the distance matrix. For example, the distance value of the distance matrix (S1, S2) can be √2, which is the distance between S1 and S2. Similarly, the distance value of the distance matrix (S1, S3) can be 2, which is the distance between S1 and S3.

[0072] Referring to the example image (IX), the brightness of the corresponding region can be changed according to the distance value of the distance matrix. Since the RP method represents the distance between points located on a spatial trajectory using a distance matrix, the image represented by the RP method can have a symmetrical shape.

[0073] Figure 3b illustrates the process by which the time-series data (D) in Figure 2 is converted into image data (I) using the RP method. The image (I) converted by the RP method can more clearly represent the characteristics compared to the first graph (G1) which shows the time-series data.

[0074] Figure 4 shows a method for extracting characteristic values ​​from a battery cell diagnostic device according to one embodiment disclosed in this document. The extraction unit 300 receives an image (I) and can extract feature values ​​from the received image (I) via the first neural network (NN1).

[0075] The first neural network (NN1) may include convolutional layers (CL) and pooling layers (PL). A convolutional layer (CL) is a layer that includes a filter and an active function, and can generate a feature map from an image (I). A pooling layer (PL) can remove noise from the extracted feature map and adjust the size of the feature map.

[0076] For the sake of explanation, the first neural network (NN1) is shown as containing one convolutional layer (CL) and one pooling layer (PL). However, the first neural network (NN1) can contain multiple convolutional layers (CL) and pooling layers (PL), and these convolutional layers (CL) and pooling layers (PL) can have various structures depending on the purpose of the first neural network (NN1).

[0077] The number of feature maps generated from an image (I) can vary depending on the number of filters included in the convolutional layer (CL). In other words, a single image may have multiple feature maps output after passing through the convolutional layer (CL) and the pooling layer (PL), and these feature maps can be referred to as layers (L1, L2, L3, L4, L5~Ln).

[0078] The extraction unit 300 can extract feature values ​​(F) by flattening and concatenating the feature map generated from the image (I). The generated feature values ​​can be used to represent time series data (D).

[0079] Figure 5 shows a method for calculating the State of Health (SOH) of a battery cell diagnostic device according to one embodiment disclosed in this document. The computation unit 400 receives feature values ​​(F), and the second neural network (NN2) can learn based on the received feature values ​​(F). The learned second neural network (NN2) can then compute or predict the State of Health (SOH) based on the feature values ​​(F). The second neural network (NN2) may include a hidden layer (HL), and although it is shown exemplary as having a cyclic structure, according to the embodiment, a neural network with a different structure can be selected. The hidden layer (HL) can be learned via a recurrence algorithm and can predict or compute the SOH.

[0080] Figure 6 shows a method for calculating the State of Health (SOH) of a battery cell diagnostic device according to another embodiment disclosed in this document. The calculation unit 400 receives operating conditions (X) from the collection unit 100 and can calculate a prediction function (PF) via a third neural network (NN3). The third neural network (NN3) may be an RNN, for example, and the calculated prediction function (PF) can correspond to any feature value.

[0081] The calculation unit 400 can concatenate the feature value (F) with the prediction function (PF). For example, the feature value (F) extracted in Figure 2 is the operating condition X(t s i s , C s Because it is based on time series data (D) collected by ), it can be linked to a prediction function (PF) obtained from operating conditions (X).

[0082] The second neural network (NN2) included in the arithmetic unit 400 can learn based on concatenated feature values ​​and prediction functions (F, PF), and can calculate the State of Health (SOH) based on the concatenated feature values ​​and prediction functions (F, PF). By using the prediction function (PF) in its learning, the second neural network (NN2) can reflect changes in SOH due to operating conditions.

[0083] Figure 7 is a flowchart showing the operation method of a battery cell diagnostic device according to one embodiment disclosed in this document. Referring to Figure 7, the battery cell diagnostic device can collect time-series data based on the operating conditions of the battery cell (S100). As mentioned above, the operating conditions of the battery cell may refer to conditions that are maintained constant during the charging and discharging operation of the battery cell. The collected time-series data may also be data that shows the operating characteristics, which are result values ​​obtained during a charging and discharging experiment based on the operating conditions of the battery cell, over time.

[0084] For example, the operating conditions may include the charge / discharge current of the battery cell, the temperature during charge / discharge of the battery cell, the charge / discharge voltage of the battery cell, and the number of charge / discharge cycles of the battery cell. Time-series data may be data obtained by sampling operating characteristics at predetermined time intervals.

[0085] The battery cell diagnostic device can convert time-series data into an image corresponding to a spatial trajectory of a predetermined dimension (S200). The battery cell diagnostic device can convert time-series data into an image using the RP method. The RP method may be a method of imaging time-series data by converting time-series data into a spatial trajectory of a predetermined dimension and representing the distance between points located on the converted spatial trajectory with a distance matrix. According to the embodiment, the predetermined dimension may be two-dimensional.

[0086] The battery cell diagnostic device can extract feature values ​​from the image (S300). The battery cell diagnostic device may use a first neural network to extract feature values, and exemplary the first neural network may be a convolutional neural network including convolutional layers and pooling layers. More specifically, the battery cell diagnostic device may use AlexNet, GoogLeNet, or ResNet, etc., to extract feature values ​​from the image.

[0087] The battery cell diagnostic device can calculate the lifespan of a battery cell based on the extracted feature values ​​(S400). The battery cell diagnostic device can train a second neural network based on the feature values ​​and use the trained second neural network to predict the State of Health (SOH) of the battery cell. The second neural network may be a deep neural network including hidden layers, and more specifically, the battery cell diagnostic device can use an RNN or LSTM to predict the SOH.

[0088] Figure 8 is a flowchart showing the operation method of a battery cell diagnostic device according to another embodiment disclosed in this document. Figure 8 illustrates how the battery cell diagnostic device predicts the State of Health (SOH) of a battery cell by reflecting a prediction function.

[0089] The battery cell diagnostic device can predict the State of Health (SOH) using a method different from the one described in Figure 7 (S400') after extracting feature values ​​from the image (S300).

[0090] The battery cell diagnostic device can generate a prediction function for operating conditions based on a third neural network (S410). The prediction function is a value generated by calculating the operating conditions via a third neural network. The battery cell diagnostic device concatenates the prediction function with the corresponding feature value (S420) and can predict the battery cell life based on the prediction function and feature value (S430). The battery cell diagnostic device can improve the prediction accuracy of the SOH of the second neural network by learning the second neural network by concatenating the prediction function and feature value.

[0091] Figure 9 is a graph comparing the life prediction accuracy based on the operation method of the diagnostic device according to one embodiment disclosed in this document. The True graph in Figure 9 plots the SOH value calculated from multiple operating conditions and time-series data with a maximum value of 1, while the Predict graph plots the SOH value predicted by the battery cell diagnostic device based on the aforementioned operating conditions and time-series data, also with a maximum value of 1. The horizontal axis of the graph may represent the number of samples used for prediction.

[0092] Figures 9A and 9B are graphs showing the prediction of SOH based on a model trained on time-series data without image transformation. However, B is a graph of a model trained to reflect a prediction function based on operating conditions. Model B can achieve higher prediction accuracy compared to model A.

[0093] Figures 9C and 9D show graphs predicting State of Health (SOH) based on models trained by image transformation of time-series data. However, graph D reflects a prediction function based on operating conditions. It can be seen that the prediction accuracy of models trained after image transformation (C, D) is improved compared to models trained without image transformation (A, B). Furthermore, prediction accuracy can be further improved by incorporating a prediction function into the training.

[0094] Figure 10 is a block diagram showing the hardware configuration of a computing system for performing the operation method of a battery cell diagnostic device according to one embodiment disclosed in this document.

[0095] Referring to Figure 10, the computing system 1000 according to one embodiment disclosed in this document may include an MCU 1010, a memory 1020, an input / output interface 1030, and a communication interface 1040.

[0096] The MCU1010 may be a processor that executes various programs stored in memory 1020 (for example, a battery pack voltage or current acquisition program, a relay control program included in the battery pack, a battery cell remaining life calculation program, a battery cell capacity degradation diagnosis program, a battery cell resistance degradation judgment program, etc.), processes various information including battery cell remaining life information, battery cell capacity degradation information, and battery cell resistance degradation information through such programs, and operates the battery cell diagnostic device shown in Figure 1 above.

[0097] Memory 1020 can store various programs related to the collection and diagnosis of battery log information. Memory 1020 can also store various information such as battery current, voltage, charge / discharge condition information, battery cell voltage information during charge / discharge cycle intervals within a set number of cycles, and battery cell dQ / dV information during charge / discharge cycle intervals within a set number of cycles.

[0098] Multiple such memory 1020s may be provided as needed. Memory 1020 may be volatile memory or non-volatile memory. As volatile memory, RAM, DRAM, SRAM, etc., can be used for memory 1020. As non-volatile memory, ROM, PROM, EAROM, EPROM, EEPROM, flash memory, etc., can be used for memory 1020. The examples of memory 1020 listed above are merely illustrative and are not limiting.

[0099] The input / output interface 1030 can provide an interface that connects input devices (not shown), such as keyboards, mice, and touch panels, with output devices (not shown), such as displays, and the MCU 1010, enabling data transmission and reception.

[0100] The communication interface 1040 is configured to send and receive various data with a server and may be various devices that support wired or wireless communication. For example, a battery cell diagnostic device can send and receive relay control programs contained in a battery pack, current, current, dQ / dV value, charge / discharge condition information, remaining life, capacity degradation, or resistance degradation information of various battery packs from an external server separately provided via the communication interface 1040.

[0101] Thus, the computer program according to one embodiment disclosed in this document may be recorded in memory 1020 and processed by MCU 1010 to be realized as a module that performs, for example, the functions shown in Figure 1.

[0102] The above description is merely illustrative of the technical concept disclosed in this document, and any person with ordinary skill in the art to which the embodiments disclosed in this document belong can make various modifications and variations without departing from the essential characteristics of the embodiments disclosed in this document.

[0103] Therefore, the embodiments disclosed herein are for illustrative purposes only, not to limit, the technical ideas disclosed herein, and such embodiments do not limit the scope of the technical ideas disclosed herein. The scope of protection for the technical ideas disclosed herein shall be interpreted in accordance with the claims described below, and all technical ideas within an equivalent scope shall be interpreted as being included in the scope of rights of this document.

Claims

1. A collection unit that collects time-series data of operating characteristics, which are one of the charge / discharge voltage, impedance, internal resistance, or capacity of a battery cell, obtained by a charge / discharge experiment under operating conditions including at least one of the operating temperature, charge / discharge current, and charge / discharge cycles of the battery cell. A transformation unit that generates a two-dimensional spatial trajectory composed of points representing the changes in the time-series data by the RP method, and converts a distance matrix composed of points located on the spatial trajectory into an image, wherein the transformation unit assigns the distance matrix to each pixel of the image and changes the brightness or color of each pixel of the image according to the distance value of the assigned distance matrix, An extraction unit that, via a first neural network which is a convolutional neural network including a convolutional layer, generates a feature map, which is two-dimensional data in which local patterns contained in the image are emphasized by performing a filter operation on the image in the convolutional layer, and extracts feature values, which are one-dimensional continuous data, by flattening and concatenating the feature map. A calculation unit that calculates the lifespan of the battery cell by processing the feature values ​​in the hidden layer using a recursive algorithm via a second neural network including a hidden layer having a cyclic structure, A battery cell diagnostic device, including a battery cell diagnostic device.

2. The battery cell diagnostic device according to claim 1, wherein the first neural network further includes a pooling layer for denoising and resizing the feature map.

3. The battery cell diagnostic device according to claim 1, wherein the second neural network is a deep neural network including multiple hidden layers.

4. The battery cell diagnostic device according to any one of claims 1 to 3, wherein the calculation unit generates a prediction function that reflects the trend of changes in the lifespan of the battery cell due to the operating conditions by recursively processing the operating conditions corresponding to the time series data corresponding to the feature values ​​in the hidden layer via a third neural network including a hidden layer having a cyclic structure, and causes the second neural network to learn based on the concatenated prediction function and the feature values.

5. The battery cell diagnostic device according to claim 4, wherein the third neural network is a deep neural network including multiple hidden layers.

6. A step of collecting time-series data of operating characteristics, which are one of the charge-discharge voltage, impedance, internal resistance, or capacity of the battery cell, obtained by a charge-discharge experiment under operating conditions including at least one of the operating temperature, charge-discharge current, and charge-discharge cycles of the battery cell, A step of generating a two-dimensional spatial trajectory composed of points representing the changes in the time-series data by the RP method, and converting a distance matrix composed of points located on the spatial trajectory into an image, wherein the distance matrix is ​​assigned to each pixel of the image, and the brightness or color of each pixel of the image is changed according to the distance value of the assigned distance matrix, The first neural network, which is a convolutional neural network including a convolutional layer, performs a filter operation on the image in the convolutional layer to generate a feature map, which is two-dimensional data in which local patterns contained in the image are emphasized, and the feature map is flattened and concatenated to extract feature values, which are one-dimensional continuous data. The steps include: calculating the lifespan of the battery cell by processing the feature values ​​in the hidden layer using a recursive algorithm via a second neural network including a hidden layer having a cyclic structure; A battery cell diagnostic method, including the following.

7. The battery cell diagnostic method according to claim 6, wherein the second neural network is a deep neural network including multiple hidden layers.

8. The step of calculating the lifespan is: The steps include: generating a predictive function that reflects the trend of changes in the battery cell's lifespan due to the operating conditions by recursively processing the operating conditions corresponding to the time-series data corresponding to the feature values ​​using a third neural network including a hidden layer having a cyclic structure; The steps include concatenating the prediction function and the feature values, A battery cell diagnostic method according to claim 6 or 7, comprising the step of training the second neural network based on the prediction function and the feature values.

Images