Artificial Intelligence-Based Battery Fault Prediction System for Electric Vehicles

Abstract

The present invention relates to an artificial intelligence-based electric vehicle battery defect prediction system that uses an artificial intelligence model based on operational data and environmental data of an electric vehicle battery to predict in advance the possibility of abnormal conditions and defects occurring in the battery, and calculates the remaining lifespan to determine the optimal maintenance time. More specifically, the system is characterized by comprising: a data collection unit that collects operational data including voltage, current, temperature, State of Charge (SOC), State of Health (SOH), and charge / discharge cycle information measured at the battery cell or module level; a data preprocessing unit that receives data transmitted from the data collection unit, performs noise removal, missing value correction, and time series synchronization, and generates input features that reflect the battery condition; an artificial intelligence prediction unit that predicts the probability of defect occurrence and remaining useful life (RUL) of the battery using a deep learning model based on the data processed by the data preprocessing unit; a judgment unit that classifies the battery condition into normal, warning, or dangerous states and determines the maintenance time based on the output result of the artificial intelligence prediction unit; and an output unit that provides the judgment result obtained through the judgment unit to a user or an external system.

Description

Technology Field

[0001] The present invention relates to an artificial intelligence-based electric vehicle battery defect prediction system that uses an artificial intelligence model based on operational and environmental data of an electric vehicle battery to predict in advance the possibility of abnormal conditions and defects occurring in the battery, calculates the remaining lifespan, and determines the optimal maintenance time. Background Technology

[0002] Generally, electric vehicles are powered by lithium-ion batteries, and the performance and safety of the battery are directly linked to the reliability of the entire vehicle.

[0003] However, conventional battery management systems (BMS) rely primarily on simple monitoring of voltage, current, and temperature to determine abnormalities, and thus have limitations in accurately predicting internal battery degradation or potential defects in advance.

[0004] In particular, although battery degradation occurs due to a combination of various factors such as charge / discharge patterns, temperature environments, and driving habits, conventional technology fails to analyze such multivariate data in an integrated manner, resulting in the problem of inefficiently setting maintenance timing. Prior art literature

[0005] Republic of Korea Registered Patent No. 2835355 (Registration Date: July 14, 2025) Republic of Korea Registered Patent No. 2864750 (Registration Date: September 22, 2025) The problem to be solved

[0006] To solve the aforementioned conventional problems, the present invention aims to provide an AI-based electric vehicle battery defect prediction system that uses artificial intelligence based on various operational data of an electric vehicle battery to predict the probability of defect occurrence and remaining lifespan of the battery, and thereby automatically calculates the optimal maintenance timing. means of solving the problem

[0007] As a means of solving the problem of the aforementioned conventional problems, the artificial intelligence-based electric vehicle battery defect prediction system according to the present invention comprises: a data collection unit (100) that collects operational data including voltage, current, temperature, State of Charge (SOC), State of Health (SOH), and charge / discharge cycle information measured at the battery cell or module unit; a data preprocessing unit (200) that receives data transmitted from the data collection unit (100), performs noise removal, missing value correction, and time series synchronization, and generates input features reflecting the battery state; an artificial intelligence prediction unit (300) that predicts the probability of battery defect occurrence and remaining useful life (RUL) using a deep learning model based on the data processed by the data preprocessing unit (200); a judgment unit (400) that classifies the battery state into normal, warning, or dangerous states and determines the maintenance timing based on the output result of the artificial intelligence prediction unit (300); and an output unit (500) that provides the judgment result obtained through the judgment unit (400) to a user or an external system. It is characterized by the fact that

[0008] In addition, the data collection unit (100) additionally collects vehicle driving pattern information and external environment data and transmits them to the data preprocessing unit (200), and the data preprocessing unit (200) preferably generates a derived feature representing the battery state, including a voltage change rate, a temperature rise rate, an estimated internal resistance value, and a charge / discharge efficiency, and provides it as an input feature for an artificial intelligence model.

[0009] In addition, the artificial intelligence prediction unit (300) includes a Long Short-Term Memory (LSTM) model or a transformer-based model for learning time series data, and it is desirable to learn the degradation progression pattern and internal resistance change trend of the battery.

[0010] In addition, the artificial intelligence prediction unit (300) preferably includes an Explainable Artificial Intelligence (XAI) function to analyze the correlation between input features and prediction results and derive the main cause of the defect occurrence. Effects of the invention

[0011] According to the present invention, a data collection unit, a data preprocessing unit, an artificial intelligence prediction unit, a judgment unit, and an output unit are organically linked and operated in a manner distinct from conventional methods, thereby providing a technical effect of predicting battery defects in advance and maximizing maintenance efficiency. Brief explanation of the drawing

[0012] FIG. 1 is a block diagram illustrating an artificial intelligence-based electric vehicle battery defect prediction system according to the present invention. Figure 2 is a conceptual diagram between the data collection unit and the data preprocessing unit for Figure 1. Figure 3 is a conceptual diagram between the data preprocessing unit and the artificial intelligence prediction unit for Figure 1. FIG. 4 is a conceptual diagram of the artificial intelligence prediction unit, judgment unit, and output unit for FIG. 1. FIG. 5 is a graph illustrating the maintenance timing by determining the battery status based on the judgment unit for FIG. 4. Specific details for implementing the invention

[0013] Various embodiments are described in more detail below with reference to the attached drawings. The embodiments described in this specification may be modified in various ways. Specific embodiments may be depicted in the drawings and described in detail in the detailed description. However, specific embodiments disclosed in the attached drawings are intended only to facilitate understanding of various embodiments. Therefore, the technical concept is not limited by specific embodiments disclosed in the attached drawings, and it should be understood that it includes all equivalents or substitutions that fall within the spirit and scope of the invention.

[0015] Hereinafter, an artificial intelligence-based electric vehicle battery defect prediction system according to the present invention (hereinafter briefly referred to as the 'system') will be described in detail with reference to the attached drawings.

[0016] First, the system (1) according to the present invention includes a data collection unit (100), a data preprocessing unit (200), an artificial intelligence prediction unit (300), a judgment unit (400), and an output unit (500), as shown in FIG. 1, and is configured to be linked with a battery management system (BMS) equipped in an electric vehicle through the organic combination relationship of the above-mentioned components, and can determine in advance the possibility of a battery defect occurring by collecting, analyzing, and predicting various operational data generated in real-time or periodically at the unit of a battery cell or module mounted in the vehicle.

[0017] To explain in more detail, the data collection unit (100) is a component that collects various sensor data reflecting the physical and operational states of the battery, and acquires operational data including voltage, current, and temperature data measured at the battery cell or module level, as well as the State of Charge (SOC), State of Health (SOH) indicating the degree of degradation of the battery, and charge / discharge cycle information.

[0018] The data collection unit (100) can be configured to additionally collect driving pattern information of the vehicle (e.g., rapid acceleration, rapid deceleration, average driving speed) and external environment data (e.g., ambient temperature, humidity, etc.) to comprehensively reflect various factors affecting battery degradation.

[0019] The data collection unit (100) in the present invention transmits acquired operation data, operation pattern information, and external environment data to the data preprocessing unit (200) (see FIG. 2).

[0020] And, the data preprocessing unit (200) uses the information transmitted through the data collection unit (100) described above as input features to correct or replace noise and missing values ​​included in the input features, and performs synchronization of the data based on the time axis to ensure consistency of the time series data, as illustrated in FIGS. 2 and FIGS. 3.

[0021] At this time, the data preprocessed through the data preprocessing unit (200) is used as an input feature of the artificial intelligence prediction unit (300) described later.

[0022] The data preprocessing unit (200) can generate derived features such as voltage change rate, temperature rise rate, internal resistance estimate, and charge / discharge efficiency in order to generate input features that can effectively reflect the battery state.

[0023] At this time, the above-mentioned derived feature can be configured to reflect the temporal change pattern between different physical quantities, and, for example, can be generated as a composite feature reflecting a combined indicator of the rate of change of voltage and the rate of temperature rise in a specific time interval, or a correlation between the rate of increase of internal resistance and the rate of decrease of capacity according to the charge-discharge cycle.

[0024] For example, the data preprocessing unit (200) above is a time interval Rate of change of voltage during ( ) and temperature rise rate( By combining ), the following Degradation Acceleration Index (DAI) can be calculated.

[0025]

[0026] Here, and is a weight that reflects the influence of each physical quantity.

[0027] In addition, the data preprocessing unit (200) has an internal resistance increase rate according to the charge / discharge cycle ( ) and battery capacity reduction rate( By combining ), the following Composite Degradation Index (CDI) can be calculated.

[0028]

[0029] Here, and is a weight that reflects the contribution of each degradation factor.

[0030] This is used as an input feature for the artificial intelligence model to improve prediction accuracy and enhance the model's generalization performance.

[0031] At this time, the data preprocessing unit (200) does not stop at calculating simple individual variables, but calculates a battery degradation acceleration index based on the correlation between the voltage change rate and the temperature rise rate over time intervals, and can generate a composite degradation feature by combining the internal resistance increase rate and capacity decrease rate according to the charge / discharge cycle. In addition, the derived feature is configured to reflect the interdependence between multiple input data, so that the abnormal state of the battery can be expressed more precisely compared to a feature based on a single physical quantity.

[0032] Furthermore, in the aforementioned preprocessing process, data scaling and normalization are performed to ensure the stability of model training.

[0033] That is, the data preprocessing unit (200) can be configured to improve the learning stability and prediction accuracy of the artificial intelligence model by scaling the input features to a predetermined range or normalizing them based on the mean and variance in order to prevent learning bias and convergence instability that may occur due to scale imbalance between input features having different units and ranges.

[0034] In addition, the artificial intelligence prediction unit (300) may be configured to include a deep learning-based model suitable for time series data analysis, as illustrated in FIG. 3 and FIG. 4.

[0035] For example, changes in battery state over time can be learned by applying a Long Short-Term Memory (LSTM) structure or a transformer-based model, and through this, the degree of battery degradation, the trend of increasing internal resistance, and the pattern of decreasing effective capacity can be comprehensively analyzed.

[0036] As a result, the artificial intelligence prediction unit (300) can quantitatively calculate the remaining useful life (RUL) along with the probability of defect occurrence in the battery.

[0037] In particular, according to an embodiment of the present invention, the artificial intelligence prediction unit (300) has a plurality of artificial intelligence models having different structures or parameters depending on the type of battery (e.g., lithium-ion, lithium iron phosphate, etc.), usage environment, or operating conditions, and can selectively apply the optimal model according to the input features.

[0038] At this time, the artificial intelligence prediction unit (300) may be configured to select at least one of a plurality of artificial intelligence models or apply a weighted combination of a plurality of models according to predefined selection criteria including the distribution of input features, the length of time series data, or the operating conditions of the battery, thereby reducing deviations in prediction performance and deriving stable prediction results even in various operating environments.

[0039] This configuration has the effect of significantly improving prediction accuracy compared to conventional technology using a single model, and can ensure universality for various vehicle and battery conditions.

[0040] And, the above-described judgment unit (400) receives result information output from the above-described artificial intelligence prediction unit (300) and classifies the battery status into multiple grades based on the probability of defect occurrence and remaining lifespan.

[0041] According to the embodiment, the judgment unit (400) can classify the battery state into a normal state, a warning state, and a danger state, and can pre-set or dynamically adjust a threshold corresponding to each state.

[0042] In addition, the above-mentioned judgment unit (400) calculates the optimal maintenance time by considering the remaining lifespan and usage pattern, thereby preventing unnecessary early replacement and simultaneously preventing sudden breakdowns.

[0043] In addition, the above judgment unit (400) may be configured not only to simply compare the probability of a defect occurring and the remaining lifespan, but also to dynamically adjust the threshold value by simultaneously considering the change trend of the probability of a defect occurring and the remaining lifespan, thereby allowing the maintenance timing to be determined differentially according to the rate of increase of the probability of a defect occurring, even when having the same remaining lifespan.

[0044] As illustrated in FIG. 5 as an example of the judgment unit (400), the performance of the battery decreases non-linearly over time or usage cycles, and the judgment unit (400) sets a Normal Zone, a Warning Zone, and a Danger Zone based on the performance change curve, and can determine the maintenance time based on the time of entry into the Warning Zone or the remaining lifespan.

[0045] And, the output unit (500) performs the function of transmitting the judgment result of the judgment unit (400) described above to a user or an external system.

[0046] For example, information regarding the battery status and recommended maintenance times may be provided to the driver in a visual or auditory form through a display installed inside the vehicle or a mobile application installed on a mobile communication device (e.g., a smartphone), and may be configured to provide an immediate warning notification if a dangerous condition is determined.

[0047] In addition, the output unit (500) can transmit data by linking with a vehicle manufacturer server or a maintenance management system to enable remote monitoring and preventive maintenance services.

[0048] Furthermore, the system of the present invention may include a structure linked with a cloud server and can continuously train and update an artificial intelligence model by integrally analyzing data collected from multiple vehicles.

[0049] This cloud-based learning architecture enhances the model's generalization performance by reflecting data from various driving conditions and environments, and provides prediction results optimized for individual vehicles.

[0050] Meanwhile, the artificial intelligence prediction unit (300) in the present invention, as described above, does not merely perform the function of predicting the possibility of a defect occurring in the battery, but may additionally include an Explainable Artificial Intelligence (XAI) function that analyzes the cause of the defect occurrence in other embodiments.

[0051] Specifically, the artificial intelligence prediction unit (300) may be configured to analyze the correlation between input features and prediction results to identify key features that affect the occurrence of defects among the battery's operation data, and thereby derive the main causal factors that caused a specific defect to occur.

[0052] For example, the artificial intelligence prediction unit (300) can quantitatively evaluate which of the multiple input features, such as battery temperature rise pattern, rapid charging frequency, depth of discharge, and voltage deviation between cells, has the greatest influence on the probability of defect occurrence, and can determine the cause of the defect based on this.

[0053] In addition, the above-mentioned explainable artificial intelligence function is not limited to calculating the importance of a single input feature, but can be configured to analyze contribution patterns based on combinations of multiple input features to identify feature combinations corresponding to specific defect types, thereby enabling the distinction of defect types based on different causes even in the same defect occurrence state.

[0054] In addition, the artificial intelligence prediction unit (300) can be implemented by analyzing the internal decision process of the learned model to calculate the importance of each input feature, and, for example, can visually or numerically express the contribution to the defect prediction result by applying explainability techniques such as SHAP (Shapley Additive Explanation) values ​​or LIME (Local Interpretable Model-agnostic Explanations).

[0055] Accordingly, the judgment unit (400) may be configured not only to classify the state of the battery as normal, warning, or danger, but also to provide the cause of the state determined, and may generate cause information in the form of, for example, 'accelerated degradation due to a high-temperature environment', 'increase in internal resistance due to repeated rapid charging', or 'local abnormality due to imbalance of a specific cell'.

[0056] Furthermore, the output unit (500) can provide cause analysis information to a user or an external system along with the judgment result of the judgment unit (400) described above, thereby suggesting a more specific maintenance direction beyond a simple warning level.

[0057] For example, depending on the specific cause, customized response measures such as checking the cooling system, improving charging habits, or replacing specific cells can be suggested.

[0058] In addition, the system according to the present embodiment can continuously improve the accuracy of future defect prediction and the reliability of cause analysis by accumulating the above-mentioned cause analysis results in a database and learning defect cases of the same or similar patterns.

[0059] As described above, the system (1) according to the present invention is differentiated from the conventional system in that the data collection unit (100), data preprocessing unit (200), artificial intelligence prediction unit (300), judgment unit (400), and output unit (500) are organically linked and operated, thereby realizing the technical effect of predicting battery defects in advance and maximizing maintenance efficiency.

[0061] As described above, the present invention has been explained with specific details such as specific components, limited embodiments, and drawings; however, this is provided merely to aid in a more comprehensive understanding of the invention, and the invention is not limited to the above embodiments. A person skilled in the art can make various modifications and variations from this description.

[0062] Accordingly, the scope of the present invention should not be limited to the described embodiments, and all things equivalent to or having equivalent variations to the claims set forth below, as well as the claims set forth below, shall be considered to fall within the scope of the concept of the present invention. Explanation of the symbols

[0063] 1: Artificial intelligence-based electric vehicle battery defect prediction system according to the present invention 100: Data Collection Unit 200: Data Preprocessing Unit 300: Artificial Intelligence Prediction Department 400: Judgment Department 500: Output section

Claims

Claim 1 In an artificial intelligence-based electric vehicle battery defect prediction system for predicting the possibility of defect occurrence in a battery installed in an electric vehicle, the prediction system is configured to collect operational data including voltage, current, temperature, State of Charge (SOC), State of Health (SOH), and charge / discharge cycle information measured at the battery cell or module level, and additionally collect vehicle driving pattern information and external environment data to comprehensively reflect factors affecting battery degradation, and a data collection unit (100) that transmits the acquired operational data, driving pattern information, and external environment data to a data preprocessing unit (200); a data preprocessing unit (200) that receives data transmitted from the data collection unit (100), performs noise removal, missing value correction, and time series synchronization, generates input features reflecting the battery state, uses the information transmitted through the data collection unit (100) as input features to correct or replace noise and missing values ​​included in said input features, and performs data synchronization based on the time axis to ensure consistency of time series data; and the data processed by the data preprocessing unit (200) Based on a deep learning model, an artificial intelligence prediction unit (300) predicts the probability of defect occurrence and remaining useful life (RUL) of the battery, includes a Long Short-Term Memory (LSTM) model or a transformer-based model for time series data learning, and learns the degradation progression pattern and internal resistance change trend of the battery; a judgment unit (400) classifies the battery status into normal, warning, or dangerous states based on the output result of the artificial intelligence prediction unit (300) and determines the maintenance timing; and an output unit (500) provides the judgment result obtained through the judgment unit (400) to a user or an external system; and the data preprocessing unit (200) includes a time interval ( Rate of change of voltage during ) ) and temperature rise rate( By combining ), the Degradation Acceleration Index (DAI) as shown in [Mathematical Formula 1] below is calculated, and the rate of increase in internal resistance according to charge / discharge cycles ( ) and battery capacity reduction rate( By combining ), a Composite Degradation Index (CDI) as shown in [Equation 2] below is calculated and used as an input feature for an artificial intelligence model, and [Equation 1] (Here and is a weight reflecting the influence of each physical quantity)[Equation 2] (Here and The above-determined judgment unit (400) classifies the battery state into a normal state, a warning state, and a danger state, pre-sets or dynamically adjusts a threshold corresponding to each state, calculates an optimal maintenance time considering the remaining lifespan and usage patterns to prevent unnecessary early replacement and simultaneously prevent sudden failures, and dynamically adjusts the threshold considering the probability of defect occurrence and the trend of change in remaining lifespan to differentially determine the maintenance time according to the rate of increase of the probability of defect occurrence even when having the same remaining lifespan, sets a Normal Zone, a Warning Zone, and a Danger Zone based on a performance change curve of the battery that decreases non-linearly according to time or usage cycles, and determines the maintenance time based on the time of entry into the warning zone or the remaining lifespan. This characterizes an artificial intelligence-based electric vehicle battery defect prediction system. Claim 2 delete Claim 3 delete Claim 4 An artificial intelligence-based electric vehicle battery defect prediction system according to claim 1, wherein the artificial intelligence prediction unit (300) includes an Explainable Artificial Intelligence (XAI) function to analyze the correlation between input features and prediction results and derive the main cause of defect occurrence.

Images