Battery chemistry-adaptive multivariate data-based battery state diagnosis apparatus and method

The battery chemical type adaptive diagnostic device addresses the challenge of varying lithium-ion battery types by using multivariate data to identify and predict the condition of batteries, enhancing diagnostic accuracy and reliability through adaptive analysis and continuous model retraining.

KR103043870B1Active Publication Date: 2026-11-06DIVE EDGE TECHNOLOGY CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
KR1020260117210
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2026-06-26
Publication Date
2026-11-06
Estimated Expiration
2046-06-26

AI Technical Summary

Technical Problem

Conventional battery condition diagnostic technologies often fail to accurately diagnose the state of lithium-ion batteries of different chemical types due to the use of uniform analysis methods, leading to reduced diagnostic accuracy and compromised reliability in anomaly detection and remaining lifespan predictions.

Method used

A battery chemical type adaptive multivariate data-based diagnostic device that identifies the chemical type of a battery using condition data, calculates derived variables specific to each type, and performs anomaly detection and remaining life prediction through multivariate degradation history time series data, incorporating adaptive thresholds and models for improved accuracy.

Benefits of technology

The device enhances diagnostic reliability and efficiency by accurately identifying battery chemical types and predicting remaining life, integrating anomaly detection and life prediction, and continuously improving performance through retraining with actual data post-maintenance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 112026078033036-PAT00003_ABST
    Figure 112026078033036-PAT00003_ABST
Patent Text Reader

Abstract

The present invention relates to a battery condition diagnostic device based on adaptive multivariate data for battery chemical types, wherein the device comprises: a data collection unit that collects condition data including voltage, temperature, and current from a battery; a chemical type identification control unit that identifies the chemical type of the battery by calculating the flatness ratio of the nominal voltage range or OCV-SOC (Open Circuit Voltage - State of Charge) curve of the battery based on the condition data and determines an analysis path corresponding to the identified chemical type; an adaptive derived variable calculation unit that calculates different derived variables predefined to reflect the characteristics of each chemical type according to the determined analysis path; and an anomaly grade determination unit that determines the anomaly grade of the battery by inputting multivariate input data for anomaly detection, including time series data of the derived variables, into an anomaly detection model and performing inference of anomalies. It includes a life prediction unit that generates multivariate degradation history time series data for life prediction representing the cycle-by-cycle degradation trend of the above-mentioned derived variables, and inputs the multivariate degradation history time series data for life prediction into a remaining life prediction model to calculate an inferred RUL (Remaining Useful Life) value.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present invention relates to battery diagnostic technology, and more specifically, to a battery chemical type adaptive multivariate data-based battery condition diagnostic device and method capable of diagnosing the condition of a battery by identifying the chemical type of the battery based on condition data collected from the battery, calculating derived variables reflecting the characteristics of each chemical type according to an analysis path corresponding to the identified chemical type, and inferring the grade of abnormal signs and remaining lifespan. Background Technology

[0003] With the recent expansion of lithium-ion battery applications, particularly in electric vehicles, energy storage systems, and industrial equipment, the importance of condition diagnostic technologies for ensuring battery safety and reducing maintenance costs is increasing. To accurately diagnose battery conditions, various technologies are being proposed that utilize state data—such as voltage, current, and temperature—to detect abnormal signs or predict remaining lifespan. In particular, active research is underway to estimate battery degradation and predict failures in advance using AI-based data analysis techniques.

[0004] Meanwhile, lithium-ion batteries include different chemical types, such as NMC (Nickel Manganese Cobalt) batteries and LFP (Lithium Iron Phosphate) batteries, depending on the composition of the cathode material, and voltage characteristics, temperature change characteristics, and degradation mechanisms differ depending on the chemical type. For example, in NMC batteries, cell voltage deviation and temperature change rate can be used as key indicators of the degradation state, whereas in LFP batteries, OCV-SOC characteristics appear flat, so there are limitations in accurately diagnosing the state based solely on voltage change, and the characteristics of the voltage change rate (dQ / dV) curve for unit capacity change that occurs during the charge-discharge process can be used as an important diagnostic indicator.

[0005] However, conventional battery condition diagnostic technologies are often developed for specific battery chemical types, which can lead to reduced diagnostic accuracy by applying the same analysis methods to batteries of different chemical types. Furthermore, the reliability of anomaly detection results or remaining lifespan predictions may be compromised because these technologies fail to adequately reflect the characteristics associated with different battery chemical types. In particular, in environments where various types of batteries are operated together, there is a need for integrated condition diagnostic technology capable of automatically applying diagnostic methods suitable for the specific characteristics of each chemical type.

[0006] Furthermore, when anomaly detection and remaining life prediction are performed independently, it is difficult to efficiently utilize degradation information derived from the same state data, and there are limitations in continuously improving the performance of diagnostic models using data accumulated during operation. Therefore, there is an increasing demand for battery condition diagnostic technology capable of automatically identifying battery chemical types and performing integrated anomaly detection and remaining life prediction using analysis paths corresponding to the identified chemical types.

[0007] Korean Registered Patent No. 10-2849365 relates to a battery abnormality diagnosis device and method, and may include a sensor unit for measuring the voltage and current of a battery cell, a differential data calculation unit for calculating differential data regarding the capacity and voltage of the battery cell, and an abnormality diagnosis unit for diagnosing an abnormality of the battery cell based on the differential data and classifying the type of the abnormality. Prior art literature

[0009] Korean Registered Patent No. 10-2849365 (August 19, 2025) The problem to be solved

[0010] One embodiment of the present invention aims to provide a battery chemical type adaptive multivariate data-based battery condition diagnosis device and method capable of identifying the chemical type of a battery based on condition data collected from a battery, and detecting abnormal signs using derived variables that reflect the characteristics of each chemical type according to an analysis path corresponding to the identified chemical type.

[0011] One embodiment of the present invention aims to provide a battery condition diagnosis device and method based on multivariate data adaptive to battery chemical types, which can accurately predict the condition of batteries of various chemical types by inferring the remaining life using multivariate degradation history time series data that reflects the characteristics of each chemical type of battery.

[0012] One embodiment of the present invention aims to provide a battery chemical type adaptive multivariate data-based battery condition diagnosis device and method that improve the reliability of battery condition diagnosis and efficiently manage the condition of batteries of various chemical types by integrally performing battery chemical type identification, anomaly detection, and remaining life prediction. means of solving the problem

[0014] Among the embodiments, the battery chemical type adaptive multivariate data-based battery condition diagnostic device comprises: a data collection unit that collects condition data including voltage, temperature, and current from a battery; a chemical type identification control unit that identifies the chemical type of the battery by calculating the flatness ratio of the nominal voltage range or OCV-SOC (Open Circuit Voltage - State of Charge) curve of the battery based on the condition data and determines an analysis path corresponding to the identified chemical type; an adaptive derived variable calculation unit that calculates different derived variables predefined to reflect the characteristics of each chemical type according to the determined analysis path; and an anomaly grade determination unit that determines the anomaly grade of the battery by inputting multivariate input data for anomaly detection, including time series data of the derived variables, into an anomaly detection model and performing inference of anomalies. It includes a life prediction unit that generates multivariate degradation history time series data for life prediction representing the cycle-by-cycle degradation trend of the above-mentioned derived variables, and inputs the multivariate degradation history time series data for life prediction into a remaining life prediction model to calculate an inferred RUL (Remaining Useful Life) value.

[0015] The chemical type identification control unit identifies the battery as a battery of the first chemical type when the unit cell nominal voltage range of the battery corresponds to a first voltage range, and identifies the battery as a battery of the second chemical type when the unit cell nominal voltage range corresponds to a second voltage range different from the first voltage range, and if the identification of the chemical type is uncertain based only on the nominal voltage range, the chemical type can be identified by additionally using the flatness ratio of the OCV-SOC curve.

[0016] The above adaptive derived variable calculation unit can calculate a first derived variable including at least one of a cell voltage deviation representing a voltage difference between multiple cells and a temperature change rate representing a change in temperature over time when the battery is identified as a Nickel Manganese Cobalt (NMC) battery, and can calculate a second derived variable including a phase transition peak characteristic of the voltage change rate (dQ / dV) curve for a change in unit capacity during charging and discharging and a spatial temperature deviation between multiple cells when the battery is identified as a Lithium Iron Phosphate (LFP) battery.

[0017] The above abnormality grade determination unit can generate an abnormal event in the multivariate input data for abnormality detection when the peak position of the phase transition peak characteristic extracted from the charge / discharge characteristics in the analysis path where the battery is identified as an LFP battery moves by more than a preset ratio compared to the normal reference, or when the peak height of the phase transition peak characteristic decreases by more than a preset ratio compared to the normal reference, and determine the abnormality grade of the battery based on the abnormal event.

[0018] The above anomaly detection grade determination unit generates a rule-based detection result by applying an adaptive threshold value, which is dynamically adjusted according to the chemical type and cumulative charge / discharge cycles of the battery, to the above-described variable, and obtains a model-based detection result from the above-described anomaly detection model that includes at least one of the type of anomaly, anomaly probability, or risk score of the battery, determines the anomaly grade of the battery using the rule-based detection result and the model-based detection result, maintains, up-corrects, or down-corrects the anomaly grade of the battery according to whether the rule-based detection result and the model-based detection result match or the degree of difference, and controls the generation or output of an alarm signal within a preset response time if the corrected anomaly grade is higher than or equal to a reference grade.

[0019] The above abnormal sign grade determination unit can automatically transmit a signal to an external system for restricting the operation or assigning maintenance of the target equipment equipped with the battery when the above abnormal sign grade is determined to be higher than the standard grade corresponding to immediate replacement.

[0020] The above-mentioned life prediction unit can simultaneously calculate the RUL inference value and a confidence interval representing the uncertainty of the RUL inference value through Monte Carlo Dropout, which performs multiple inferences while maintaining a Dropout active state during the inference process of the above-mentioned remaining life prediction model.

[0021] It may further include a self-reinforcing learning control unit that utilizes the actual abnormal sign type and actual remaining life measurement value fed back after maintenance is completed as labels, automatically inputs them into the retraining dataset of the abnormal sign detection model and the remaining life prediction model, and updates the abnormal sign detection model or the remaining life prediction model using the retraining dataset.

[0022] The above battery can be applied to a battery mounted on at least one of an electric two-wheeled vehicle, an electric forklift, an electric cart, an energy storage system (ESS), an unmanned delivery robot, and industrial mobile equipment.

[0023] Among the embodiments, a battery condition diagnosis method based on multivariate data adaptive to battery chemical type performed in a battery condition diagnosis device based on multivariate data adaptive to battery chemical type comprises: a data collection step of collecting condition data including voltage, temperature, and current from a battery through a data collection unit; a chemical type identification control step of identifying the chemical type of the battery and determining an analysis path corresponding to the identified chemical type by calculating the flatness ratio of the nominal voltage range or OCV-SOC (Open Circuit Voltage - State of Charge) curve of the battery based on the condition data through a chemical type identification control unit; an adaptive derived variable calculation step of calculating different derived variables predefined to reflect the characteristics of each chemical type according to the determined analysis path through an adaptive derived variable calculation unit; and an anomaly grade determination step of determining the anomaly grade of the battery by inputting multivariate input data for anomaly detection, including time series data of the derived variables, into an anomaly detection model and performing anomaly inference through an anomaly detection grade determination unit. It includes a life prediction step that generates multivariate degradation history time series data for life prediction representing the cycle-by-cycle degradation trend of the derived variable through a life prediction unit, and inputs the multivariate degradation history time series data for life prediction into a remaining life prediction model to calculate an inferred RUL (Remaining Useful Life) value. Effects of the invention

[0025] The disclosed technology may have the following effects. However, this does not mean that a specific embodiment must include all of the following effects or only the following effects; therefore, the scope of the rights of the disclosed technology should not be understood as being limited by this.

[0026] A battery condition diagnosis device and method based on multivariate data adaptive to battery chemical type according to one embodiment of the present invention can identify the chemical type of the battery based on condition data collected from the battery, and detect abnormal signs using derived variables that reflect characteristics of each chemical type according to an analysis path corresponding to the identified chemical type.

[0027] A battery condition diagnosis device and method based on multivariate data adaptive to battery chemical type according to one embodiment of the present invention can accurately predict the condition of batteries of various chemical types by inferring the remaining life using multivariate degradation history time series data that reflects the characteristics of each chemical type of battery.

[0028] A battery condition diagnosis device and method based on multivariate data adaptive to battery chemical type according to one embodiment of the present invention improves the reliability of battery condition diagnosis and enables efficient management of the condition of batteries of various chemical types by integrally performing identification of battery chemical type, detection of abnormal signs, and prediction of remaining lifespan.

[0029] A battery condition diagnosis device and method based on multivariate data adaptive to battery chemical type according to one embodiment of the present invention can improve diagnosis accuracy and remaining life prediction performance as the operating period increases by continuously retraining an anomaly detection model and a remaining life prediction model using actual anomaly types and actual remaining life measurements fed back after maintenance is completed. Accordingly, the reliability of battery condition diagnosis can be continuously improved by reflecting data accumulated in various operating environments. Brief explanation of the drawing

[0031] FIG. 1 is a drawing illustrating a battery condition diagnostic system according to one embodiment of the present invention. Figure 2 is a diagram illustrating the system configuration of the battery status diagnostic device of Figure 1. Figure 3 is a diagram illustrating the functional configuration of the battery status diagnostic device of Figure 1. FIG. 4 is a flowchart illustrating the functional configuration of a battery status diagnostic device according to the present invention. FIG. 5 is a flowchart illustrating the process of adaptively diagnosing abnormal signs and predicting remaining lifespan according to the battery chemical type of the battery condition diagnostic device according to the present invention. Specific details for implementing the invention

[0032] The description of the present invention is merely an example for structural or functional explanation, and therefore the scope of the present invention should not be interpreted as being limited by the examples described in the text. That is, since the examples are subject to various modifications and may take various forms, the scope of the present invention should be understood to include equivalents capable of realizing the technical concept. Furthermore, the objectives or effects presented in the present invention do not imply that a specific example must include all of them or only such effects; therefore, the scope of the present invention should not be understood as being limited by them.

[0033] Meanwhile, the meaning of the terms described in this application should be understood as follows.

[0034] Terms such as "first," "second," etc., are intended to distinguish one component from another, and the scope of rights shall not be limited by these terms. For example, the first component may be named the second component, and similarly, the second component may be named the first component.

[0035] When it is stated that one component is "connected" to another component, it should be understood that it may be directly connected to that other component, or that there may be other components in between. Conversely, when it is stated that one component is "directly connected" to another component, it should be understood that there are no other components in between. Meanwhile, other expressions describing the relationships between components, such as "between" and "exactly between," or "adjacent to" and "directly adjacent to," should be interpreted in the same way.

[0036] A singular expression should be understood to include a plural expression unless the context clearly indicates otherwise, and terms such as "include" or "have" are intended to specify the existence of the implemented features, numbers, steps, actions, components, parts, or combinations thereof, and should be understood not to preclude the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0037] In each step, identifiers (e.g., a, b, c, etc.) are used for convenience of explanation and do not describe the order of the steps; the steps may occur differently from the specified order unless a specific order is clearly indicated in the context. That is, the steps may occur in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.

[0038] The present invention may be implemented as computer-readable code on a computer-readable recording medium, and the computer-readable recording medium includes all types of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc. Additionally, the computer-readable recording medium may be distributed across networked computer systems, so that computer-readable code can be stored and executed in a distributed manner.

[0039] Unless otherwise defined, all terms used herein have the same meaning as generally understood by those skilled in the art to which this invention pertains. Terms defined in commonly used dictionaries should be interpreted as having meanings consistent with the context of the relevant technology and should not be interpreted as having an ideal or overly formal meaning unless explicitly defined in this application.

[0041] FIG. 1 is a drawing illustrating a battery condition diagnostic system according to one embodiment of the present invention.

[0042] Referring to FIG. 1, the battery status diagnostic system (100) may include a user terminal (110), a battery status diagnostic device (130), and a database (150).

[0043] The user terminal (110) may correspond to a computing device capable of providing execution results and related information of the battery condition diagnostic device (130). To this end, the user terminal (110) may install a dedicated application or software and receive and process data regarding the battery condition data, chemical type identification results, abnormality indicator grade, RUL (Remaining Useful Life) inference value, and confidence interval in real time through a dedicated application or software linked with the battery condition diagnostic system (100). In addition, the dedicated application may provide the abnormality indicator grade, RUL inference value, expected replacement time, battery degradation trend, etc., in the form of a visualization and dashboard, thereby providing the user with the diagnostic results, alarm information, and maintenance decision information of the battery condition diagnostic system (100).

[0044] The user terminal (110) may be composed of a single or multiple units, and if composed of multiple units, it may include a first user terminal, a second user terminal, ..., and the nth user terminal (n is a natural number). For example, the user terminal (110) may be implemented as a smartphone, laptop, or computer capable of operating in connection with the battery status diagnostic device (130), but is not necessarily limited thereto and may be implemented as various devices including a tablet PC. In addition, the user terminal (110) may be connected to the battery status diagnostic device (130) via a network, but is not necessarily limited thereto, and multiple user terminals (110) may be connected to the battery status diagnostic device (130) simultaneously.

[0045] The battery condition diagnostic device (130) may be implemented as a server corresponding to a computer or program that performs the battery condition diagnostic method based on multivariate data adaptive to the battery chemical type according to the present invention. For example, the battery condition diagnostic device (130) may be implemented as a server that provides functions for collecting condition data, identifying the battery chemical type, detecting abnormal signs, and predicting remaining lifespan, and each step of the battery condition diagnostic method based on multivariate data adaptive to the battery chemical type according to the present invention may be performed on the server.

[0046] Additionally, the battery status diagnostic device (130) is not necessarily limited to this and can be implemented as a server that performs artificial intelligence-based abnormal sign inference, RUL inference, and diagnostic result management functions. The battery status diagnostic device (130) can be connected to the user terminal (110) via a wired network or a wireless network such as Bluetooth, WiFi, or LTE, and can transmit and receive data with the user terminal (110) through the wired or wireless network.

[0047] Additionally, the battery status diagnostic device (130) may be implemented to operate in connection with an independent external system (not shown in FIG. 1). For example, the battery status diagnostic device (130) may operate in conjunction with a battery management system and a vehicle control system, and is not necessarily limited thereto, but may operate in conjunction with an energy management system and a cloud server. That is, the battery status diagnostic device (130) of the present invention may be implemented in a structure capable of interoperating with various external systems to comprehensively manage the state of the battery and utilize the diagnostic results in various systems.

[0048] In one embodiment, the battery condition diagnostic device (130) may be applied to batteries mounted on electric two-wheeled vehicles, electric forklifts, electric carts, energy storage systems (ESS), unmanned delivery robots, and industrial mobile equipment. Additionally, the battery condition diagnostic device (130) may be configured to diagnose the battery condition by automatically identifying chemical types and selecting analysis paths for each chemical type, even in an environment where multiple batteries are mixed and operated. For example, the battery condition diagnostic device (130) may simultaneously manage batteries of electric forklifts and unmanned delivery robots operated in a logistics warehouse and apply analysis paths corresponding to the chemical types of each battery. The database (150) may correspond to a storage device that stores various information required during the operation of the battery condition diagnostic device (130). For example, the database (150) may store various information such as condition data, chemical type information, derived variable data, anomaly detection results, RUL inference values, and learning data. Additionally, the database (150) can support the learning and inference of the abnormal sign detection model and the remaining life prediction model of the battery condition diagnostic device (130) based on the stored information. The database (150) is linked with the battery condition diagnostic device (130) to store or update status data and diagnostic result data, and can also transmit the data to the user terminal (110) as needed. Additionally, the database (150) can store maintenance history data and actual remaining life measurement data, and perform retraining of the abnormal sign detection model and the remaining life prediction model based on the stored maintenance history data and actual remaining life measurement data.

[0049] In FIG. 1, the database (150) is shown as a device independent of the battery condition diagnostic device (130), but is not necessarily limited thereto and can be implemented by being included in the battery condition diagnostic device (130).

[0051] Figure 2 is a diagram illustrating the system configuration of the battery status diagnostic device of Figure 1.

[0052] Referring to FIG. 2, the battery status diagnostic device (130) may include a processor (210), memory (230), user input / output unit (250), network input / output unit (270), and communication port unit (290).

[0053] The processor (210) can execute a battery state diagnosis procedure based on multivariate data adaptive to battery chemical type according to an embodiment of the present invention, manage memory (230) that is read or written during this process, and schedule the synchronization time between volatile memory and non-volatile memory in memory (230). The processor (210) can control the overall operation of the battery state diagnosis device (130) and is electrically connected to the memory (230), user input / output unit (250), and network input / output unit (270) to control the data flow between them. The processor (210) can be implemented as a CPU (Central Processing Unit) or GPU (Graphics Processing Unit) of the battery state diagnosis device (130).

[0054] The memory (230) may include an auxiliary storage device implemented as non-volatile memory such as an SSD (Solid State Disk) or HDD (Hard Disk Drive) and used to store all data required for the battery state diagnostic device (130), and may include a main memory device implemented as volatile memory such as RAM (Random Access Memory). Additionally, the memory (230) may store a set of instructions that execute the battery chemical type adaptive multivariate data-based battery state diagnostic method according to the present invention by being executed by an electrically connected processor (210).

[0055] The user input / output unit (250) includes an environment for receiving user input and an environment for outputting specific information to the user, and may include an input device including an adapter such as a touch pad, touch screen, virtual keyboard, or pointing device, and an output device including an adapter such as a monitor or touch screen. In one embodiment, the user input / output unit (250) may correspond to a computing device connected via remote access, and in such case, the battery status diagnostic device (130) may be performed as an independent server.

[0056] The network input / output unit (270) provides a communication environment for connecting to a user terminal (110) through a network and may include an adapter for communication such as a LAN (Local Area Network), MAN (Metropolitan Area Network), WAN (Wide Area Network), and VAN (Value Added Network). Additionally, the network input / output unit (270) may be implemented to provide short-range communication functions such as WiFi and Bluetooth, or wireless communication functions of 4G or higher, for wireless transmission of data.

[0057] The communication port section (290) may correspond to an interface for performing data transmission and reception between the battery status diagnostic device (130) and an external device. The communication port section (290) may be connected to a battery management system (BMS), a user terminal (110), a database (150), and an external system to transmit and receive status data or diagnostic result information. For example, the communication port section (290) may receive status data including voltage, temperature, and current from the battery management system and transmit an abnormal sign grade or RUL inference value to the user terminal (110) or an external system. Additionally, the communication port section (290) may transmit and receive data using a wired communication method or a wireless communication method, and may be connected to an external device using at least one of TCP / IP-based communication, CAN communication, RS-485 communication, Bluetooth communication, WiFi communication, and LTE communication.

[0059] Figure 3 is a diagram illustrating the functional configuration of the battery status diagnostic device of Figure 1.

[0060] Referring to FIG. 3, the battery condition diagnostic device (130) may include a data collection unit (310), a chemical type identification control unit (320), an adaptive derived variable calculation unit (330), an abnormal sign grade determination unit (340), a lifespan prediction unit (350), a self-reinforcing learning control unit (360), and a control unit (370).

[0061] At this time, the embodiments of the present invention are not required to include all of the above functional configurations simultaneously; depending on each embodiment, some of the above configurations may be omitted, or some or all of the above configurations may be selectively included. Furthermore, one embodiment of the present invention may be implemented as an independent module that selectively includes some of the above configurations, and a battery state diagnosis method based on multivariate data adaptive to battery chemical type may be performed through the interoperability between the modules. The operation of each configuration is described in detail below.

[0063] The data collection unit (310) can collect status data including voltage, temperature, and current from the battery. First, the data collection unit (310) can periodically receive status data including voltage, temperature, and current from a sensor or battery management system connected to the battery. The data collection unit (310) can generate status data in the form of a time series by collecting measurement values ​​for a plurality of unit cells along with time information. Additionally, the data collection unit (310) can check for missing data and measurement ranges in the collected status data, and if abnormal measurement values ​​are included, exclude the data or request re-collection. For example, the data collection unit (310) can generate status data in chronological order by collecting the voltage value of each unit cell, the current value of the battery pack, and the temperature value of a plurality of locations at a 1-second interval.

[0064] Next, the data collection unit (310) can sort the collected state data based on the same time standard and organize the data obtained from different sensors into a single data frame. In this process, the data collection unit (310) can be configured to correct the time error between measurement points and unify the units of the data so that the state data can be used based on the same standard in subsequent processing. For example, the data collection unit (310) can convert voltage data into V units, temperature data into °C units, and current data into A units, and then correct the error between measurement points occurring at different sensors to generate a state data frame corresponding to the same time standard. Additionally, the data collection unit (310) can be configured to store the collection time information corresponding to each state data together so that time series analysis is possible. For example, the data collection unit (310) can sort voltage data, temperature data, and current data based on the same timestamp to generate a single state data set.

[0065] Next, the data collection unit (310) can store sorted state data in a memory area and then provide it so that a component performing subsequent processing can use the state data. The data collection unit (310) can maintain time-series data that reflects state changes over time by storing data while maintaining the order of collection of state data. Additionally, the data collection unit (310) can maintain a data set that reflects the latest state by sequentially adding new state data to existing state data when new state data is collected. For example, the data collection unit (310) can maintain a set of the latest state data by sequentially storing the most recent 1,000 state data and removing the oldest data when new state data is collected.

[0066] Meanwhile, the data collection unit (310) can collect status data using wired or wireless communication. The data collection unit (310) can receive status data after forming a communication connection with a battery management system, a vehicle controller, or an external sensor module, and in the event of a communication failure, it can request retransmission while maintaining previously collected data. Additionally, the data collection unit (310) can be configured to check the error detection code of the received status data to determine whether the data is corrupted, and then store only normal status data. For example, the data collection unit (310) can receive voltage, temperature, and current data from the battery management system using CAN communication, and in the event of a communication error, it can request retransmission of the same data.

[0067] In one embodiment, the data collection unit (310) may be implemented in a configuration that directly receives status data from a battery management system. In this case, the data collection unit (310) may receive voltage data per multiple unit cell, temperature data per multiple location, and current data per battery pack unit managed by the battery management system in units of communication frames. Additionally, the data collection unit (310) may distinguish voltage, temperature, and current based on identification information included in the received communication frames, and store the distinguished status data by corresponding it to the same collection time information. For example, the data collection unit (310) may receive a CAN communication frame from the battery management system, classify cell voltage data, temperature data, and current data according to the frame identifier, and then organize them into a single status data frame.

[0068] In one embodiment, the data collection unit (310) may be implemented in a configuration that collects status data from a sensor module separately attached to the battery. In this case, the data collection unit (310) receives measurement values ​​from a voltage sensor, a temperature sensor, and a current sensor, respectively, and if the sampling periods for each sensor are different, it may sort the status data by interpolating the measurement values ​​to match the reference sampling period or by maintaining the most recent measurement value. Additionally, the data collection unit (310) may be configured to mark the measurement value as error data if the measurement value of the sensor module falls outside the allowable range, and to exclude the status data frame containing the error data from subsequent processing. For example, the data collection unit (310) may sort the voltage data measured by the voltage sensor at a 1-second interval and the temperature data measured by the temperature sensor at a 5-second interval to match the reference sampling period, and create a status data frame by including current data at the same point in time.

[0069] In one embodiment, the data collection unit (310) may be implemented in a configuration that collects status data from historical data stored in the internal storage space or external storage device of the battery status diagnosis device (130). In this case, the data collection unit (310) may read status data including voltage, temperature, and current stored during past operation in chronological order and sort the read status data according to the analysis criteria of the battery currently being diagnosed. Additionally, if the collection cycle, data format, and unit of the historical data are different, the data collection unit (310) may convert them into the same data format and provide them for subsequent processing. For example, the data collection unit (310) may read status data stored in a CSV file, a database table, or a log file, sort it based on the collection time information, and then convert it into status data that fits the input format of an anomaly detection model or a remaining life prediction model.

[0070] The chemical type identification control unit (320) can identify the chemical type of the battery by calculating the nominal voltage range of the battery or the flatness ratio of the OCV-SOC (Open Circuit Voltage - State of Charge) curve based on state data, and determine an analysis path corresponding to the identified chemical type. Here, the chemical type of the battery may include NMC (Nickel Manganese Cobalt) batteries, LFP (Lithium Iron Phosphate) batteries, etc. The chemical type identification control unit (320) can calculate the nominal voltage range of the battery using state data received from the data collection unit (310). For example, the chemical type identification control unit (320) can analyze the voltage value per unit cell using voltage information included in the state data and determine the nominal voltage range of the battery based on the voltage distribution for a plurality of unit cells. In addition, the chemical type identification control unit (320) can primarily identify the chemical type of the battery by comparing the calculated nominal voltage range with reference information for each chemical type stored in advance. For example, the chemical type identification control unit (320) can identify the battery as a specific chemical type if the calculated nominal voltage range matches or is included in the reference nominal voltage range of a specific chemical type stored in advance.

[0071] At this time, the chemical type identification control unit (320) can additionally calculate the flatness ratio of the OCV-SOC curve when it is difficult to clearly distinguish the chemical type based solely on the nominal voltage range. The chemical type identification control unit (320) can generate an OCV-SOC curve using the change characteristics of the open circuit voltage according to the charge state, and calculate the flatness ratio of the OCV-SOC curve by calculating the ratio of the voltage change range in a specific SOC section to the voltage change range in the entire SOC section. In addition, the chemical type identification control unit (320) can re-determine the chemical type by comparing the calculated flatness ratio of the OCV-SOC curve with reference values ​​for each chemical type. For example, the chemical type identification control unit (320) can compare the calculated flatness ratio of the OCV-SOC curve with reference information corresponding to multiple chemical types, and identify the battery as a chemical type with high flatness characteristics as a result of the comparison.

[0072] Subsequently, the chemical type identification control unit (320) can determine the final chemical type based on the identification result using the nominal voltage range and the flatness ratio of the OCV-SOC curve. The chemical type identification control unit (320) can select an analysis path corresponding to the finally determined chemical type and provide the selected analysis path information so that it can be used in a subsequent processing step. Additionally, the chemical type identification control unit (320) can be configured so that subsequent components can use the same analysis path by storing the identification result corresponding to the analysis path in a memory area. For example, the chemical type identification control unit (320) can determine a first analysis path if identified as an NMC battery and determine a second analysis path if identified as an LFP battery.

[0073] Here, the first analysis path may include data processing procedures and analysis criteria corresponding to the first chemical type, and the second analysis path may include data processing procedures and analysis criteria corresponding to the second chemical type. The chemical type identification control unit (320) may select the first analysis path or the second analysis path according to the identified chemical type so that the characteristics of different chemical types are reflected in the subsequent processing. For example, if the chemical type identification control unit (320) is identified as a battery of a specific chemical type, it may determine an analysis path corresponding to that chemical type and provide the determined analysis path to a subsequent component.

[0074] Meanwhile, the chemical type identification control unit (320) can verify or update the chemical type identification results using newly collected status data during operation. The chemical type identification control unit (320) can reanalyze the status data at regular intervals and compare it with the existing identification results, and if different results are obtained, it can reset the analysis path based on the latest status data. In addition, the chemical type identification control unit (320) can control the analysis to be performed suitable for the current battery condition by reflecting the reset analysis path in the subsequent processing process. For example, the chemical type identification control unit (320) can re-identify the chemical type by analyzing newly collected status data after battery replacement and determine a new analysis path corresponding to the changed chemical type.

[0075] In one embodiment, the chemical type identification control unit (320) may identify the battery as a battery of the first chemical type if the unit cell nominal voltage range of the battery corresponds to a first voltage range, and may identify the battery as a battery of the second chemical type if the unit cell nominal voltage range corresponds to a second voltage range that is different from the first voltage range. Here, the battery of the first chemical type may include NMC batteries, NCA batteries, etc., and the battery of the second chemical type may include LFP batteries, LMFP batteries, etc. Additionally, the chemical type identification control unit (320) may be configured to identify batteries of the third chemical type or higher, including all-solid-state batteries, and to select an analysis path corresponding to the identified chemical type. The chemical type identification control unit (320) may first determine whether there is a correspondence with the first voltage range using the nominal voltage range calculated from the voltage value per unit cell. The chemical type identification control unit (320) may classify the battery as a battery of the first chemical type if the calculated nominal voltage range is included in the first voltage range, and may additionally determine whether there is a correspondence with the second voltage range if it is not included in the first voltage range. Additionally, the chemical type identification control unit (320) can classify the battery as a battery of the second chemical type if it falls within the second voltage range. For example, the chemical type identification control unit (320) can determine the chemical type by first performing a comparison with respect to the first voltage range, and then, if it does not match the first voltage range, performing a comparison with respect to the second voltage range.

[0076] Next, the chemical type identification control unit (320) can determine the chemical type if the nominal voltage range clearly corresponds to one of the voltage ranges. On the other hand, if the calculated nominal voltage range is located in the boundary area between the first voltage range and the second voltage range, or overlaps with multiple voltage ranges, it can determine that the identification of the chemical type is uncertain. Additionally, the chemical type identification control unit (320) can generate and store state information indicating that the identification is uncertain. For example, if multiple chemical types are derived as candidates, the chemical type identification control unit (320) may not immediately determine the chemical type but set it to an uncertain state.

[0077] Furthermore, the chemical type identification control unit (320) can generate identification result information corresponding to the chemical type when the chemical type is determined. The chemical type identification control unit (320) can include the chemical type identification time, the nominal voltage range used, and the corresponding chemical type information in the generated identification result information. Additionally, the chemical type identification control unit (320) can be configured to store the generated identification result information so that it can be compared with the identification result of the status data collected thereafter. For example, the chemical type identification control unit (320) can verify the consistency of chemical type identification by storing the current identification result and the previous identification result together.

[0078] In one embodiment, the chemical type identification control unit (320) may identify the chemical type by additionally using the flatness ratio of the OCV-SOC curve when the identification of the chemical type is uncertain based solely on the nominal voltage range. First, the chemical type identification control unit (320) may evaluate the reliability of the primary chemical type identification result based on the nominal voltage range. The chemical type identification control unit (320) may determine that the identification of the chemical type is uncertain if the calculated nominal voltage range overlaps with reference information corresponding to multiple chemical types, or if multiple candidate chemical types are derived based on the nominal voltage range. Additionally, if the chemical type identification control unit (320) determines that it is difficult to determine the chemical type based solely on the nominal voltage range, it may perform a secondary identification procedure using the flatness ratio of the OCV-SOC curve. For example, the chemical type identification control unit (320) may select the flatness ratio of the OCV-SOC curve as an additional identification criterion when both the first chemical type and the second chemical type are derived as candidates based on the nominal voltage range.

[0079] Next, the chemical type identification control unit (320) can generate an OCV-SOC curve using voltage information and charge state information included in the state data. The chemical type identification control unit (320) can arrange the change in open circuit voltage corresponding to the change in charge state in a time series form and analyze the voltage change characteristics according to the change in charge state from the generated OCV-SOC curve. In addition, the chemical type identification control unit (320) can calculate the flatness ratio of the OCV-SOC curve by distinguishing between the section where the voltage change is maintained gradually and the section where the voltage change occurs rapidly in the OCV-SOC curve. For example, the chemical type identification control unit (320) can analyze the slope distribution for the entire section of the OCV-SOC curve and calculate the flatness ratio of the OCV-SOC curve using the ratio of the section where the slope is below a reference value.

[0080] Next, the chemical type identification control unit (320) can calculate the suitability of candidate chemical types by comparing the flatness ratio of the calculated OCV-SOC curve with reference information for each chemical type. The chemical type identification control unit (320) can calculate the difference value or similarity of the flatness ratio for each of the multiple candidate chemical types and select the chemical type having the highest suitability. In addition, the chemical type identification control unit (320) can determine the final chemical type by combining the primary identification result based on the nominal voltage range and the secondary identification result based on the flatness ratio of the OCV-SOC curve. For example, the chemical type identification control unit (320) can identify the battery as a battery of the second chemical type if the comparison result of the flatness ratio of the OCV-SOC curve shows high similarity to the reference information of the second chemical type when the first chemical type and the second chemical type are derived as candidates.

[0081] Furthermore, the chemical type identification control unit (320) can set an analysis path corresponding to the chemical type using the finally determined chemical type information. The chemical type identification control unit (320) can be configured to store the final identification result and the flatness ratio calculation result of the OCV-SOC curve together so that they can be used as reference information in the future chemical type identification process for the same battery. In addition, the chemical type identification control unit (320) can verify the consistency of the chemical type identification result by comparing it with the stored identification result when new state data is collected. For example, the chemical type identification control unit (320) can control the maintenance of the existing chemical type information if the flatness ratio of the newly generated OCV-SOC curve matches the existing identification result during the subsequent operation process.

[0082] The adaptive derived variable calculation unit (330) can calculate different derived variables that are predefined to reflect the characteristics of each chemical type according to the determined analysis path. First, the adaptive derived variable calculation unit (330) can check the analysis path information transmitted from the chemical type identification control unit (320) and select a derived variable calculation rule corresponding to the confirmed analysis path. Here, the derived variable calculation rule may include information on which physical quantity to extract from the state data, which calculation formula to apply, and which data interval to use, and may include, for example, calculation procedures for performing differentiation, average value calculation, deviation calculation, rate of change calculation, or correlation analysis on voltage data, temperature data, and current data. For example, when the first analysis path is selected, the adaptive derived variable calculation unit (330) may select a first derived variable calculation rule that includes a calculation procedure for calculating cell voltage deviation and temperature rate of change, and when the second analysis path is selected, select a second derived variable calculation rule for calculating voltage change characteristics and temperature distribution characteristics during charging and discharging.

[0083] Specifically, the adaptive derived variable calculation unit (330) can calculate multiple derived variables representing characteristics of each chemical type using voltage, temperature, and current included in the state data. Here, the derived variables may include cell voltage deviation representing the voltage difference between multiple cells, temperature change rate representing the amount of temperature change over time, characteristic values ​​of the voltage change rate curve for unit capacity change during charging and discharging, spatial temperature deviation between multiple cells, rate of change of charging and discharging current, and characteristic values ​​representing the amount of internal resistance change and charging and discharging efficiency. The adaptive derived variable calculation unit (330) can select some of the multiple derived variables or calculate additional new derived variables according to the analysis path. For example, the adaptive derived variable calculation unit (330) can calculate cell voltage deviation and temperature change rate using voltage data and temperature data in a specific analysis path, and calculate voltage change characteristics and spatial temperature deviation based on charging and discharging data in another analysis path.

[0084] Next, the adaptive derived variable calculation unit (330) can generate time series data for each derived variable by aligning the calculated derived variables based on the same time standard. The adaptive derived variable calculation unit (330) can form a time series structure by storing calculation time information corresponding to each derived variable and arranging the continuously calculated derived variables in chronological order. Additionally, the adaptive derived variable calculation unit (330) can generate a multivariate data structure by aligning different types of derived variables to correspond to the same time point. For example, the adaptive derived variable calculation unit (330) can generate multivariate time series data by aligning cell voltage deviation, temperature change rate, and current change rate to correspond to the same timestamp.

[0085] Furthermore, the adaptive derived variable calculation unit (330) can calculate new derived variables by applying a derived variable calculation rule corresponding to the changed analysis path when the analysis path is changed. The adaptive derived variable calculation unit (330) can prevent the mixing of characteristics of different chemical types by managing the derived variables generated from the analysis path before the change and the derived variables generated from the analysis path after the change separately. In addition, the adaptive derived variable calculation unit (330) can be configured to generate new time series data based on the time of the change of the analysis path so that it can be used in subsequent processing. For example, when a new analysis path is selected, the adaptive derived variable calculation unit (330) can generate a new data set that is distinct from the existing derived variable data, and generate multivariate input data for anomaly detection or multivariate degradation history time series data for lifespan prediction using the derived variables corresponding to the changed analysis path.

[0086] In one embodiment, the adaptive derived variable calculation unit (330) can calculate a first derived variable including at least one of a cell voltage deviation representing the voltage difference between multiple cells and a temperature change rate representing the amount of temperature change over time when the battery is identified as a Nickel Manganese Cobalt (NMC) battery, and can calculate a second derived variable including a phase transition peak characteristic of the voltage change rate (dQ / dV) curve for unit capacity change during charging and discharging and a spatial temperature deviation between multiple cells when the battery is identified as a Lithium Iron Phosphate (LFP) battery. The adaptive derived variable calculation unit (330) can calculate the first derived variable when the battery is identified as an NMC battery. The adaptive derived variable calculation unit (330) can generate a characteristic value representing the degree of imbalance between cells by calculating the cell voltage deviation using voltage data of multiple cells. Additionally, the adaptive derived variable calculation unit (330) can calculate the temperature change rate by comparing temperature data in chronological order and calculating the amount of temperature change per unit time. The adaptive derived variable calculation unit (330) can generate a first derived variable including at least one of cell voltage deviation and temperature change rate, and can store the generated first derived variable in a time series form. For example, the adaptive derived variable calculation unit (330) can generate a first derived variable that reflects when voltage imbalance between multiple cells increases or the rate of temperature rise changes during the charging and discharging process.

[0087] Specifically, the adaptive derived variable calculation unit (330) may use the maximum, minimum, or average values ​​of the voltage values ​​of multiple cells to calculate the cell voltage deviation, and may calculate the cell voltage deviation by calculating the voltage difference between cells or the deviation relative to the reference voltage. In addition, the adaptive derived variable calculation unit (330) may calculate the temperature change rate by calculating the amount of temperature change between consecutive time points using temperature data measured at multiple time points. The adaptive derived variable calculation unit (330) may generate a first derived variable data set by aligning the calculated cell voltage deviation and temperature change rate based on the same time standard. For example, the adaptive derived variable calculation unit (330) may generate a first derived variable data set that reflects when the voltage of a specific cell shows a difference from other cells or when the rate of temperature rise increases during the charging process.

[0088] Next, the adaptive derived variable calculation unit (330) can calculate a second derived variable when the battery is identified as an LFP battery. The adaptive derived variable calculation unit (330) can generate a dQ / dV curve using voltage data and capacity data measured during the charging and discharging process, and can calculate phase transition peak characteristics by extracting at least one of the position, peak height, and peak width of the phase transition peak from the generated dQ / dV curve. Additionally, the adaptive derived variable calculation unit (330) can calculate the spatial temperature deviation between cells or modules by comparing temperature data measured at multiple locations. The adaptive derived variable calculation unit (330) can generate a second derived variable including the phase transition peak characteristics and the spatial temperature deviation. For example, the adaptive derived variable calculation unit (330) can generate a second derived variable that reflects changes in the peak shape of the dQ / dV curve or the temperature of a specific region being higher than that of another region during the charging and discharging process.

[0089] Furthermore, the adaptive derived variable calculation unit (330) can generate a data structure in the form of a time series by arranging the calculated first derived variable or second derived variable in chronological order. The adaptive derived variable calculation unit (330) can be configured to reflect changes in characteristics over time by storing the calculation time information corresponding to each derived variable together. In addition, the adaptive derived variable calculation unit (330) can maintain continuous derived variable time series data by updating the derived variable and adding it to the existing data set whenever new state data is input. For example, the adaptive derived variable calculation unit (330) can store cell voltage deviation, temperature change rate, phase transition peak characteristics, and spatial temperature deviation in chronological order and use them to generate multivariate input data for anomaly detection or multivariate degradation history time series data for life prediction. Here, multivariate input data for anomaly detection may correspond to a set of multidimensional feature vectors constructed by aligning multiple derived variables calculated at a specific point in time or during a specific period based on the same time standard, and may correspond to data in which multiple derived variables, such as cell voltage deviation, temperature change rate, phase transition peak characteristics, and spatial temperature deviation, are arranged in a time series form. Additionally, multivariate degradation history time series data for life prediction may correspond to time series data constructed by accumulating the history of changes in multiple derived variables calculated according to the progression of charge-discharge cycles, and may correspond to data arranged in chronological order or by the number of charge-discharge cycles to indicate degradation trends per cycle.

[0090] The anomaly grade determination unit (340) can determine the anomaly grade of the battery by inputting multivariate input data for anomaly detection, including time series data of derived variables, into an anomaly detection model and performing anomaly inference. First, the anomaly grade determination unit (340) can receive multivariate input data for anomaly detection, including time series data of derived variables, from the adaptive derived variable calculation unit (330). The anomaly grade determination unit (340) can normalize the received multivariate input data for anomaly detection into a format that can be input into the anomaly detection model after verifying the time sequence, whether data is missing, and the configuration of the input dimensions. Additionally, the anomaly grade determination unit (340) can perform scale adjustment so that when the unit and value ranges for each derived variable are different, they can be processed according to the same comparison standard. For example, the anomaly detection grade determination unit (340) can sort multivariate input data for anomaly detection, including cell voltage deviation, temperature change rate, phase transition peak characteristics and spatial temperature deviation, in chronological order and then convert it into an input tensor of an anomaly detection model.

[0091] At this time, the anomaly detection grade determination unit (340) can analyze the time series change pattern of the derived variable using an anomaly detection model. Here, the anomaly detection model may be implemented as an LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), Transformer, AutoEncoder, or a combination thereof, and may correspond to a model that has learned the correlation between multiple derived variables included in the multivariate input data for anomaly detection, the trend of change over time, and the pattern of deviating from the normal state. The anomaly detection model may output at least one of an anomaly type, an anomaly probability, or a risk score based on the learned time series pattern. The anomaly detection grade determination unit (340) may obtain at least one of the possibility of anomaly occurring in the battery, an anomaly type, or a risk score from the output value of the anomaly detection model. For example, the anomaly detection grade determination unit (340) may obtain a high risk score from the anomaly detection model when a time series pattern is input in which the cell voltage deviation increases and the temperature change rate rises together over a certain period.

[0092] Furthermore, the abnormal sign grade determination unit (340) can determine the abnormal sign grade of the battery using the output value of the abnormal sign detection model. The abnormal sign grade can be divided into multiple stages, such as a state with no or low abnormal signs, a state requiring observation, a state requiring maintenance, or a state requiring an alarm. The abnormal sign grade determination unit (340) can calculate the abnormal sign grade of the battery by comparing the risk score or abnormality probability output from the abnormal sign detection model with a pre-set grade standard. For example, the abnormal sign grade determination unit (340) can determine the abnormal sign grade of the battery as a state requiring observation if the risk score is above the observation standard and below the maintenance standard.

[0093] Meanwhile, the anomaly grade determination unit (340) can store the determined anomaly grade along with the derived variable information used to calculate the anomaly grade. The anomaly grade determination unit (340) can be configured to store the anomaly grade, the output value of the anomaly detection model, the time of creation of the input data, and the derived variable information used in correspondence with each other so that they can be utilized for future diagnostic history lookup or maintenance judgment. Additionally, the anomaly grade determination unit (340) can generate diagnostic result data to be provided to a user terminal or an external system when the anomaly grade is higher than or equal to a reference grade. For example, the anomaly grade determination unit (340) can store the anomaly grade calculated at a specific point in time and the derived variable items that influenced the determination of the grade together and display them on the dashboard of the user terminal.

[0094] In one embodiment, the abnormal sign grade determination unit (340) can generate an abnormal event in multivariate input data for abnormal detection when the peak position of a phase transition peak characteristic extracted from charge / discharge characteristics in an analysis path where the battery is identified as an LFP battery moves by more than a preset ratio relative to a normal reference, or when the peak height of the phase transition peak characteristic decreases by more than a preset ratio relative to a normal reference, and can determine the abnormal sign grade of the battery based on the abnormal event. First, when the abnormal sign grade determination unit (340) receives analysis path information where the battery is identified as an LFP battery from the chemical type identification control unit (320), it can set the phase transition peak characteristic extracted from charge / discharge characteristics as an abnormal detection target. The abnormal sign grade determination unit (340) can compare the peak position and peak height of the phase transition peak characteristic extracted from the charge / discharge characteristics with reference information in a normal state. For example, the abnormal sign grade determination unit (340) can calculate the amount of movement of the peak position and the amount of reduction of the peak height by comparing the peak position and peak height of the phase transition peak characteristics extracted from the current charge / discharge characteristics with reference information stored in the previous charge / discharge cycle or normal state.

[0095] Additionally, the abnormal sign grade determination unit (340) can determine whether the degradation of the battery is progressing by comparing the amount of change in the currently measured phase transition peak characteristics with reference data or phase transition peak characteristics extracted from a previous charge / discharge cycle. For example, the abnormal sign grade determination unit (340) can analyze the trend of change in peak position and peak height using phase transition peak characteristics extracted from a dQ / dV curve, and is not necessarily limited to this, but can analyze the trend of change using phase transition peak characteristics extracted from various analysis techniques based on ICA (Incremental Capacity Analysis), DVA (Differential Voltage Analysis), or charge / discharge characteristics. That is, the abnormal sign grade determination unit (340) can determine whether there is an abnormality by using phase transition peak characteristics extracted from charge / discharge characteristics as a common analysis target, thereby determining the change in electrochemical characteristics of the battery or the state of degradation progressing without being limited to a specific analysis technique.

[0096] Next, the abnormal sign grade determination unit (340) can determine whether the peak position of the phase transition peak characteristic moves by more than a preset ratio relative to the normal standard or whether the peak height of the phase transition peak characteristic decreases by more than a preset ratio relative to the normal standard. The abnormal sign grade determination unit (340) can calculate the amount of movement of the peak position or the amount of decrease of the peak height by comparing the amount of change of the phase transition peak characteristic with reference information in the normal state. Additionally, the abnormal sign grade determination unit (340) can determine that a change or deterioration of the electrochemical characteristic is progressing if the calculated amount of change is greater than or equal to the preset ratio. For example, the abnormal sign grade determination unit (340) can determine that the possibility of an abnormal state occurring has increased if the peak position of the phase transition peak characteristic continuously moves or the peak height continuously decreases during multiple charge / discharge cycles.

[0097] Furthermore, the abnormal sign grade determination unit (340) may generate an abnormal event corresponding to multivariate input data for abnormal detection when a shift in the peak position or a decrease in the peak height of the phase transition peak characteristic is detected. The abnormal event may include information indicating the time of occurrence of the abnormality, the number of occurrences of the abnormality, information on the phase transition peak characteristic, and the degree of the abnormality. Additionally, the abnormal sign grade determination unit (340) may determine the abnormal sign grade of the battery by comparing the generated abnormal event with the abnormal sign judgment criteria. For example, the abnormal sign grade determination unit (340) may determine that the state requires observation when a change in the phase transition peak characteristic occurs temporarily, and determine the abnormal sign grade as a state requiring maintenance or an alarm when abnormal events are repeatedly generated in multiple charge / discharge cycles.

[0098] In one embodiment, the anomaly grade determination unit (340) generates rule-based detection results by applying an adaptive threshold that is dynamically adjusted according to the chemical type of the battery and the cumulative number of charge / discharge cycles to the derived variable, and can obtain model-based detection results including at least one of the battery anomaly type, anomaly probability, or risk score from the anomaly detection model. The anomaly grade determination unit (340) receives the derived variable provided by the adaptive derived variable calculation unit (330) and can verify the chemical type of the battery and the cumulative number of charge / discharge cycles. The anomaly grade determination unit (340) can set an adaptive threshold for each derived variable to reflect the degradation characteristics by chemical type and changes in characteristics according to usage history. Here, the adaptive threshold may correspond to a judgment criterion value that changes dynamically according to the chemical type of the battery, the cumulative number of charge / discharge cycles, and the trend of change of the derived variable. In addition, the anomaly grade determination unit (340) can apply a judgment criterion suitable for a degraded battery by updating the adaptive threshold as the cumulative number of charge / discharge cycles increases. For example, the abnormal sign grade determination unit (340) can apply different adaptive threshold values ​​to the same derived variable depending on the chemical type of the battery or the number of accumulated charge / discharge cycles.

[0099] Next, the anomaly grade determination unit (340) can generate a rule-based detection result by comparing a derived variable with an adaptive threshold. The anomaly grade determination unit (340) can determine whether each derived variable exceeds an adaptive threshold and calculate whether an anomaly has occurred or the degree of an anomaly using the determination result. Additionally, the anomaly grade determination unit (340) can generate a rule-based detection result by combining the determination results for multiple derived variables. Here, the rule-based detection result may correspond to judgment result information indicating a normal state, an anomaly state, or the degree of an anomaly. For example, the anomaly grade determination unit (340) can generate a rule-based detection result that determines that an anomaly has occurred in the battery when the cell voltage deviation and the temperature change rate each exceed their respective adaptive thresholds.

[0100] Next, the anomaly grade determination unit (340) can obtain model-based detection results by inputting multivariate input data for anomaly detection into an anomaly detection model. The anomaly detection model may correspond to a model that has learned time-series patterns of normal and abnormal states, and may analyze the correlation and temporal change characteristics of the input derived variables. The anomaly grade determination unit (340) can obtain at least one of the battery anomaly type, anomaly probability, or risk score from the output value of the anomaly detection model. Here, the anomaly type may include information indicating an increase in internal resistance, expansion of cell imbalance, temperature anomaly, or a state of deterioration progression, and the risk score may correspond to information that quantifies the severity of the abnormal state. For example, the anomaly grade determination unit (340) can obtain a high risk score along with a specific anomaly type from the anomaly detection model.

[0101] Additionally, the anomaly grade determination unit (340) can independently generate and store rule-based detection results and model-based detection results, respectively. The anomaly grade determination unit (340) can be configured to be used in a subsequent anomaly grade determination process by managing the anomaly status information included in the rule-based detection results in correspondence with the anomaly probability or risk score included in the model-based detection results. Additionally, the anomaly grade determination unit (340) can manage the history of anomaly occurrences by storing the generation time of the rule-based detection results and the input derived variable information together. For example, the anomaly grade determination unit (340) can store the rule-based detection results and model-based detection results generated at the same time in correspondence with each other and use the stored results to comprehensively determine the abnormal state of the battery.

[0102] In one embodiment, the anomaly grade determination unit (340) determines the anomaly grade of the battery using rule-based detection results and model-based detection results, and maintains, corrects upward, or corrects downward the anomaly grade of the battery depending on whether the rule-based detection results and the model-based detection results match or the degree of difference, and if the corrected anomaly grade is higher than or equal to the reference grade, it can control the generation or output of an alarm signal within a preset response time. After receiving the rule-based detection results and model-based detection results, the anomaly grade determination unit (340) may apply threshold calculation rules corresponding to different derived variables for each chemical type of the battery. The anomaly grade determination unit (340) may calculate an adaptive threshold using the chemical type of the battery and the cumulative number of charge / discharge cycles, and generate rule-based detection results by applying the calculated adaptive threshold to the derived variables. Additionally, the anomaly grade determination unit (340) may obtain model-based detection results from an anomaly detection model. For example, the abnormal sign grade determination unit (340) may apply a first threshold calculation rule corresponding to a first derived variable for a first chemical type and a second threshold calculation rule corresponding to a second derived variable for a second chemical type.

[0103] Additionally, the anomaly grade determination unit (340) can analyze the correlation between the two results by comparing the anomaly information included in the rule-based detection result with the anomaly probability, risk score, or type of anomaly included in the model-based detection result. If the two results indicate the same anomaly state, the anomaly grade determination unit (340) can maintain the current anomaly grade or determine that the reliability of the anomaly state is high. For example, the anomaly grade determination unit (340) can determine that the two results match if an anomaly state is determined in the rule-based detection result and a high risk score is calculated in the model-based detection result. Additionally, if different derived variables are used depending on the chemical type, the anomaly grade determination unit (340) can dynamically change the adaptive threshold by reflecting the reference information corresponding to the derived variable and the degradation characteristics according to the cumulative number of charge / discharge cycles. For example, even if the type of anomaly is the same, the anomaly grade determination unit (340) can generate rule-based detection results by applying different reference values ​​depending on the chemical type of the battery or the cumulative number of charge / discharge cycles.

[0104] Next, the anomaly grade determination unit (340) can determine the anomaly grade of the battery using rule-based detection results and model-based detection results. The anomaly grade determination unit (340) can maintain the current anomaly grade, or adjust it upward or downward by analyzing whether there is a match or the degree of difference between the rule-based detection results and the model-based detection results. The anomaly grade determination unit (340) can maintain the existing anomaly grade if the degree of difference is within a standard range, and can adjust the anomaly grade upward if the risk level of the model-based detection results is higher than that of the rule-based detection results. Additionally, the anomaly grade determination unit (340) can adjust the anomaly grade downward if the risk level of the model-based detection results is lower than that of the rule-based detection results. For example, the anomaly grade determination unit (340) can adjust the existing anomaly grade upward if a normal state is derived from the rule-based detection results but a high probability of anomaly is calculated from the model-based detection results.

[0105] Next, the anomaly grade determination unit (340) may maintain, up-correct, or down-correct the anomaly grade based on whether the rule-based detection results and the model-based detection results match or the degree of difference. The anomaly grade determination unit (340) may calculate the anomaly grade corrected according to the correction rule and record the calculated anomaly grade in a storage device. For example, the anomaly grade determination unit (340) may stepwise up-correct the anomaly grade when the same type of anomaly is continuously detected and the risk score of the model-based detection results increases. Additionally, the anomaly grade determination unit (340) may be configured to determine an anomaly grade that reflects the degradation characteristics of each chemical type by using rule-based detection results generated using different threshold calculation rules for each chemical type. Accordingly, the anomaly grade determination unit (340) can reduce false detections or missed detections that may occur when the same reference value is uniformly applied to batteries of different chemical types.

[0106] Furthermore, the abnormal sign grade determination unit (340) can determine whether the corrected abnormal sign grade is higher than or equal to the reference grade. If the corrected abnormal sign grade is higher than or equal to the reference grade, the abnormal sign grade determination unit (340) can control the generation or output of an alarm signal within a preset response time. Here, the alarm signal may include a warning message provided to the user terminal (110), an indicator light control signal, an audible signal, or an event signal transmitted to an external system. Additionally, the abnormal sign grade determination unit (340) can generate and store information regarding the type of abnormal sign, the risk level, and the time of alarm occurrence along with the alarm signal. For example, if the corrected abnormal sign grade determination unit (340) determines that the corrected abnormal sign grade is higher than or equal to the reference grade, it can output a warning message to the user terminal (110) and simultaneously transmit an event signal to an external control system.

[0107] In one embodiment, the anomaly grade determination unit (340) may automatically transmit a signal to an external system for restricting operation or assigning maintenance of the target equipment equipped with a battery when the anomaly grade is determined to be higher than the standard grade corresponding to immediate replacement. The anomaly grade determination unit (340) may generate control information to be transmitted to an external system when the anomaly grade is determined to be higher than the standard grade corresponding to immediate replacement. Here, the control information may include at least one of the identification information of the target battery, the anomaly grade, the type of anomaly, the risk level information, the time of diagnosis, and the identification information of the target equipment. Additionally, the anomaly grade determination unit (340) may enable the operation restriction or assignment of maintenance for the target equipment to be performed automatically by transmitting the generated control information to an external system. For example, when the anomaly grade determination unit (340) is determined to be higher than the standard grade corresponding to immediate replacement, the anomaly grade determination unit (340) may generate control information including the identification information of the target battery and the anomaly grade and transmit it to an external system.

[0108] In addition, the external system may correspond to at least one of a Battery Management System (BMS), a Vehicle Management System, an Equipment Management System, a Maintenance Management System, or a Control Server. When the external system receives control information, the abnormal sign grade determination unit (340) may determine whether the equipment equipped with the target battery is operational or automatically select the equipment to be maintained. Furthermore, the abnormal sign grade determination unit (340) may provide information on the operation restriction of the target equipment and information on maintenance priority together so that maintenance on high-risk batteries is performed preferentially. For example, if abnormal signs occur in multiple batteries at the same time, the abnormal sign grade determination unit (340) may provide information to the external system to determine the maintenance priority based on the abnormal sign grade or risk information.

[0109] Additionally, the abnormal sign grade determination unit (340) can receive operation restriction results or maintenance assignment results from an external system and manage them in conjunction with the diagnostic history of the target battery. The abnormal sign grade determination unit (340) can be configured to store whether operation is restricted, whether maintenance is assigned, whether maintenance is completed, or whether the battery is replaced, along with the abnormal sign grade, so that this information can be utilized for verifying subsequent state diagnostic results or for generating retraining data for the self-reinforcing learning control unit (360). Furthermore, the abnormal sign grade determination unit (340) can control the re-evaluation of the abnormal sign grade of the battery or the performance of a new state diagnosis using the updated state information after the maintenance is completed. For example, when the abnormal sign grade determination unit (340) receives battery replacement completion information from an external system, it can terminate the existing abnormal sign history for the battery and create a new state diagnosis history for the replaced battery.

[0110] The life prediction unit (350) can generate multivariate degradation history time series data for life prediction that indicates the degradation trend of derived variables by cycle, and input the multivariate degradation history time series data for life prediction into a remaining life prediction model to calculate an inference value of RUL (Remaining Useful Life). The life prediction unit (350) receives time series data of derived variables provided by the adaptive derived variable calculation unit (330) and can classify the received derived variables by charge / discharge cycle units. The life prediction unit (350) can generate data items indicating the degradation trend by cycle by calculating the representative value, change amount, or cumulative value of the derived variable calculated in each cycle. In addition, the life prediction unit (350) can generate multivariate degradation history time series data for life prediction by reflecting the direction of change and the rate of change of the derived variable corresponding to each cycle. For example, the life prediction unit (350) can calculate the amount of change in cell voltage deviation, temperature change rate, phase transition peak characteristics or spatial temperature deviation for each charge / discharge cycle and arrange them according to the cycle order.

[0111] At this time, the lifespan prediction unit (350) can organize the changes in derived variables accumulated over multiple cycles into a single multivariate data structure to generate multivariate degradation history time series data for lifespan prediction. The lifespan prediction unit (350) can store cycle numbers, cycle-specific derived variable values, amounts of changes in derived variables, and degradation trend information in correspondence with each other. Additionally, if there is missing cycle data, the lifespan prediction unit (350) can be configured to maintain a continuous time series structure by performing interpolation processing or previous cycle data preservation processing. For example, if some derived variable values ​​are missing in a specific cycle, the lifespan prediction unit (350) can correct the input data of that cycle using the derived variable values ​​of an adjacent cycle.

[0112] Next, the life prediction unit (350) can preprocess the generated multivariate deterioration history time series data for life prediction so that it can be input into a remaining life prediction model. The life prediction unit (350) can normalize the range of values ​​for each derived variable, adjust the input length to the required format of the remaining life prediction model, and generate an input tensor such that multiple derived variables are arranged according to the same cycle standard. Additionally, the life prediction unit (350) can select at least one of the recent cycle interval or the entire cumulative cycle interval to use as input data for the remaining life prediction model. For example, the life prediction unit (350) can convert the multivariate deterioration history time series data for life prediction, including deterioration trends during multiple recent cycles, to fit the input format of the remaining life prediction model.

[0113] Additionally, the life prediction unit (350) can input preprocessed multivariate degradation history time series data for life prediction into a remaining life prediction model to calculate an RUL inference value. Here, the remaining life prediction model may include an LSTM, GRU, Transformer, or recurrent neural network-based model, and may correspond to a model that has learned the relationship between the long-term change pattern of cycle-specific derived variables and the degree of battery degradation progression. The remaining life prediction model can calculate an RUL inference value representing the remaining period or remaining cycles for which the battery is usable by analyzing the long-term change characteristics included in the input multivariate degradation history time series data for life prediction. Additionally, the life prediction unit (350) can record the calculated RUL inference value in a storage device and convert it into diagnostic result data that can be utilized by a user terminal or an external system. For example, the life prediction unit (350) can calculate an RUL inference value of the battery by inputting a degradation history in which the cell voltage deviation increases and the phase transition peak characteristics change during the cumulative charge / discharge process into the remaining life prediction model.

[0114] In one embodiment, the life prediction unit (350) can simultaneously calculate a RUL inference value and a confidence interval representing the uncertainty of the RUL inference value through Monte Carlo Dropout, which performs multiple inferences while maintaining a Dropout active state during the inference process of the remaining life prediction model. Here, the Dropout active state may correspond to a state in which a function is activated to probabilistically disable some nodes or some connections during the inference process of the remaining life prediction model; for example, it may correspond to a state in which different sets of nodes or sets of connections are selectively disabled each time inference is performed. Additionally, Monte Carlo Dropout may correspond to a technique that generates a distribution of output values ​​by repeatedly performing multiple inferences on the same input data while maintaining a Dropout active state, and quantifies the uncertainty of the inference value using the generated output distribution.

[0115] The life prediction unit (350) can be configured to maintain a dropout active state during the inference process of the remaining life prediction model. In this case, the remaining life prediction model can calculate RUL inference values ​​with different nodes or connections disabled for each inference, even if the same multivariate deterioration history time series data for life prediction is input, by probabilistically deactivating different nodes or connections each time inference is performed. Accordingly, the life prediction unit (350) can obtain multiple RUL inference values ​​under different inference conditions for a single input data. For example, the life prediction unit (350) can generate multiple RUL inference values ​​by repeatedly inputting the same deterioration history time series data into the remaining life prediction model and maintaining a dropout active state during each repeated inference process.

[0116] At this time, the lifespan prediction unit (350) can perform multiple inferences using Monte Carlo dropout. The lifespan prediction unit (350) can store the RUL inference values ​​calculated in each inference and generate an output distribution consisting of multiple RUL inference values. Additionally, the lifespan prediction unit (350) can determine the average or median of the RUL inference values ​​included in the output distribution as the representative RUL inference value. For example, the lifespan prediction unit (350) can perform iterative inference on the same multivariate degradation history time series data for lifespan prediction and calculate the representative RUL inference value using multiple RUL inference values ​​generated as a result of the iterative inference.

[0117] Furthermore, the life prediction unit (350) can calculate the uncertainty of the RUL inference value by using the degree of dispersion of multiple RUL inference values. The life prediction unit (350) can determine within what range the RUL inference value fluctuates by calculating the standard deviation, quantile value, or variance value of the output distribution. Additionally, the life prediction unit (350) can calculate the confidence interval of the RUL inference value using the calculated uncertainty. For example, the life prediction unit (350) can calculate a narrow confidence interval when the repeated inference results are concentrated in a narrow range, and calculate a wide confidence interval when the repeated inference results are dispersed over a wide range.

[0118] Additionally, the lifespan prediction unit (350) may store the calculated RUL inference value and the confidence interval in correspondence with each other and generate diagnostic result information that can be provided to the user terminal (110) or an external system. The lifespan prediction unit (350) may be configured to provide not only the RUL inference value but also the uncertainty of the RUL inference value so that it can be utilized in determining the battery replacement time or maintenance time. Additionally, if the confidence interval is wide, the lifespan prediction unit (350) may generate information indicating that the reliability of the RUL inference value is low or that additional status data collection is required. For example, the lifespan prediction unit (350) may display the representative RUL inference value and the confidence interval together on the user terminal (110) so that the user can check the reliability of the remaining lifespan prediction result.

[0119] The self-reinforcing learning control unit (360) can utilize the actual abnormal sign types and actual remaining life measurements fed back after maintenance is completed as labels, automatically input them into the retraining dataset of the abnormal sign detection model and the remaining life prediction model, and update the abnormal sign detection model or the remaining life prediction model using the retraining dataset. Here, self-reinforcing learning does not mean reinforcement learning, but may correspond to a self-feedback-based learning structure that improves model performance by repeatedly retraining the abnormal sign detection model and the remaining life prediction model using the actual abnormal sign types and actual remaining life measurements fed back after maintenance is completed as labels. The self-reinforcing learning control unit (360) can update the retraining dataset using newly obtained label information and retrain the abnormal sign detection model or the remaining life prediction model using the updated retraining dataset. The self-reinforcing learning control unit (360) can store the cause of failure identified during the maintenance process, information on replacement parts, the grade of abnormal signs at the time of diagnosis, and the actual remaining life measurement values ​​by correlating them with each other. Additionally, the self-reinforcing learning control unit (360) can convert the collected actual abnormal sign types and actual remaining life measurement values ​​into labels and combine them with existing status data, derived variables, and diagnosis results. For example, the self-reinforcing learning control unit (360) can assign the cell imbalance abnormal type identified after maintenance is completed and the actual remaining life measurement values ​​as labels corresponding to the past derived variable time series data of the battery.

[0120] At this time, the self-reinforcing learning control unit (360) can automatically input labeled data into the retraining datasets of the anomaly detection model and the remaining lifespan prediction model. The self-reinforcing learning control unit (360) can correspond actual anomaly types to the input data for the retraining of the anomaly detection model and actual remaining lifespan measurements to the input data for the retraining of the remaining lifespan prediction model. Additionally, the self-reinforcing learning control unit (360) can select only data suitable for retraining by checking for duplication, omission, and the validity of the labels of the data included in the retraining dataset. For example, the self-reinforcing learning control unit (360) can include only data in the retraining dataset where both maintenance history and actual measurements exist, and exclude data where actual anomaly types have not been identified from the retraining target.

[0121] Furthermore, the self-reinforcing learning control unit (360) can update an anomaly detection model or a remaining lifespan prediction model using a retraining dataset. Before updating the model, the self-reinforcing learning control unit (360) can check the performance indicators of the existing model and the distribution of the data to be retrained, and verify whether the performance of the updated model after retraining satisfies the standard conditions. Additionally, if the updated model shows higher accuracy or lower prediction error than the existing model, the self-reinforcing learning control unit (360) can register the updated model as an operational model. For example, the self-reinforcing learning control unit (360) can update the anomaly detection model using a retraining dataset with actual anomaly types added, and update the remaining lifespan prediction model using a retraining dataset with actual remaining lifespan measurements added.

[0122] Additionally, the self-reinforcing learning control unit (360) can store the model update history in the database (150) and control the control unit (370) to apply the updated model to a subsequent diagnostic process. The self-reinforcing learning control unit (360) can be configured to allow tracking of the model update process by storing the model version, the time of creation of the retraining dataset, the label information used, and the update results. Additionally, the self-reinforcing learning control unit (360) can store the updated anomaly detection model or remaining lifespan prediction model in the memory (230) so that it can be used in the subsequent inference process of the battery state diagnostic device (130). For example, the self-reinforcing learning control unit (360) can register the updated model in the database (150) and control the control unit (370) to perform anomaly detection or RUL inference using the updated model for state data collected thereafter.

[0123] Additionally, the self-reinforcing learning control unit (360) can be configured to apply the latest model in the subsequent battery condition diagnosis process by providing the anomaly detection model updated through relearning to the anomaly grade determination unit (340) and providing the updated remaining life prediction model to the life prediction unit (350). The self-reinforcing learning control unit (360) can repeatedly perform model updates using actual anomaly types and actual remaining life measurements obtained after maintenance is completed, and the updated model can be reapplied to the anomaly diagnosis and RUL inference processes for subsequently collected condition data. Accordingly, the battery condition diagnosis device (130) can form a feedback structure in which condition data collection, anomaly diagnosis, life prediction, and model relearning are repeatedly performed. For example, the self-reinforcing learning control unit (360) can update the model using newly obtained maintenance history and improve the accuracy of the subsequent battery condition diagnosis process using the updated model.

[0124] In one embodiment, the self-reinforcing learning control unit (360) can periodically monitor changes in the distribution of data included in the retraining dataset and changes in model performance, and determine the timing for performing retraining. The self-reinforcing learning control unit (360) can determine whether a model update is necessary by comparing the number of newly accumulated data with labels, the inference accuracy of the anomaly detection model, and the prediction error of the remaining lifespan prediction model with reference values. Additionally, the self-reinforcing learning control unit (360) can control the performance of retraining when a decrease in model performance is detected or when a certain number of new data is accumulated, and can maintain the continuity of the model by utilizing the existing retraining dataset and the new data together during the retraining process. For example, the self-reinforcing learning control unit (360) can perform retraining of the anomaly detection model or the remaining lifespan prediction model when the number of maintenance history and actual remaining lifespan measurements accumulated over a certain period is greater than or equal to a reference number, or when the prediction error of the remaining lifespan prediction model exceeds a reference range.

[0125] The control unit (370) controls the overall operation of the battery condition diagnostic device (130) and can manage the control flow or data flow between the data collection unit (310), the chemical type identification control unit (320), the adaptive derived variable calculation unit (330), the abnormal sign grade determination unit (340), the lifespan prediction unit (350), and the self-reinforcing learning control unit (360).

[0127] FIG. 4 is a flowchart illustrating the functional configuration of a battery status diagnostic device according to the present invention.

[0128] Referring to FIG. 4, the battery state diagnostic device (130) can collect state data including voltage, temperature, and current from the battery through the data collection unit (310) (step S410). Here, the battery state diagnostic device (130) can store the aligned state data in a memory area and then provide it so that a component performing subsequent processing can use the state data. Additionally, the battery state diagnostic device (130) can identify the chemical type of the battery by calculating the nominal voltage range of the battery or the flatness ratio of the OCV-SOC (Open Circuit Voltage - State of Charge) curve based on the state data based on the chemical type identification control unit (320), and determine an analysis path corresponding to the identified chemical type (step S430). Here, the battery state diagnostic device (130) can identify the battery as a battery of the first chemical type if the unit cell nominal voltage range of the battery corresponds to a first voltage range, and identify the battery as a battery of the second chemical type if the unit cell nominal voltage range corresponds to a second voltage range different from the first voltage range.

[0129] The battery condition diagnostic device (130) can calculate different derivative variables that are predefined to reflect the characteristics of each chemical type according to an analysis path determined based on the adaptive derivative variable calculation unit (330) (step S450). Here, if the battery is identified as a Nickel Manganese Cobalt (NMC) battery, the battery condition diagnostic device (130) can calculate a first derivative variable including at least one of a cell voltage deviation representing the voltage difference between multiple cells and a temperature change rate representing the amount of temperature change over time, and if the battery is identified as a Lithium Iron Phosphate (LFP) battery, it can calculate a second derivative variable including the phase transition peak characteristics of the voltage change rate (dQ / dV) curve for unit capacity change during charging and discharging and the spatial temperature deviation between multiple cells.

[0130] Additionally, the battery condition diagnostic device (130) can determine the abnormality grade of the battery by inputting multivariate input data for abnormality detection, including time series data of derived variables, into an abnormality detection model based on the abnormality grade determination unit (340) and performing inference of abnormalities (step S470). Here, the battery condition diagnostic device (130) can generate an abnormal event in the multivariate input data for abnormality detection when the peak position of the phase transition peak characteristic included in the second derived variable moves by more than a preset ratio relative to the normal standard in the analysis path where the battery is identified as an LFP battery, or when the peak height of the phase transition peak characteristic decreases by more than a preset ratio relative to the normal standard, and can determine the abnormality grade of the battery based on the abnormal event.

[0131] The battery condition diagnostic device (130) can generate multivariate degradation history time series data for life prediction that indicates the cycle-by-cycle degradation trend of derived variables based on the life prediction unit (350), and input the multivariate degradation history time series data for life prediction into a remaining life prediction model to calculate an inference value of RUL (Remaining Useful Life) (step S490). Here, the battery condition diagnostic device (130) can simultaneously calculate the RUL inference value and a confidence interval representing the uncertainty of the RUL inference value through Monte Carlo Dropout, which maintains a Dropout active state and performs multiple inferences during the inference process of the remaining life prediction model.

[0133] FIG. 5 is a flowchart illustrating the process of adaptively diagnosing abnormal signs and predicting remaining lifespan according to the battery chemical type of the battery condition diagnostic device according to the present invention.

[0134] Referring to FIG. 5, the battery state diagnostic device (130) can perform a state data collection step (S510). In the state data collection step (S510), the battery state diagnostic device (130) can collect state data including voltage, temperature, and current from the battery. The battery state diagnostic device (130) can receive state data from a plurality of sensors or battery management systems and generate a data set available for use in a subsequent analysis process by arranging the collected state data based on the same time reference. Additionally, the battery state diagnostic device (130) can be configured to enable time-series analysis by storing collection time information corresponding to the state data together.

[0135] Next, the battery condition diagnostic device (130) can perform a chemical type identification step (S520). In the chemical type identification step (S520), the battery condition diagnostic device (130) can calculate the nominal voltage range of the battery or the flatness ratio of the OCV-SOC curve using the condition data. The battery condition diagnostic device (130) can identify the chemical type of the battery by comparing the calculated nominal voltage range or the flatness ratio of the OCV-SOC curve with reference information for each chemical type stored in advance. Additionally, the battery condition diagnostic device (130) can determine an analysis path corresponding to the identified chemical type.

[0136] Next, the battery status diagnostic device (130) can perform a judgment step (S530). In the judgment step (S530), the battery status diagnostic device (130) can determine whether the battery is an LFP battery using the result of the chemical type identification step (S520). If the battery status diagnostic device (130) determines that it is an LFP battery, it can perform a second derived variable calculation step (S550), and if it is not an LFP battery, it can perform a first derived variable calculation step (S540). Accordingly, the battery status diagnostic device (130) can control the generation of derived variables that reflect the characteristics of each chemical type.

[0137] If it is determined that the battery is not an LFP battery, the battery state diagnostic device (130) may perform a first derived variable calculation step (S540). In the first derived variable calculation step (S540), the battery state diagnostic device (130) may calculate a first derived variable from the state data, which includes at least one of a cell voltage deviation representing the voltage difference between multiple cells and a temperature change rate representing the amount of temperature change over time. The battery state diagnostic device (130) may arrange the calculated first derived variable in chronological order and generate a data structure in the form of a time series so that it can be used in a subsequent diagnostic process.

[0138] On the other hand, if the battery is determined to be an LFP battery, the battery condition diagnostic device (130) may perform a second derived variable calculation step (S550). In the second derived variable calculation step (S550), the battery condition diagnostic device (130) may calculate a second derived variable including the phase transition peak characteristics of the voltage change rate (dQ / dV) curve for unit capacity change during charging and discharging, and the spatial temperature deviation between multiple cells. Additionally, the battery condition diagnostic device (130) may be configured to generate diagnostic data reflecting the electrochemical characteristics of the LFP battery by storing the calculated second derived variable in a time series form.

[0139] Subsequently, the battery condition diagnostic device (130) may perform a step of generating multivariate input data for anomaly detection (S560). In the step of generating multivariate input data for anomaly detection (S560), the battery condition diagnostic device (130) may generate multivariate input data for anomaly detection using time series data of a first derived variable or a second derived variable. The battery condition diagnostic device (130) may preprocess the generated multivariate input data for anomaly detection to match the input format of an anomaly detection model and may generate a data structure to maintain temporal correlation between multiple derived variables.

[0140] Subsequently, the battery condition diagnostic device (130) may perform an abnormality grade determination step (S570). In the abnormality grade determination step (S570), the battery condition diagnostic device (130) may perform inference of abnormalities by inputting multivariate input data for abnormality detection into an abnormality detection model. The battery condition diagnostic device (130) may calculate an abnormality type, an abnormality probability, or a risk score using the inference results of the abnormality detection model, and may determine the abnormality grade of the battery using the calculated results. Additionally, the battery condition diagnostic device (130) may control the generation or output of an alarm signal if the determined abnormality grade is higher than or equal to a reference grade.

[0141] Finally, the battery condition diagnostic device (130) can perform a step for calculating RUL inference values ​​(S580). In the step for calculating RUL inference values ​​(S580), the battery condition diagnostic device (130) can generate multivariate degradation history time series data for life prediction that indicates the degradation trend of derived variables per cycle. The battery condition diagnostic device (130) can calculate the RUL inference value of the battery by inputting the generated multivariate degradation history time series data for life prediction into a remaining life prediction model. Additionally, the battery condition diagnostic device (130) can calculate the RUL inference value and a confidence interval representing the uncertainty of the RUL inference value by performing multiple inferences using Monte Carlo dropout. Accordingly, the battery condition diagnostic device (130) can comprehensively diagnose the battery condition by adaptively generating derived variables according to the chemical type of the battery and performing abnormal sign diagnosis and remaining life prediction.

[0143] Although the present invention has been described above with reference to preferred embodiments, those skilled in the art will understand that various modifications and changes can be made to the invention without departing from the spirit and scope of the invention as described in the following claims. Explanation of the symbols

[0145] 100: Battery Status Diagnostic System 110: User terminal 130: Battery Status Diagnostic Device 150: Database 210: Processor 230: Memory 250: User I / O Section 270: Network I / O Section 290: Communication port section 310: Data collection unit 320: Chemical type identification control unit 330: Adaptive Derived Variable Calculation Unit 340: Anomaly Sign Grade Determination Unit 350: Life Prediction Unit 360: Self-Reinforcing Learning Control Unit 370: Control unit

Claims

Claim 1 A battery chemical type adaptive multivariate data-based battery condition diagnosis device comprising: a data collection unit that collects state data including voltage, temperature, and current from a battery; a chemical type identification control unit that identifies the chemical type of the battery by calculating the flatness ratio of the nominal voltage range or OCV-SOC (Open Circuit Voltage - State of Charge) curve of the battery based on the state data and determines an analysis path corresponding to the identified chemical type; an adaptive derived variable calculation unit that calculates different derived variables predefined to reflect the characteristics of each chemical type according to the determined analysis path; an anomaly grade determination unit that determines the anomaly grade of the battery by inputting multivariate input data for anomaly detection, including time series data of the derived variables, into an anomaly detection model and performing anomaly inference; and a life prediction unit that generates multivariate degradation history time series data for life prediction indicating the degradation trend of the derived variables per cycle and inputs the multivariate degradation history time series data for life prediction into a remaining life prediction model to calculate an RUL (Remaining Useful Life) inference value. Claim 2 A battery chemical type adaptive multivariate data-based battery condition diagnostic device according to claim 1, wherein the chemical type identification control unit identifies the battery as a battery of a first chemical type when the unit cell nominal voltage range of the battery corresponds to a first voltage range, identifies the battery as a battery of a second chemical type when the unit cell nominal voltage range corresponds to a second voltage range different from the first voltage range, and identifies the chemical type by additionally using the flatness ratio of the OCV-SOC curve when the identification of the chemical type is uncertain based solely on the nominal voltage range. Claim 3 A battery chemical type adaptive multivariate data-based battery condition diagnosis device according to claim 1, wherein the adaptive derived variable calculation unit calculates a first derived variable including at least one of a cell voltage deviation representing a voltage difference between multiple cells and a temperature change rate representing a temperature change amount over time when the battery is identified as a Nickel Manganese Cobalt (NMC) battery, and calculates a second derived variable including a phase transition peak characteristic of a voltage change rate (dV / dQ) curve for a unit capacity change during charging and discharging and a spatial temperature deviation between multiple cells when the battery is identified as a Lithium Iron Phosphate (LFP) battery. Claim 4 A battery chemical type adaptive multivariate data-based battery condition diagnostic device according to claim 1, wherein the abnormal sign grade determination unit generates an abnormal event in the multivariate input data for abnormal detection when the peak position of a phase transition peak characteristic extracted from charge / discharge characteristics in an analysis path where the battery is identified as an LFP battery moves by more than a preset ratio relative to a normal reference or when the peak height of the phase transition peak characteristic decreases by more than a preset ratio relative to a normal reference, and determines the abnormal sign grade of the battery based on the abnormal event. Claim 5 A battery chemical type adaptive multivariate data-based battery condition diagnosis device according to claim 1, wherein the abnormal sign grade determination unit generates a rule-based detection result by applying an adaptive threshold value that is dynamically adjusted according to the chemical type of the battery and the cumulative number of charge / discharge cycles to the derived variable, and obtains a model-based detection result including at least one of the abnormal sign type, abnormal probability, or risk score of the battery from the abnormal sign detection model, determines the abnormal sign grade of the battery using the rule-based detection result and the model-based detection result, maintains, up-corrects, or down-corrects the abnormal sign grade of the battery according to whether the rule-based detection result and the model-based detection result match or the degree of difference, and controls to generate or output an alarm signal within a preset response time if the corrected abnormal sign grade is higher than or equal to a reference grade. Claim 6 A battery chemical type adaptive multivariate data-based battery condition diagnostic device according to claim 1, wherein the abnormal sign grade determining unit automatically transmits a signal to an external system for restricting operation or assigning maintenance of the target equipment equipped with the battery when the abnormal sign grade is determined to be higher than a standard grade corresponding to immediate replacement. Claim 7 A battery chemical type adaptive multivariate data-based battery condition diagnostic device, wherein, in claim 1, the life prediction unit simultaneously calculates the RUL inference value and a confidence interval representing the uncertainty of the RUL inference value through Monte Carlo Dropout, which maintains a Dropout active state and performs multiple inferences during the inference process of the remaining life prediction model. Claim 8 A battery chemical type adaptive multivariate data-based battery condition diagnosis device according to claim 1, further comprising a self-reinforcing learning control unit that utilizes actual abnormal sign types and actual remaining life measurements fed back after maintenance is completed as labels, automatically inputs them into a retraining dataset of the abnormal sign detection model and the remaining life prediction model, and updates the abnormal sign detection model or the remaining life prediction model using the retraining dataset. Claim 9 A battery chemical type adaptive multivariate data-based battery condition diagnostic device according to claim 1, characterized in that the battery is applied to a battery mounted on at least one of an electric two-wheeled vehicle, an electric forklift, an electric cart, an energy storage device (ESS), an unmanned delivery robot, and industrial mobile equipment. Claim 10 A battery condition diagnosis method based on adaptive multivariate data for battery chemical type, performed in a battery condition diagnosis device based on adaptive multivariate data for battery chemical type, comprising: a data collection step of collecting condition data including voltage, temperature, and current from a battery through a data collection unit; a chemical type identification control step of identifying the chemical type of the battery and determining an analysis path corresponding to the identified chemical type by calculating the flatness ratio of the nominal voltage range or OCV-SOC (Open Circuit Voltage - State of Charge) curve of the battery based on the condition data through a chemical type identification control unit; an adaptive derived variable calculation step of calculating different derived variables predefined to reflect the characteristics of each chemical type according to the determined analysis path through an adaptive derived variable calculation unit; and an anomaly grade determination step of determining the anomaly grade of the battery by inputting multivariate input data for anomaly detection, including time series data of the derived variables, into an anomaly detection model and performing inference of an anomaly. A battery condition diagnosis method based on multivariate data adaptive to battery chemical type. The method comprises a life prediction step of generating multivariate degradation history time series data for life prediction that indicates the cycle-by-cycle degradation trend of the derived variable through a life prediction unit, and inputting the multivariate degradation history time series data for life prediction into a remaining life prediction model to calculate an inferred RUL (Remaining Useful Life).

Citation Information

Patent Citations

  • Battery overcharge early warning method and system based on big data

    CN110148991A

  • Battery type identification and detection method and system based on charging and discharging characteristics

    CN119986415A

  • Residual value reading method and residual value reading device for determining whether battery cells can be reused after use

    KR1020240077618A

  • Method and System for Grading a Battery Based on SOH

    KR1020260059248A

  • ESS battery state diagnosis and lifespan prediction device and method

    WO2021040236A1