Defect type classification system and defect type classification method

Abstract

The failure type classification system includes a measurement unit connected to a battery for which a failure type is to be detected and measures at least one electrical characteristic of the battery for a predetermined inspection period to generate measurement data, a conversion unit converts the measurement data to generate input data, and a failure type prediction module that performs machine learning on learning data and determines a failure type of the battery based on the input data. The input data is data suitable for an input node of the failure type prediction module, and the failure type is classified according to a cause of a short circuit failure of the battery.

Description

[Technical field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of priority based on Korean Patent Application No. 10-2021-0105497 dated August 10, 2021, and all contents disclosed in the documents of that Korean patent application are incorporated herein by reference.

[0002] The present disclosure relates to a defect type classification system and a defect type classification method. [Background technology]

[0003] In the multiple processes for producing batteries, a check for the presence or absence of defects is performed for each production process. For example, factors required for checking the presence or absence of defects are specified for each production process, and a measuring device capable of measuring the factors is used to detect the presence or absence of a battery defect. In this way, when a battery defect is detected using an existing method using a measuring device, only the presence or absence of a defect in the measurement section where the measurement was performed and the measured values ​​such as the battery voltage and current at the time of measurement are known. The cause of the detected battery defect cannot be known. In order to know the cause of the defect, an operator must directly disassemble and analyze the defective battery to understand the cause. Summary of the Invention [Problem to be solved by the invention]

[0004] It is an object of the present invention to provide a system and method capable of classifying battery defect types. [Means for solving the problem]

[0005] The failure type classification system includes a measurement unit connected to a battery for which a failure type is to be detected and measures at least one electrical characteristic of the battery for a predetermined inspection period to generate measurement data, a conversion unit converts the measurement data to generate input data, and a failure type prediction module that performs machine learning on learning data and determines a failure type of the battery based on the input data. The input data is data suitable for an input node of the failure type prediction module, and the failure type is classified according to a cause of a short circuit failure of the battery.

[0006] The measurement unit may measure a battery voltage, which is a voltage across the battery, over the test period and store the measurement data indicative of a measurement result.

[0007] The conversion unit may analyze the measurement data and extract at least one of a maximum value, a minimum value, an average value, and a median value of the battery voltage, and a peak number of the battery voltage, to generate the input data.

[0008] The measurement data is based on the results of measuring the battery voltage, which is the voltage across the battery, over a predetermined inspection period, and the input data may include at least one of a maximum value, a minimum value, an average value, and a median value of the battery voltage, and a number of peaks of the battery voltage.

[0009] The failure type classification system may further include a database for storing the learning data, and the database may store input data and a failure type based on a voltage across a battery when a short circuit failure occurs in the battery.

[0010] The defect type prediction module may calculate a deviation, which is a distance difference between at least one of electrical characteristics measured for a target battery that is a target for classifying a defect type, and each of a plurality of values ​​belonging to the learning data, derive a predetermined number of reference data in ascending order of the calculated deviations, and classify the defect type of the target battery into the class that is the largest as a result of counting the classes to which each of the derived reference data belongs.

[0011] The defect type prediction module may receive at least one of the electrical characteristics from the learning data, and may learn to predict a defect type corresponding to one of the input electrical characteristics as one of a plurality of classes.

[0012] According to another aspect of the present invention, a failure type classification method includes a step of: a failure type prediction module learning to classify a failure type according to a cause of a short circuit failure of a battery based on learning data; a measurement unit measuring at least one of electrical characteristics of the battery for a predetermined inspection period to generate measurement data; a conversion unit converting the measurement data to generate input data; and the failure type prediction module determining a failure type of the battery based on the input data, and the input data may be data suitable for an input node of the failure type prediction module.

[0013] The step of generating the measurement data may include measuring a battery voltage, which is a voltage across the battery, over the test period and generating the measurement data indicative of a measurement result.

[0014] The step of generating the input data may include analyzing the measurement data to extract at least one of a maximum value, a minimum value, an average value, and a median value of the battery voltage, and a number of peaks of the battery voltage, to generate the input data.

[0015] The measurement data is based on the results of measuring the battery voltage, which is the voltage across the battery, over a predetermined inspection period, and the input data may include at least one of a maximum value, a minimum value, an average value, and a median value of the battery voltage, and a number of peaks of the battery voltage.

[0016] The failure type classification method may further include storing input data and a failure type based on a voltage across the battery when a short circuit failure occurs in the battery, as the learning data, in a database.

[0017] The step of determining the defect type may include a step of calculating a deviation, which is a distance difference between at least one of electrical characteristics measured for a target battery that is a target for classifying a defect type, and each of a plurality of values ​​belonging to the learning data; deriving a predetermined number of reference data in ascending order of the calculated deviations; and classifying the defect type of the target battery into the class that is the largest as a result of counting the classes to which each of the derived reference data belongs.

[0018] The step of learning to classify the defect type may include a step of inputting at least one of the electrical characteristics from the learning data, and learning to predict a defect type corresponding to one of the input electrical characteristics as one of a plurality of classes. Effect of the Invention

[0019] The present invention provides a system and method for classifying defects that can detect causes of defects in batteries that occur during the production process.

[0020] There is a difference in the skill level of battery disassembly and the accuracy of analyzing the defect to derive the cause between operators who detect the cause of the defect. The present invention increases the accuracy of the defect type analysis and can maintain a certain level of accuracy without deviation in accuracy. In addition, the time required for operators to analyze the defect type can be significantly reduced. [Brief description of the drawings]

[0021] [Figure 1] 1 is a block diagram showing a defect type classification system according to an embodiment; [Diagram 2] 11 is a graph showing a method for generating input data from measurement data obtained by measuring a battery voltage according to one embodiment. [Diagram 3] 11 is a diagram illustrating an operation of a defect type prediction module according to an embodiment; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0022] The present invention includes two stages for effectively learning the feature values ​​of a defective battery. The first machine learning stage is a stage in which the defect type prediction module learns the feature values ​​of a defective battery by machine learning, and the second classification stage is a stage in which the machine-learned defect type prediction module determines the cause of a defect in a battery in which a defect has occurred. In the first machine learning stage, the defect type prediction module learns the values ​​of electrical features such as the voltage and current of the measured defective battery and the cause of the defect in the defective battery. In the second classification stage, after measuring the values ​​of the electrical features of the battery in which a defect has occurred, the measured feature values ​​are input to the defect type prediction module, and the defect type prediction module can determine the cause of the defect based on the input feature values.

[0023] Hereinafter, the embodiments disclosed in the present specification will be described in detail with reference to the attached drawings, and the same or similar components are denoted by the same or similar reference numerals, and the duplicated description thereof will be omitted. The suffixes "module" and / or "part" for components used in the following description are given or mixed for the sake of ease of specification preparation only, and do not have any meaning or role that is different from each other. In addition, when describing the embodiments disclosed in the present specification, if it is determined that a detailed description of such known technology may obscure the gist of the embodiments disclosed in the present specification, the detailed description will be omitted. In addition, the attached drawings are merely for the purpose of making the embodiments disclosed in the present specification easily understandable, and it should be understood that the attached drawings do not limit the technical ideas disclosed in the present specification, and include all modifications, equivalents, or alternatives included in the idea and technical scope of the present invention.

[0024] Terms including ordinal numbers such as first, second, etc. may be used to describe various components, but the components are not limited by the terms. The terms are used only to distinguish one component from another.

[0025] When a component is referred to as being "coupled" or "connected" to another component, it should be understood that it may be directly coupled or connected to the other component, but that there may be other components in between. In contrast, when a component is referred to as being "directly coupled" or "directly connected" to another component, it should be understood that there are no other components in between.

[0026] It should be understood that in this application, the terms "comprise" or "have" are intended to specify the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, but do not preclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0027] FIG. 1 is a block diagram illustrating a defect type classification system according to an embodiment.

[0028] As shown in FIG. 1, the defect type classification system 1 includes a measurement unit 10, a conversion unit 20, a defect type prediction module 30, and a database 40.

[0029] The measurement unit 10 is connected to the battery 2 for which a defect type is to be detected, measures at least one of the electrical characteristics of the battery 2 for a predetermined inspection period, generates measurement data based on the measurement results, and stores the measurement data. The measurement unit 10 may include a memory 11 for storing the measurement data. The memory 11 may be a non-volatile memory for storing the measurement data. When the battery for which a defect type is to be detected is replaced, the measurement unit 10 may delete the contents stored in the memory 11 and write new measurement data.

[0030] For example, the measurement unit 10 can measure the battery voltage, which is the voltage between the positive and negative electrodes of the battery 2, over an inspection period and store measurement data indicating the battery voltage measured over the inspection period in the memory 11. The present invention is not limited to using only the battery voltage to detect the type of defect. At least one or more of the battery voltage, the battery current flowing through the battery, the internal resistance of the battery, the inductance of the battery, and the capacitance of the battery can be used to detect the type of defect of the battery.

[0031] The conversion unit 20 converts the measurement data to generate input data suitable for the defect type prediction module 30. The input data is determined according to values ​​to be input to input nodes of the defect type prediction module 30. The conversion unit 20 reads the measurement data from the memory 11, analyzes the read measurement data, and extracts features of the analyzed measurement data to generate input data.

[0032] FIG. 2 is a graph showing a method for generating input data from measurement data obtained by measuring a battery voltage according to one embodiment.

[0033] FIG. 2 is a non-limiting example illustrating one of the various ways to generate input data.

[0034] As shown in FIG. 2, during the inspection time, the measurement unit 10 measures the voltage across the battery 2 to generate measurement data relating to the battery voltage, and stores the data in the memory 11.

[0035] The conversion unit 20 reads out the measurement data from the memory 11, analyzes the read out measurement data, and extracts the maximum, minimum, average, and median values ​​that are characteristics of the measurement data, as well as the number of peaks from the measurement data, to generate input data. In other words, the input data is the maximum, minimum, average, and median values ​​of the battery voltage during the inspection period, as well as the number of peaks in the battery voltage.

[0036] The defect type prediction module 30 performs machine learning on the learning data stored in the database 40. The database 40 classifies and stores input data for each defect type, and whenever a defect occurs in the battery, the input data and the defect type are updated in the database 40. At this time, the input data may include at least one of the maximum, minimum, average, and median values ​​of the battery voltage, and the number of peaks of the battery voltage.

[0037] The defect type prediction module 30 can perform machine learning to predict a defect type by inputting input data stored for each defect type into an input node. A cycle at which the defect type prediction module 30 performs machine learning can be determined according to the amount of learning data updated in the database 40. For example, every time a significant amount of learning data is updated in the database 40, the defect type prediction module 30 can perform machine learning including the updated learning data.

[0038] The defect type prediction module 30 may be implemented by a processor that executes a program that implements one of a supervised machine learning algorithm, an unsupervised machine learning algorithm, a semi-supervised machine learning algorithm, and a reinforced machine learning algorithm. When the defect type prediction module 30 is a processor that executes a program that implements a neural network algorithm, the defect type prediction module 30 may perform neural network learning so as to classify defect types using learning data to which a label is attached. The defect type prediction module 30 may repeatedly perform neural network learning to predict a defect type with the highest probability among a plurality of defect types that may occur in a classification target based on the learning data. The defect type prediction module 30 may adjust the activation function of each node of the neural network and the weighting between nodes so that the predicted defect type approaches an actual defect type through repeated neural network learning.

[0039] The failure type prediction module 30 constructed by such machine learning can determine a failure type based on input data. For example, when the input data is at least one of a maximum value, a minimum value, a median value, and an average value of a battery voltage and a peak number of the battery voltage, the failure type prediction module 30 can apply each element constituting the input data to a corresponding input node to determine one of a plurality of failure types. The number of input nodes can be determined according to the number of elements constituting the input data. The plurality of failure types are classified according to the cause of a short circuit failure of the battery, and can include, for example, a folded tab, a crack, a dent, a separator wrinkle, etc. The crack and the dent can mean that a part constituting a battery cell is cracked or dented, respectively.

[0040] FIG. 3 is a diagram illustrating an operation of the defect type prediction module according to an embodiment.

[0041] The defect type prediction module according to an embodiment may be implemented using a K-Nearest-Neighbor classification (KNN) model among machine learning models. In addition, in the classification method shown in FIG. 3, input data is the maximum value of the battery voltage and the number of peaks of the battery voltage.

[0042] In the machine learning stage, the defect type prediction module 30 is input with the maximum value of the battery voltage, which is the value on the X-axis, and the number of peaks of the battery voltage, which is the value on the Y-axis, and is trained to output one of "dent" and "crack" as the cause of the defect. As shown in Fig. 3, the defect type prediction module 30 can perform training for a battery (class A) whose cause of defect is "dent" represented by a star, and a battery (class B) whose cause of defect is "crack" represented by a triangle.

[0043] After completing machine learning, the defect type prediction module 30 receives the maximum battery voltage value (Xi) and the number of battery voltage peaks (Yi) corresponding to the positions indicated by squares in FIG. 3, and can predict the cause of the battery defect based on the input.

[0044] Specifically, the failure type prediction module 30 receives the maximum battery voltage value (Xi) and the number of peaks (Yi) of the battery voltage measured for a battery to be classified into a failure type (hereinafter, a target battery).

[0045] As shown in FIG. 3(b), the defect type prediction module 30 can calculate a distance difference (hereinafter, deviation) between a position (hereinafter, target position) defined by the maximum value (Xi) of the battery voltage and the number of peaks (Yi) of the battery voltage and each of all positions (hereinafter, comparison positions) corresponding to the learned data. The defect type prediction module 30 can derive a predetermined number of reference data in ascending order of a plurality of deviations between the target position and the comparison position. For example, the defect type prediction module 30 can calculate a first and a second distance difference on the X-axis and the Y-axis between any one of the target position and the comparison position, and calculate a deviation according to a result of square-rooting the value obtained by squaring and adding the first and second distance differences calculated. The defect type prediction module 30 can calculate a plurality of deviations between the target position and the comparison position in this manner. The defect type prediction module 30 can arrange all the calculated deviations in ascending order, derive a predetermined number of deviations in ascending order, and determine data of the comparison position corresponding to the derived deviations as the reference data. However, the method for calculating the deviation in the present invention is not limited to the above method.

[0046] As shown in (c) of Fig. 3, the defect type prediction module 30 can determine that the derived reference data in class A and class B are located within the dotted circle. The defect type prediction module 30 can count the classes to which each of the reference data belongs, and classify the defect type of the target battery into the class with the larger count result. In (c) of Fig. 3, the count result of class A is 1, and the count result of class B is 2, so the defect type of the target battery can be predicted to be class B (cracked).

[0047] In the example of FIG. 3, the dimensions for the defect type are determined by two axes, the X axis and the Y axis. However, the number of dimensions for the defect type can be increased by increasing the number of axes such as the X axis, the Y axis, the Z axis, etc. according to the number of input data.

[0048] By using the above method, the defect type prediction module 30 can classify the target battery into class B (cause of defect: crack) when the maximum battery voltage of the target battery is 37V and the number of battery peaks is 3. Alternatively, the defect type prediction module 30 can classify the target battery into class A (cause of defect: dent) when the maximum battery voltage of the target battery is 47V and the number of battery peaks is 5.

[0049] Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited to these embodiments, and various modifications and improvements made by those having ordinary skill in the art to which the present invention pertains also fall within the scope of the present invention. [Explanation of symbols]

[0050] 1. Defect classification system 2 Battery 10 Measuring part 11. Memory 20 Conversion unit 30 Defect type prediction module

Claims

1. a measurement unit connected to a battery for which a defect type is to be detected, for measuring a battery voltage across the battery for a predetermined inspection period to generate measurement data; a conversion unit that converts the measurement data to generate input data including at least one of a maximum value, a minimum value, an average value, and a median value of the battery voltage, and a number of peaks of the battery voltage; a failure type prediction module that performs machine learning on learning data and determines a failure type of the battery based on the input data; The input data is data suitable for an input node of the failure type prediction module, and the failure type is classified according to a cause of a short circuit failure of the battery. Defect classification system.

2. The defect classification system of claim 1 , wherein the measurement unit stores the measurement data indicating a measurement result.

3. The conversion unit is 3. The fault type classification system according to claim 2, wherein the input data is generated by analyzing the measurement data and extracting at least one of a maximum value, a minimum value, an average value, and a median value of the battery voltage, and a number of peaks of the battery voltage.

4. Further comprising a database for storing the learning data; The database includes: The fault type classification system according to claim 1 , further comprising: storing input data and a fault type based on a voltage across the battery when a short circuit fault occurs in the battery.

5. A measurement unit connected to a battery for which a defective type is to be detected, measuring at least one of the electrical characteristics of the battery over a predetermined inspection period to generate measurement data; A conversion unit that converts the measurement data to generate input data; a failure type prediction module that performs machine learning on learning data and determines a failure type of the battery based on the input data; the input data is data suitable for an input node of the failure type prediction module, and the failure type is classified according to a cause of a short circuit failure of the battery; The defect type prediction module includes: A defect type classification system which calculates a deviation which is a distance difference between at least one of electrical characteristics measured for a target battery which is the subject of classifying a defect type and each of a plurality of values ​​belonging to the learning data, derives a predetermined number of reference data in ascending order from the calculated deviations, and classifies the defect type of the target battery into the class which is the largest as a result of counting the classes to which each of the derived reference data belongs.

6. The defect type prediction module includes:

6. The defect type classification system according to claim 5, wherein at least one of the electrical characteristics is input from the learning data, and the system is trained to predict a defect type corresponding to one of the input electrical characteristics as one of a plurality of classes.

7. A failure type prediction module learns to classify a failure type according to a cause of a short circuit failure of the battery based on the learning data; A measurement unit measures a battery voltage, which is a voltage across a battery, for a predetermined inspection period to generate measurement data; A conversion unit converts the measurement data to generate input data including at least one of a maximum value, a minimum value, an average value, and a median value of the battery voltage, and a peak number of the battery voltage; the failure type prediction module determines a failure type of the battery based on the input data; The input data is data suitable for an input node of the defect type prediction module.

8. The step of generating the measurement data includes:

8. The method of claim 7, further comprising the step of generating said measurement data indicative of a measurement result.

9. The step of generating input data comprises:

9. The method of claim 8, further comprising: analyzing the measurement data to extract at least one of a maximum value, a minimum value, an average value, and a median value of the battery voltage, and a number of peaks of the battery voltage, to generate the input data.

10. 8. The method of claim 7, further comprising storing in a database, as the learning data, input data based on a voltage across the battery when a short circuit fault occurs in the battery and a fault type.

11. A step in which a failure type prediction module learns to classify a failure type according to a cause of a short circuit failure of a battery based on the learning data; A measurement unit measures at least one of the electrical characteristics of the battery over a predetermined test period to generate measurement data; A conversion unit converts the measurement data to generate input data; the failure type prediction module determines a failure type of the battery based on the input data; The input data is data suitable for an input node of the defect type prediction module; The step of determining the defect type includes: calculating a deviation between at least one of electrical characteristics measured for a target battery to be classified as a defective type and each of a plurality of values ​​belonging to the learning data; deriving a predetermined number of reference data in ascending order of the calculated deviations, and classifying a defect type of the target battery into a class with the largest result of counting classes to which each of the derived reference data belongs.

12. The step of learning to classify defect types includes:

12. The method of claim 11, further comprising: inputting at least one of the electrical characteristics from the learning data, and learning to predict a failure type corresponding to one of the input electrical characteristics as one of a plurality of classes.

Images