Vehicle Fault Prediction System and Method

Abstract

The present invention relates to a vehicle failure prediction system that predicts vehicle failures based on time-series data collected from an electronic control unit (ECU) mounted on a vehicle, comprising: a data collection unit that collects sensor data, driving data, or diagnostic data from the electronic control unit; a data preprocessing unit that sorts the data collected from the data collection unit by time and converts it into time-series data by performing at least one of noise removal, missing value correction, or normalization; a time-series learning model unit configured to learn the temporal correlation of the time-series data; an anomaly detection model unit that calculates the degree of anomaly in the time-series data based on normal driving data; a failure prediction unit that calculates at least one of the probability of failure, the risk of failure, or the remaining lifespan by combining the learning result of the time-series learning model unit and the anomaly score of the anomaly detection model unit; and an output unit that provides the result of the failure prediction unit to a user. and a control unit connected to the data collection unit, data preprocessing unit, time series learning model unit, anomaly detection model unit, fault prediction unit, and output unit, respectively, and controlling the operation of the data collection unit, data preprocessing unit, time series learning model unit, anomaly detection model unit, fault prediction unit, and output unit; characterized by including

Description

Technology Field

[0001] The present invention relates to a vehicle failure prediction system and method that predicts the probability of failure and the remaining lifespan of a vehicle by analyzing time-series data collected from an electronic control unit (ECU) installed in a vehicle based on artificial intelligence. Background Technology

[0003] Recently, as the electrification and intelligence of automobiles, agricultural machinery, construction machinery, special vehicles, and eco-friendly vehicles have progressed rapidly, numerous electronic control units (ECUs) are being installed. These ECUs have established themselves as key elements controlling major vehicle functions, such as engine operation, transmission, braking, steering, battery management, and emission reduction.

[0004] These electronic control units continuously generate various sensor signals and status information during operation, and this information can be utilized as important data for maintaining vehicle performance and ensuring safety. However, conventional vehicle maintenance and diagnostic technologies have largely remained reactive, relying on warning lights or fault codes to identify problems only after a failure has actually occurred. This has led to issues such as unexpected breakdowns during operation, increased risk of safety accidents, unnecessary maintenance costs, and economic losses due to vehicle downtime.

[0005] In particular, despite the fact that recent vehicle systems have become more complex in structure and control logic, and that a single component failure can lead to a chain reaction of functional degradation, existing rule-based inspection methods or simple threshold comparison methods have limitations in effectively detecting subtle signs of abnormality that appear prior to a failure.

[0006] Furthermore, data generated during vehicle operation has the characteristic of changing continuously over time and exhibits various variability even in normal conditions depending on the driving environment, driving habits, age, and load conditions, making it difficult to clearly distinguish between normal and abnormal states.

[0007] Consequently, persistent issues have been raised regarding precursory phenomena that could lead to actual failures being misidentified as normal fluctuations, or conversely, unnecessary warnings being issued even when the system is in a normal state. Furthermore, in realistic operating environments, failure data is often insufficiently accumulated, and failure types and causes are frequently not clearly labeled; consequently, it is difficult to perform highly reliable failure predictions relying solely on simple statistical analysis or rule definitions. This problem is particularly pronounced in vehicles that operate for long periods and are significantly affected by inconsistent operating conditions and working environments, such as commercial vehicles, agricultural machinery, and construction equipment.

[0008] Meanwhile, across the automotive industry, the maintenance paradigm is shifting from reactive maintenance to predictive maintenance. There is a growing demand to predict the likelihood of failures in advance to plan maintenance schedules, reduce unnecessary parts replacements, and simultaneously improve vehicle uptime and safety. In particular, in vehicle management environments involving the simultaneous operation of multiple vehicles or remote management settings, there is an increasing need for technologies capable of monitoring the status of individual vehicles in real-time or near-real-time and evaluating failure risk from a long-term perspective.

[0009] Accordingly, there is growing interest in technology that systematically analyzes the vast amount of operational and status data generated by vehicles to predict future failure potential and remaining service life, going beyond simple diagnosis of current status.

[0010] However, to apply these prediction technologies to actual vehicle environments, there are numerous challenges to overcome, including ensuring data reliability, addressing communication constraints, security and privacy issues, scalability for various vehicle types and control structures, and the division of roles between vehicle terminals and external analysis environments.

[0011] In particular, security vulnerabilities that may arise during the data linkage process between the vehicle's interior and external environments are becoming a critical consideration, coupled with recently strengthened cybersecurity regulations. Furthermore, since data structures and operational methods vary depending on the vehicle manufacturer, model, and intended use, a technical foundation capable of broad application is required, rather than being limited to specific vehicles or environments.

[0012] Against this backdrop, there is a continuous demand for new approaches that can effectively utilize various time-series data accumulated during vehicle operation to detect failure precursors early and simultaneously improve vehicle maintenance efficiency and safety. Prior art literature

[0014] Republic of Korea Registered Patent No. 10-1371312 (March 10, 2014) Republic of Korea Registered Patent No. 10-2552699 (July 10, 2023) The problem to be solved

[0015] Accordingly, the present invention has been devised to solve the above-mentioned problems, and the objective of the present invention is to provide a vehicle failure prediction system and method capable of analyzing time-series data collected from an electronic control unit mounted on a vehicle to recognize abnormal signs appearing in advance prior to the actual occurrence of a failure and to predict the likelihood of a vehicle failure.

[0016] Another objective of the present invention is to provide a vehicle failure prediction system and method that possess scalability and practicality, applicable to various vehicles with different vehicle types, operating conditions, and usage environments, by providing a prediction technology applicable even in environments where sufficient failure data is not available.

[0018] However, the technical problems to be solved by the present invention are not limited to those mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art to which the present invention belongs from the description below. means of solving the problem

[0020] The present invention was created to improve upon the problems of the prior art described above. In a vehicle failure prediction system that predicts vehicle failures based on time-series data collected from an electronic control unit (ECU) mounted on a vehicle, the system comprises: a data collection unit that collects sensor data, driving data, or diagnostic data from the electronic control unit; a data preprocessing unit that sorts the data collected from the data collection unit by time and converts it into time-series data by performing at least one of noise removal, missing value correction, or normalization; a time-series learning model unit configured to learn the temporal correlation of the time-series data; an anomaly detection model unit that calculates the degree of anomaly in the time-series data based on normal driving data; a failure prediction unit that calculates at least one of the probability of failure, the risk of failure, or the remaining lifespan through an anomaly score from the anomaly detection model unit, which determines whether the input time-series data deviates from a normal range based on the characteristics of the normal driving data learned by the time-series learning model unit; and an output unit that provides the result of the failure prediction unit to a user. and may include a control unit connected to each of the above data collection unit, data preprocessing unit, time series learning model unit, anomaly detection model unit, fault prediction unit, and output unit to control the operation of the above data collection unit, data preprocessing unit, time series learning model unit, anomaly detection model unit, fault prediction unit, and output unit.

[0022] In addition, in one embodiment, the data collection unit may collect data of the electronic control unit using at least one of CAN, CAN-FD, LIN, FlexRay, or DoIP communication methods.

[0024] In addition, in one embodiment, the time series learning model unit may include at least one of LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), or Temporal CNN-based neurons.

[0026] In addition, in one embodiment, the anomaly detection model unit may calculate a reconstruction error or anomaly score using at least one of an AutoEncoder, a Variational AutoEncoder, or a One-Class SVM-based model.

[0028] In addition, in one embodiment, the failure prediction unit can predict the remaining useful life (RUL) along with the probability of failure occurring.

[0030] In addition, in one embodiment, the output unit can visually output a warning level, maintenance recommendation information, or fault type through at least one of a diagnostic screen, a web-based dashboard, or a mobile terminal.

[0032] In addition, in one embodiment, the time series learning model unit and the anomaly detection model unit may be executed in a server or cloud environment outside the vehicle, and the data collection unit and the data preprocessing unit may have a distributed processing structure deployed in the vehicle or diagnostic device.

[0034] In addition, in one embodiment, the data of the electronic control unit may be encrypted when transmitted from the vehicle or diagnostic device to the server.

[0036] As another embodiment according to the present invention, a vehicle failure prediction method for predicting a vehicle failure using time series data collected from an electronic control unit (ECU) mounted on a vehicle comprises: a data collection step for collecting sensor data, driving data, or diagnostic data from the electronic control unit during vehicle operation; a data preprocessing step for converting the collected data into time series data by sorting the data based on time and performing at least one of noise removal, missing value correction, or normalization after the data collection step; a time series feature learning step for learning the temporal pattern of the time series data using a time series learning model after the data preprocessing step; an anomaly detection step for calculating an anomaly score of the time series data by comparing it with normal driving data after the time series feature learning step; a failure prediction step for predicting at least one of the probability of a vehicle failure, the risk of failure, or the remaining lifespan by combining the anomaly score and the time series feature after the anomaly detection step; and a result output step for providing a failure prediction result to a user after the failure prediction step.

[0038] In addition, in one embodiment, the time series feature learning step may be performed using at least one of an LSTM or a GRU-based neural network.

[0040] In addition, in one embodiment, the anomaly detection step can calculate the reconstruction error relative to normal data using an AutoEncoder-based model.

[0042] In addition, in one embodiment, the failure prediction step can calculate the remaining lifespan as a quantitative value along with the probability of failure occurrence.

[0044] In addition, in one embodiment, the result output step may provide at least one of different warning grades or maintenance recommendation messages depending on the failure risk. Effects of the invention

[0046] According to one embodiment of the present invention, early detection of failure precursors based on data accumulated during vehicle operation can effectively reduce safety accidents and economic losses caused by unexpected vehicle breakdowns or stoppages during operation, and improve maintenance efficiency by reasonably predicting the timing of maintenance.

[0047] In addition, according to one embodiment of the present invention, it can be flexibly applied to various vehicles and operating environments, thereby increasing the reliability of vehicle management and reducing unnecessary inspections or excessive maintenance, and providing the effect of contributing to improved vehicle uptime and reduced operating costs in the long term.

[0049] However, the effects obtainable from the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art from the description below. Brief explanation of the drawing

[0051] The following drawings attached to this specification illustrate preferred embodiments of the present invention and serve to further enhance understanding of the technical concept of the present invention together with the detailed description of the invention provided below; therefore, the present invention should not be interpreted as being limited only to the matters described in such drawings. FIG. 1 is a conceptual diagram of a vehicle failure prediction system according to an embodiment of the present invention. Figure 2 is a block diagram of the components of a vehicle failure prediction system. Figure 3 is a flowchart of data collection and preprocessing of an electronic control unit. Figure 4 is a structural diagram of the time series learning model unit and the anomaly detection model unit. Figure 5 is a diagram illustrating an example of the result output of the fault prediction unit. FIG. 6 is a flowchart illustrating a vehicle failure prediction method according to an embodiment of the present invention. Specific details for implementing the invention

[0052] Below, with reference to the attached drawings, embodiments of the present invention are described in detail so that those skilled in the art can easily implement the invention. However, since the description of the present invention is merely an example for structural or functional explanation, the scope of the present invention should not be interpreted as being limited by the embodiments described in the text. That is, since the embodiments 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 embodiment 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.

[0053] The meaning of the terms described in this invention should be understood as follows.

[0054] Terms such as "first" and "second" 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. When a component is referred to as being "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 a component is referred to as being "directly connected" to another component, it should be understood that there are no other components in between. Meanwhile, other expressions describing the relationship between components, such as "between" and "exactly between," or "adjacent to" and "directly adjacent to," shall be interpreted in the same manner.

[0055] 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 set-up 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.

[0056] 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 invention.

[0058] FIG. 1 is a conceptual diagram of a vehicle fault prediction system according to an embodiment of the present invention, FIG. 2 is a block diagram of the components of a vehicle fault prediction system, FIG. 3 is a flowchart of data collection and preprocessing of an electronic control unit, FIG. 4 is a structural diagram of a time series learning model unit and an anomaly detection model unit, and FIG. 5 is a diagram illustrating an example of the result output of a fault prediction unit.

[0059] As illustrated in FIGS. 1 to 5, in a vehicle failure prediction system that predicts vehicle failures based on time series data collected from an electronic control unit (ECU) mounted on a vehicle, the present invention may include a data collection unit (100), a data preprocessing unit (200), a time series learning model unit (300), an anomaly detection model unit (400), a failure prediction unit (500), an output unit (600), and a control unit (700).

[0060] Referring to FIG. 1, sensor data, driving data, and diagnostic data generated from the electronic control unit are transmitted to a data collection unit (100), undergo preprocessing such as time alignment, noise removal, missing value correction, and normalization, and then encrypted and transmitted to a server outside the vehicle. The server (50) calculates the probability of failure, remaining lifespan, or risk level through time series learning and anomaly detection, and is configured to utilize the results for vehicle maintenance by providing them in the form of visual displays, warning notifications, and maintenance guides through a diagnostic device, web, or mobile terminal.

[0061] The data collection unit (100) can collect sensor data, driving data, or diagnostic data from an electronic control unit (ECU). Specifically, the data collection unit (100) is configured to collect various data generated during vehicle operation in real-time or periodically from a plurality of electronic control units (ECUs) mounted on a vehicle (30), and is configured to acquire at least one of sensor data, driving status data, and diagnostic-related data. At this time, the data collection unit (100) receives signals transmitted through an internal vehicle communication network and may include buffering, timestamp assignment, or data packet separation functions to stably acquire data even if the period, format, and resolution of data generated from different electronic control units differ. Additionally, it may further include retransmission request, temporary storage, or error flag assignment functions to maintain a continuous data flow even in situations where data loss or communication delay occurs.

[0062] The data collection unit (100) can collect data from an electronic control unit using at least one of CAN, CAN-FD, LIN, FlexRay, or DoIP communication methods. Specifically, the data collection unit (100) is configured to use at least one of CAN, CAN-FD, LIN, FlexRay, or DoIP communication methods in order to reliably collect various data generated during operation from an electronic control unit mounted on a vehicle (30). Here, CAN (Controller Area Network) is a representative automotive network that transmits and receives data based on messages between electronic control units inside a vehicle, and can be utilized to collect data in major control areas such as engine and braking based on high reliability and real-time capabilities.

[0063] CAN-FD (Controller Area Network with Flexible Data-rate) is an extension of the existing CAN that allows for flexible adjustment of data length per frame and transmission speed, enabling more efficient collection of high-resolution sensor data or large-capacity time-series data. LIN (Local Interconnect Network) is an auxiliary communication network with a relatively simple structure and low communication speed, making it suitable for cost-effectively collecting data generated from auxiliary control devices or convenience devices.

[0064] FlexRay is a high-speed, high-reliability communication method based on time synchronization, which enables deterministic data transmission in control areas where safety is critical, and can be utilized for collecting precise status information. Additionally, DoIP (Diagnostics over Internet Protocol) is a communication method that transmits diagnostic data based on the Internet Protocol (IP), enabling the high-speed collection of large-capacity diagnostic information or remote diagnostic data through an Ethernet environment. In this way, the data collection unit (100) can be configured to enable reliable time-series data collection even under various vehicle types and operating conditions by selectively or in combination applying different communication methods depending on the communication environment of the vehicle (30) and the characteristics of the electronic control unit.

[0066] Referring to Fig. 3, data generated during operation is first collected from the electronic control unit, and the collected data is sorted according to time standards so that data from different occurrence points can be compared on the same time axis. Subsequently, data reliability is ensured by removing noise caused by sensor errors or disturbances and performing corrections for missing values, and the range of each data point is adjusted to a constant level through normalization and scaling processes. Through this series of processing steps, time-series data is finally generated and configured to be utilized in the learning and analysis stages for fault prediction.

[0067] The data preprocessing unit (200) can convert the data collected from the data collection unit (100) into time-series data by sorting the data based on time and performing at least one of noise removal, missing value correction, or normalization. Specifically, the data preprocessing unit (200) is configured to convert raw data transmitted from the data collection unit (100) into time-series data in an analyzable form, and performs at least one of sorting based on time, noise removal, missing value correction, or normalization.

[0068] At this time, the data preprocessing unit (200) can improve the reliability of the data by aligning data with different sampling periods along a common time axis or by removing noise caused by abnormal spike values ​​or sensor errors. In addition, if some data is missing, an interpolation or statistical estimation method using values ​​from adjacent time points can be applied, and the range of each data can be converted to a constant scale to prevent a specific signal from having an excessive influence during the subsequent learning process.

[0070] The time series learning model unit (300) may be configured to learn the temporal correlation of time series data. Specifically, the time series learning model unit (300) is configured to learn the temporal change patterns and correlations included in the preprocessed time series data and operates to model the data flow that appears in the normal operating state of the vehicle (30). By performing learning by considering the trend of change over a continuous time interval rather than a value at a single point in time, the time series learning model unit (300) can effectively reflect performance degradation that progresses gradually or abnormal patterns that appear repeatedly. In addition, it may have a structure that can be repeatedly learned or updated using data accumulated over a certain period so that the normal state range can be flexibly learned even if the operating conditions or driving environment of the vehicle (30) change.

[0071] The time series learning model unit (300) may include at least one of a Long Short-Term Memory (LSTM), a Gated Recurrent Unit (GRU), or a Temporal CNN-based neural network. Specifically, the time series learning model unit (300) is configured to include a neural network having a recurrent structure or a time-axis convolutional structure to learn the characteristics of vehicle operation data that change over time, and at least one of a Long Short-Term Memory (LSTM), a Gated Recurrent Unit (GRU), or a Temporal Convolutional Neural Network (Temporal CNN) may be used as such a neural network.

[0072] LSTM has a structure capable of effectively learning long-term temporal dependencies by selectively controlling the storage and deletion of past information through input, forget, and output gates, and is suitable for modeling long-term accumulated change patterns such as the gradual performance degradation of a vehicle. GRU has a recurrent structure similar to LSTM, but improves computational efficiency by learning temporal dependencies with a simpler structure using update and reset gates, and enables stable learning even in environments with limited computational resources.

[0073] Additionally, the Temporal CNN has a structure advantageous for extracting local patterns that appear repeatedly within a certain interval by performing convolution operations along the time axis, and is configured to effectively capture abnormal signals that occur repeatedly under specific conditions while driving. This time series learning model unit (300) can be applied alone or in combination depending on the driving conditions or data characteristics of the vehicle (30), thereby improving the accuracy and reliability of fault prediction in various vehicle (30) environments.

[0075] The anomaly detection model unit (400) can calculate the degree of anomaly in time-series data based on normal driving data. Specifically, the anomaly detection model unit (400) is configured to determine whether input data deviates from the normal range based on the time-series data pattern in a normal driving state, and calculates the degree of anomaly by quantifying the difference or distortion between the data. At this time, the anomaly detection model unit (400) is configured to distinguish between normal fluctuations due to changes in driving conditions and actual failure precursor phenomena by making a judgment based on the results of learning the complex characteristics of the normal state, rather than a simple threshold comparison. In addition, the calculated anomaly score can be processed in a cumulative or weighted manner to reflect not only sudden anomalies at a specific point in time but also minute changes accumulated over a long period.

[0076] The anomaly detection model unit (400) can calculate a reconstruction error or an anomaly score using at least one of an AutoEncoder, a Variational AutoEncoder, or a One-Class SVM-based model. Specifically, the anomaly detection model unit (400) is configured to calculate a reconstruction error or an anomaly score using at least one of an AutoEncoder, a Variational AutoEncoder, or a One-Class SVM-based model, as a configuration for determining whether the vehicle's operating state deviates from a normal range based on data collected during normal driving.

[0077] An AutoEncoder is a neural network structure that learns the characteristics of normal data by compressing input data into a low-dimensional latent space and then restoring it to its original form. By calculating the difference between the input data and the restored data as a reconstruction error, it can effectively detect abnormal states that deviate from normal patterns. A Variational AutoEncoder is an extension of the AutoEncoder that has a structure for learning latent variables based on probabilistic distributions. By modeling the distribution range of normal data and calculating the degree of deviation from that distribution as an anomaly score, it enables more stable anomaly detection even in driving environments with high uncertainty.

[0078] In addition, One-Class Support Vector Machine (One-Class SVM) is a machine learning technique that forms a boundary region within a feature space using only normal data and determines data outside that boundary as an anomaly, thereby enabling effective anomaly detection even in situations where sufficient failure data is not secured. This anomaly detection model unit (400) can calculate an anomaly score by applying a single model or combining the results of multiple models, and can be configured to more reliably determine vehicle failure precursors by considering the trend of change in the anomaly score over time.

[0080] The time series learning model unit (300) and the anomaly detection model unit (400) may have a distributed processing structure in which they are executed in a server or cloud environment outside the vehicle (30), and the data collection unit (100) and the data preprocessing unit (200) are deployed in the vehicle (30) or the diagnostic device. Specifically, the time series learning model unit (300) and the anomaly detection model unit (400) are configured to be executed in a server (50) or cloud environment located outside the vehicle (30), and the data collection unit (100) and the data preprocessing unit (200) have a distributed processing structure in which they are deployed and operated inside the vehicle (30) or in a diagnostic device connected to the vehicle.

[0081] According to this distributed processing structure, the vehicle (30) or diagnostic device can reduce unnecessary data transmission volume and minimize the impact of communication delays or limitations of internal vehicle computing resources by collecting data generated during operation in real time and performing basic preprocessing. On the other hand, learning and anomaly detection functions with high computational load are performed on an external server (50) or in a cloud environment, enabling precise analysis using high-performance computing resources.

[0082] In addition, in a server (50) or cloud environment, data collected from multiple vehicles (30) can be managed comprehensively or models can be periodically updated, thereby simultaneously improving prediction accuracy by reflecting differences in the individual environments of the vehicles (30) and ensuring the scalability of the system (10). This configuration reduces the burden on the internal vehicle system while enabling the provision of stable fault prediction services even in remote management and large-scale vehicle operation environments.

[0084] Data from the electronic control unit can be transmitted in an encrypted manner when transferred from the vehicle (30) or diagnostic device to the server (50). Specifically, the data from the electronic control unit is configured to be transmitted in an encrypted manner when transferred from the vehicle or diagnostic device to a server (50) or cloud environment outside the vehicle (30). In this case, encryption is applied as a means to prevent unauthorized access or tampering by a third party during the data transmission process, and ensures that sensitive data, such as vehicle identification information, driving history, or diagnostic information, is safely protected. Additionally, the encrypted data can be configured so that decryption and analysis are possible only after undergoing a normal authentication procedure on the server (50) side, thereby ensuring both the confidentiality and integrity of the data. This encrypted transmission structure allows the security level of vehicle data to be maintained even in remote diagnostic and cloud-based analysis environments, thereby enabling the provision of reliable fault prediction services under various vehicle environments and communication conditions.

[0086] The fault prediction unit (500) can calculate at least one of the probability of failure, the risk of failure, or the remaining lifespan through an anomaly score from an anomaly detection model unit (400) that determines whether the input time series data deviates from the normal range based on the characteristics of the normal driving data learned by the time series learning model unit (300). Specifically, the fault prediction unit (500) is configured to comprehensively determine the failure status of the vehicle by combining the learning result of the time series learning model unit (300) and the anomaly score calculated by the anomaly detection model unit (400), and is configured to calculate at least one of the probability of failure, the risk of failure, or the remaining lifespan. At this time, the fault prediction unit (500) can derive a more reliable prediction result by considering both the pattern change over time and the anomaly accumulation trend together, rather than relying on a single indicator. In addition, the fault prediction result can be calculated to reflect not only the current state of the vehicle (30) but also the level of risk that may occur within a certain period in the future, so it can be configured to be utilized for preventive maintenance judgment.

[0088] The fault prediction unit (500) can predict the remaining useful life (RUL) along with the probability of a failure occurring. Specifically, the fault prediction unit (500) is configured to quantitatively calculate the probability of a failure occurring and to predict the remaining useful life (RUL) by comprehensively analyzing the current vehicle state and the trend of change over time based on time-series data accumulated during the operation of the vehicle (30). At this time, the fault prediction unit (500) can calculate the probability of a failure occurring within a certain period in the future by considering the accumulation trend of the abnormal score, the rate of change, or the pattern of repeated occurrence, rather than just determining whether there is an abnormality at a single point in time, and thereby the failure risk can be classified in stages.

[0089] In addition, the remaining lifespan can be defined as the period or amount of operation during which a vehicle or a specific component is expected to maintain normal function, and can be calculated by comparing past operation history with the current state and gradually updating it. This failure prediction unit (500) can be practically utilized for planning maintenance timing, managing parts replacement cycles, and improving vehicle operation efficiency by providing the prediction results as numerical and temporal information rather than simple warning signals.

[0091] Referring to FIG. 5, the vehicle (30) identification information, along with the risk of failure, expected failure parts, remaining lifespan, and maintenance recommendations, are visually displayed on a diagnostic device, web, or mobile terminal through the output unit (600). If the risk of failure is high, the user can immediately recognize it through a warning display or notification. This output screen intuitively conveys the status of the vehicle (30) to support the driver or manager in performing inspection and maintenance at an appropriate time.

[0092] The output unit (600) can provide the results of the fault prediction unit (500) to the user. Specifically, the output unit (600) is configured to intuitively provide the results calculated by the fault prediction unit (500) to the user, and visually outputs the prediction information through a diagnostic screen, a web-based interface, or a mobile terminal. At this time, the output unit (600) may display different warning grades according to the fault risk level, or provide a message guiding the recommended maintenance time and target parts. In addition, to enhance user understanding, it may include visual elements such as graphs, icons, or color displays, and may be configured to provide comparison information with past driving history or the trend of changes in the prediction results as needed.

[0094] The output unit (600) can visually output a warning level, maintenance recommendation information, or fault type through at least one of a diagnostic screen, a web-based dashboard, or a mobile terminal. Specifically, the output unit (600) is configured to provide a user with an intuitive understanding of the prediction results transmitted from the fault prediction unit (500), and is configured to visually output information through at least one of a diagnostic screen, a web-based dashboard, or a mobile terminal.

[0095] At this time, the output unit (600) may display warning grades that are distinguished according to the probability of failure or the level of failure risk, or provide maintenance recommendation information regarding the timing and targets for recommended inspection or replacement. Additionally, the output unit (600) may be configured to include not only text information but also visual elements such as graphs, icons, color displays, or trends of change over time, so that the user can grasp the current status and future risk level of the vehicle (30) at a glance. Furthermore, the output unit (600) may be configured to more efficiently support vehicle management and maintenance decision-making by adjusting the range of information displayed according to user authority or usage environment, or by providing comparison information with past prediction results.

[0097] The control unit (700) is connected to the data collection unit (100), data preprocessing unit (200), time series learning model unit (300), anomaly detection model unit (400), fault prediction unit (500), and output unit (600), respectively, and can control the operation of the data collection unit (100), data preprocessing unit (200), time series learning model unit (300), anomaly detection model unit (400), fault prediction unit (500), and output unit (600). Specifically, the control unit (700) is configured to control the overall operation of the data collection unit (100), data preprocessing unit (200), time series learning model unit (300), anomaly detection model unit (400), fault prediction unit (500), and output unit (600), and manages the data flow and processing order between each component. The control unit (700) can dynamically adjust the data collection cycle, analysis frequency, or output method according to the vehicle (30) operating state or the system (10) settings, and can also control the system to perform prediction calculations only when specific conditions are met. In addition, it can be configured to improve the stability and reliability of the entire system (10) by including a function to detect an error state when a system (10) malfunction occurs and to restrict the operation of the corresponding component or switch to a safety mode.

[0098] The control unit (700) is equipped with a wireless communication module and is linked with an external mobile terminal, etc., so that operation can be controlled from the outside or the control status can be checked in real time. The wireless communication module can be implemented with various communication technologies.

[0100] FIG. 6 is a flowchart illustrating a vehicle failure prediction method according to an embodiment of the present invention.

[0101] As illustrated in FIG. 6, in a vehicle failure prediction method that predicts vehicle failures using time series data collected from an electronic control unit (ECU) mounted on a vehicle, the present invention may comprise a data collection step (S100), a data preprocessing step (S200), a time series feature learning step (S300), an anomaly detection step (S400), a failure prediction step (S500), and a result output step (S600).

[0102] The data collection step (S100) is a step of collecting sensor data, driving data, or diagnostic data from an electronic control unit while the vehicle is in operation. Specifically, the data collection step (S100) is a step of collecting sensor data, driving status data, or diagnostic data from an electronic control unit mounted on the vehicle (30) while the vehicle (30) is in operation, and is a step of securing information that reflects the actual driving environment and state of the vehicle (30). In this step (S100), since data generated from different electronic control units may be generated at different cycles and in different formats, time information is assigned to each collected data so that the chronological order can be clearly identified in subsequent steps. In addition, temporary storage or re-collection processing may be performed together to maintain the continuity of the data flow even in the event of communication delays or temporary data omissions.

[0103] The data preprocessing step (S200) is a step of converting the data collected after the data collection step (S100) into time-series data by sorting the collected data by time and performing at least one of noise removal, missing value correction, or normalization. Specifically, the data preprocessing step (S200) is a step for converting the collected raw data into time-series data in a form suitable for analysis, and performs at least one of sorting by time, noise removal, missing value correction, or normalization.

[0104] In this step (S200), data with different sampling periods are aligned along a common time axis to enable comparison, and abnormal values ​​caused by sensor errors or external environmental factors are removed or mitigated. Additionally, if some data is missing, data continuity is ensured by correcting it using values ​​from adjacent time points or statistical estimation, and the data range is adjusted to a constant scale to prevent specific signals from having an excessive influence in the subsequent learning phase.

[0105] The time series feature learning step (S300) is a step of learning the temporal patterns of time series data using a time series learning model after undergoing the data preprocessing step (S200). Specifically, the time series feature learning step (S300) is a step of learning the patterns and correlations of vehicle (30) operation data that change over time by using the preprocessed time series data as input. In this step (S300), learning is performed by considering the trend of change over a continuous time interval rather than the value at a single point in time, and through this, characteristics that appear repeatedly in a normal operating state or tendencies that change gradually can be extracted as features. In addition, the learning model can be updated or supplemented using data accumulated over a certain period so that the range of the normal state can be flexibly reflected even if the driving conditions or usage environment of the vehicle (30) change.

[0106] The time series feature learning step (S300) can be performed using at least one of an LSTM or GRU-based neural network. Specifically, the time series feature learning step (S300) is performed using at least one of an LSTM (Long Short-Term Memory) or GRU (Gated Recurrent Unit)-based neural network to effectively learn the characteristics of vehicle (30) operation data that change over time. In this step (S300), the LSTM selectively controls the storage and removal of past information through input gates, forget gates, and output gates, thereby enabling stable learning of operation characteristics accumulated over a long period or patterns of gradual performance degradation.

[0107] GRU enables fast and stable learning even in environments with limited computational resources by efficiently reflecting temporal dependencies through a relatively simple structure using update gates and reset gates. Additionally, the time series feature learning step (S300) is configured so that the learning results can be gradually supplemented or updated according to changes in the driving conditions or driving environment of the vehicle (30), thereby enabling the continuous extraction of features effective for distinguishing between normal and abnormal patterns even in various vehicle (30) states.

[0109] The anomaly detection step (S400) is a step that calculates an anomaly score of the time series data by comparing it with normal driving data after undergoing the time series feature learning step (S300). Specifically, the anomaly detection step (S400) is a step that determines whether the input time series data deviates from the normal range based on the characteristics of the normal driving data learned through the time series feature learning step (S300). In this step (S400), the difference between the input data and the normal data is quantified to calculate an anomaly score, and the presence of an anomaly is determined through criteria that reflect the complex characteristics of the normal state, rather than a simple threshold comparison. Additionally, to distinguish between changes caused by temporary disturbances and actual failure precursors, the temporal change trend or cumulative trend of the anomaly score may be considered together.

[0110] The anomaly detection step (S400) can calculate the reconstruction error relative to normal data using an AutoEncoder-based model. Specifically, the anomaly detection step (S400) is performed using an AutoEncoder-based model to determine whether the vehicle's operating state deviates from the normal range based on data collected during normal driving. Here, the AutoEncoder is a neural network structure trained to compress input data into a low-dimensional latent space and then restore it to its original form, and is suitable for effectively learning the common characteristics and patterns of normal data. In this step (S400), when normal data is input, the difference between the original data and the restored data appears relatively small, whereas when data deviating from the normal state is input, the difference becomes large. This difference is defined as a reconstruction error to quantify the degree of anomaly.

[0111] This reconstruction error can be calculated by considering not only the value at a single point in time but also the trend of change or cumulative trend over time, and can be utilized to distinguish between temporary disturbances and actual failure precursors. Through this, the anomaly detection step (S400) is configured to enable reliable anomaly determination based on the normal state, even in environments where sufficient failure data is not available.

[0113] The failure prediction step (S500) is a step that predicts at least one of the probability of failure, the risk of failure, or the remaining lifespan of the vehicle (30) by combining the anomaly score and time series features after passing through the anomaly detection step (S400). Specifically, the failure prediction step (S500) is a step that comprehensively predicts the failure state of the vehicle (30) by combining the anomaly score calculated in the anomaly detection step (S400) and the temporal pattern information derived in the time series feature learning step (S300). In this step (S500), the probability of failure or the risk of failure can be calculated by considering not only the degree of anomaly at the current point in time but also the trend of change from the past to the present. Additionally, the remaining lifespan is defined as the period during which the vehicle (30) or a specific component is expected to maintain normal function, and by predicting this as a quantitative value, it can be utilized to determine the timing of future maintenance.

[0115] The failure prediction step (S500) can calculate the remaining lifespan as a quantitative value along with the probability of failure occurrence. Specifically, the failure prediction step (S500) is performed by comprehensively analyzing time-series data accumulated during the operation of the vehicle (30) and anomaly detection results to calculate the possibility of a failure occurring within a certain period in the future in the form of a failure probability, and at the same time, to quantify and calculate the period during which the vehicle (30) or a specific component is expected to maintain normal function as the remaining useful life (RUL).

[0116] At this time, the probability of failure can be calculated stepwise by reflecting the magnitude, rate of change, or cumulative trend of the abnormal score, and the remaining lifespan can be expressed as a numerical value in units of time or driving volume based on the result of comparing past driving history and current state. In addition, these prediction results can be periodically updated according to changes in the driving conditions or usage environment of the vehicle (30), so as to provide prediction information that is continuously supplemented rather than limited to a judgment at a single point in time. Through this, the failure prediction step (S500) provides a basis for quantitative judgment that can be practically utilized for planning maintenance timing, managing parts replacement cycles, and improving the operational efficiency of the vehicle (30).

[0118] The result output step (S600) is a step that provides the fault prediction result to the user after passing through the fault prediction step (S500). Specifically, the result output step (S600) is a step that provides the prediction result derived from the fault prediction step (S500) to the user, and visualizes the information so that it can be directly utilized for vehicle (30) management or maintenance decision-making. In this step (S600), different warning grades may be provided according to the fault risk level, or guidance information related to the recommended time for inspection or maintenance may be provided. In addition, the trend of change in the prediction result may be displayed as a graph or visual display so that the user can intuitively understand the condition of the vehicle (30), and comparison information with past results may be provided as needed.

[0119] The result output step (S600) may provide at least one of different warning grades or maintenance recommendation messages depending on the failure risk. Specifically, the result output step (S600) is a step that provides information to enable the user to intuitively recognize the status of the vehicle (30) based on the failure probability, failure risk, or remaining lifespan calculated in the failure prediction step (S500). In this step (S600), multiple warning grades that are distinguished from each other according to the level of failure risk may be set, and corresponding to each warning grade, at least one of maintenance recommendation messages regarding whether an inspection is necessary, the timing of maintenance, or driving restrictions may be provided. In addition, the result output step (S600) may be configured to clearly convey differences in risk levels by utilizing visual elements such as colors, icons, or stage indicators in addition to simple text display. Furthermore, information on the trend of changes in the prediction results or comparison information with previous output results may be provided together, thereby supporting the user in continuously monitoring the trend of the vehicle (30) status and making more rational maintenance decisions.

[0121] In addition, the present invention is extended to be applicable to various vehicles such as passenger cars, commercial vehicles, agricultural machinery, construction machinery, and special vehicles, and is configured to predict failures based on electronic control unit data collected according to the type and operating environment of the vehicle (30). Specifically, in the case of agricultural machinery, failures are predicted by collecting driving and work status data from the engine control unit, transmission control unit, and hydraulic control unit mounted on the tractor, combine, or rice transplanter, and in the case of construction machinery and heavy equipment, abnormalities are determined using data including work load, hydraulic pressure, and work cycle information generated from the engine control unit, work device control unit, and hydraulic control unit mounted on the excavator, loader, crane, or forklift.

[0122] Furthermore, for electric or hybrid vehicles, fault prediction is performed based on voltage, current, and temperature data collected from the battery management system, inverter control unit, and motor control unit. Moreover, in areas related to exhaust emissions and environmental regulations, fault prediction is achieved by considering abnormal signs in exhaust gas aftertreatment systems, including EGR, DPF, and SCR, thereby enabling reliable fault prediction across various vehicle types and regulatory environments.

[0124] In addition, the present invention is configured to enable predictive analysis and integrated management of multiple vehicles (30) by transmitting vehicle failure prediction results to a remote server (50) or a cloud system, wherein data collected inside the vehicle (30) is transmitted to an external server after basic preprocessing, and the AI-based failure prediction analysis is performed on the server side, having a distributed processing structure.

[0125] At this time, encryption is applied during the data transmission process between the vehicle (30) or the diagnostic device and the server (50) to ensure the confidentiality and integrity of the electronic control unit data, and at least one of public key-based encryption, symmetric key encryption, or seed-key-based authentication methods may be used. In addition, the time-series learning model and the anomaly detection model are not stored in the vehicle (30) or the diagnostic device but are stored and executed on the server (50) to reduce the risk of model leakage, and are updated via a remote update method and managed separately from the firmware of the vehicle (30) or the diagnostic device, thereby enabling compliance with cyber security regulations. Furthermore, the viewing or control of the fault prediction results may be restricted according to user authority, vehicle (30) identification information, or manufacturer policy, so that security and management efficiency can be simultaneously ensured in various vehicle (30) operating environments.

[0127] As described above, the detailed description of the preferred embodiments of the present invention disclosed is provided to enable those skilled in the art to implement and practice the present invention. Although the present invention has been described with reference to preferred embodiments, those skilled in the art will understand that various modifications and changes can be made to the present invention without departing from the scope of the invention. For example, those skilled in the art may utilize each configuration described in the embodiments described above in combination with one another. Accordingly, the present invention is not intended to be limited to the embodiments shown herein, but to be given the broadest scope consistent with the principles and novel features disclosed herein.

[0128] The present invention may be embodied in other specific forms without departing from the spirit and essential features of the invention. Accordingly, the above detailed description should not be interpreted restrictively in all respects but should be considered exemplary. The scope of the invention shall be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the invention are included within the scope of the invention. The invention is not intended to be limited to the embodiments shown herein, but to be given the broadest possible scope consistent with the principles and novel features disclosed herein. Furthermore, embodiments may be constructed by combining claims that are not explicitly related in the claims, or by including them as new claims through amendments made after filing. Explanation of the symbols

[0129] 10: Vehicle Failure Prediction System 30 : Vehicle 50 : Server 100 : Data Collection Unit 200 : Data Preprocessing Section 300 : Time series learning model section 400 : Anomaly Detection Model Section 500 : Fault Prediction Unit 600 : Output section 700 : Control unit

Claims

Claim 1 A vehicle failure prediction system for predicting vehicle failures based on time-series data collected from an electronic control unit (ECU) mounted on a vehicle, comprising: a data collection unit for collecting sensor data, driving data, or diagnostic data from the electronic control unit; a data preprocessing unit for sorting the data collected from the data collection unit by time and converting it into time-series data by performing at least one of noise removal, missing value correction, or normalization; a time-series learning model unit configured to learn the temporal correlation of the time-series data; an anomaly detection model unit for calculating the degree of anomaly in the time-series data based on normal driving data; a failure prediction unit for calculating at least one of the probability of failure, the risk of failure, or the remaining lifespan through an anomaly score of the anomaly detection model unit, which determines whether the input time-series data deviates from a normal range based on the characteristics of the normal driving data learned by the time-series learning model unit; and an output unit for providing the result of the failure prediction unit to a user. and a control unit connected to each of the data collection unit, data preprocessing unit, time series learning model unit, anomaly detection model unit, fault prediction unit, and output unit to control the operation of the data collection unit, data preprocessing unit, time series learning model unit, anomaly detection model unit, fault prediction unit, and output unit; wherein the fault prediction unit predicts the remaining useful life (RUL) along with the probability of failure; the data collection unit receives signals transmitted through an internal vehicle communication network and includes at least one of buffering, timestamp assignment, or data packet separation functions to stably acquire data even if the period, format, and resolution of data generated from different electronic control units differ; and includes at least one of retransmission request, temporary storage, or error flag assignment functions to maintain a continuous data flow even in situations where data loss or communication delay occurs.The data preprocessing unit can align data with different sampling periods along a common time axis, remove noise caused by abnormal spike values ​​or sensor errors, and apply interpolation or statistical estimation methods using values ​​from adjacent time points in cases where some data is missing. It also prevents specific signals from having an excessive influence during the subsequent learning process by converting the range of each data to a constant scale. The time series learning model unit is configured to repeatedly learn or update the temporal change patterns of the normal state using time series data accumulated over a certain period during normal operation, and is configured to flexibly learn the range of the normal state to distinguish between normal fluctuations and actual failure precursors resulting from changes in vehicle operating conditions. The anomaly detection model unit calculates an anomaly score based on the complex characteristics of the normal state, determining the degree to which the input time series data deviates from the normal range. It processes the anomaly score using a cumulative or weighting method to reflect not only sudden anomalies at a specific point in time but also minute changes accumulated over a long period. The failure prediction unit considers both pattern changes over time and the anomaly accumulation trend. The failure prediction result reflects not only the vehicle's current state but also the risk level that may occur within a certain period in the future, and considers the accumulation trend, rate of change, or recurrence of the anomaly score. A vehicle failure prediction system characterized by calculating the probability of a failure occurring within a future period by considering at least one of the patterns, wherein the remaining lifespan is calculated by gradually updating it by comparing past driving history with the current state, and providing the prediction results as numerical and temporal information. Claim 2 A vehicle fault prediction system according to claim 1, wherein the data collection unit collects data of an electronic control unit using at least one of CAN, CAN-FD, LIN, FlexRay, or DoIP communication methods. Claim 3 A vehicle failure prediction system according to claim 1, wherein the time series learning model unit comprises at least one of a Long Short-Term Memory (LSTM), a Gated Recurrent Unit (GRU), or a Temporal CNN-based neural network. Claim 4 A vehicle failure prediction system according to claim 1, wherein the anomaly detection model unit calculates a reconstruction error or anomaly score using at least one of an AutoEncoder, a Variational AutoEncoder, or a One-Class SVM-based model. Claim 5 delete Claim 6 A vehicle failure prediction system according to claim 1, wherein the output unit visually outputs a warning level, maintenance recommendation information, or failure type through at least one of a diagnostic screen, a web-based dashboard, or a mobile terminal. Claim 7 A vehicle fault prediction system according to claim 1, characterized in that the time series learning model unit and the anomaly detection model unit are executed in a server or cloud environment outside the vehicle, and the data collection unit and the data preprocessing unit have a distributed processing structure deployed in the vehicle or diagnostic device. Claim 8 A vehicle fault prediction system according to claim 7, characterized in that the data of the electronic control unit is encrypted when transmitted from the vehicle or diagnostic device to the server. Claim 9 A vehicle failure prediction method for predicting vehicle failures using time-series data collected from an electronic control unit (ECU) mounted on a vehicle, comprising: a data collection step for collecting sensor data, driving data, or diagnostic data from the electronic control unit during vehicle operation; a data preprocessing step for converting the collected data into time-series data by sorting the data based on time and performing at least one of noise removal, missing value correction, or normalization after the data collection step; a time-series feature learning step for learning the temporal pattern of the time-series data using a time-series learning model after the data preprocessing step; an anomaly detection step for calculating an anomaly score of the time-series data through comparison with normal driving data after the time-series feature learning step; and a failure prediction step for predicting at least one of the probability of vehicle failure, failure risk, or remaining lifespan by combining the anomaly score and the time-series features after the anomaly detection step. The method comprises a result output step that provides a fault prediction result to a user after passing through the fault prediction step; wherein the fault prediction step calculates the remaining lifespan as a quantitative value along with the probability of failure occurrence, and the data collection step receives a signal transmitted through an internal vehicle communication network and includes at least one of buffering, timestamp assignment, or data packet separation functions to stably acquire data even if the period, format, and resolution of data generated from different electronic control units differ, and includes at least one of retransmission request, temporary storage, or error flag assignment functions to maintain a continuous data flow even in situations where data loss or communication delay occurs, and the data preprocessing step may align data with different sampling periods along a common time axis, remove noise caused by abnormal spike values ​​or sensor errors, and apply interpolation using values ​​from adjacent time points or statistical estimation methods in cases where some data is missing.Vehicle failure prediction is characterized by preventing specific signals from having an excessive influence during the subsequent learning process by converting the range of each data to a constant scale, the time series feature learning step repeatedly learns or updates the temporal change pattern of the normal state using time series data accumulated over a certain period in a normal operating state, and flexibly learns the range of the normal state to distinguish between normal fluctuations and actual failure precursors due to changes in vehicle operating conditions, the anomaly detection step calculates an anomaly score based on the degree to which the input time series data deviates from the normal range based on the complex characteristics of the normal state, and processes the anomaly score in a cumulative or weighted manner to reflect not only sudden anomalies at a specific point in time but also minute changes accumulated over a long period, the failure prediction step considers both pattern changes over time and the anomaly accumulation trend together, the failure prediction result reflects the risk level that may occur within a certain period in the future as well as the current state of the vehicle, and calculates the probability of a failure occurring within a certain period in the future by considering at least one of the accumulation trend of the anomaly score, the rate of change, or the pattern of repeated occurrence, and the remaining life is calculated by gradually updating by comparing past operating history with the current state, and the prediction result is provided as numerical and temporal information. method., Claim 10 A vehicle failure prediction method according to claim 9, wherein the time series feature learning step is performed using at least one of an LSTM or a GRU-based neural network. Claim 11 A vehicle failure prediction method according to claim 9, wherein the anomaly detection step calculates a reconstruction error relative to normal data using an AutoEncoder-based model. Claim 12 delete Claim 13 A vehicle failure prediction method according to claim 9, wherein the result output step provides at least one of different warning grades or maintenance recommendation messages depending on the failure risk.

Images