Methods for reducing event memory data to assist in fault detection in one or more motor vehicles, as well as control device and vehicle system

Abstract

The invention relates to a method for reducing event memory data to support fault detection in one or more motor vehicles (10), wherein the event memory data (7) each represent a fault-related event. First, event memory data (7) is received from at least one control unit (2) of the one motor vehicle (10) or the multiple motor vehicles (10). Subsequently, at least one NLP (Natural Language Processing) model is applied to the event memory data (7), thereby reducing the event memory data (7) by removing redundant event memory data (7) and / or combining similar event memory data (7), wherein the redundant event memory data (7) comprise semantically at least partially similar fault-related events.Finally, the remaining event memory data (7) is transmitted to a workshop and / or an offboard diagnostic system to assist in fault detection.

Description

[0001] The invention relates to a method for reducing event memory data (event memory data sets) to support fault detection in one or more motor vehicles, as well as a control device and a vehicle system.

[0002] To detect faults in a motor vehicle, diagnostic systems can be implemented in one or more of the vehicle's control units, or diagnostics (on-board diagnostics) can be performed. These diagnostic systems automatically conduct cyclical checks in the vehicle for the respective control unit and store the resulting data as event data in internal memory. The event data will contain or describe one or at least one fault-related event, for example, in natural language. If, for example, a check does not produce an expected event or result, this event is supplemented with additional data, such as the time of occurrence, environmental data, and / or an event code, and stored as event data, for example, in non-volatile memory. A diagnostic tool can then be used to, for example,A workshop can read the event memory data, including the event(s), and / or determine the corresponding cause or fault. However, such events often do not occur in isolation, but in combination with other events from the same or other control units. Therefore, directly determining the cause of the fault or the vehicle's fault behavior when reading the event memory is not immediately possible.

[0003] Instead, a comprehensive analysis or investigation is required to identify the actual cause. This process can be very time-consuming and / or, in some cases, lead to unnecessary repairs to the vehicle, as well as high warranty and / or repair costs for the vehicle manufacturer and / or the user.

[0004] The increasing networking within a vehicle's infrastructure leads to a rise in the number of diagnostics and the resulting event memory entries. In particular, the networking of control units and the resulting (complex) relationships between occurring faults or events, which are then represented as event memory data, can complicate manual fault diagnosis. This can lead to longer workshop visits and / or extended vehicle downtime and / or higher costs. Since some events occur only as a consequence of other events and do not contribute to fault diagnosis, so-called readout authorizations have been introduced. These readout authorizations are lists containing codes assigned to an event, indicating to the workshop that certain event memory entries do not need to be considered for the diagnosis.Each read-through authorization must be manually entered into a system and is then distributed, for example, via a customer service center to a workshop information system of the workshop.

[0005] For example, a vehicle breaks down on the road. The initial readout of the event memory of all control units reveals, for instance, 138 entries. Overwrite authorization exists for 18 of these entries, leaving the workshop with 120 (relevant) event memory entries. The workshop must now determine the actual cause of the vehicle breakdown by performing a guided fault diagnosis. This process requires considerable effort, which can lead to extended vehicle downtime and / or high costs.

[0006] DE 10 2005 019 335 A1 discloses a method and a device for evaluating events from the operation of at least one vehicle.

[0007] DE 10 2018 209 108 A1 discloses a fast fault analysis for technical devices using machine learning.

[0008] US 2022 / 0207931A1 describes a method and system for predicting completion times for vehicle maintenance.

[0009] US 2024 / 0203168A1 describes a system and procedure for the automatic prediction and planning of vehicle repairs.

[0010] The aforementioned state of the art may leave room for further developments that evaluate events and / or event storage data more reliably and / or better, in particular filtering them.

[0011] The present invention is based on the objective of automatically providing or carrying out a fast and / or reliable prioritization of event memory data of a motor vehicle, in particular with regard to a vehicle problem.

[0012] The problem is solved by the subject matter of the independent patent claims. Advantageous further developments of the invention are described by the dependent patent claims, the following description, and the figure.

[0013] The invention relates to a method for (automatically) reducing or diminishing or filtering event memory data or event memory entries to support fault detection in one or more motor vehicles, wherein the event memory data each represent a fault-related event or, for example, describe or comprise (at least partially) it in natural language form, wherein the following steps are carried out (e.g., by means of a control device of a vehicle system of the motor vehicle): - Receiving event memory data from at least one control unit of one or more motor vehicles, e.g. by means of the control device or a receiving device of the control device or vehicle system, - Applying at least one (machine-based) NLP (Natural Language Processing) model to the event memory data, thereby reducing or filtering the event memory data by removing redundant event memory entries and / or grouping similar event memory data, where the redundant event memory entries semantically (similar in meaning or content) at least partially encompass (describe) similar error-related events. - Transmitting or routing the remaining or residual event memory data to a workshop and / or an off-board diagnostic system to assist in fault detection, e.g., via a communication unit or communication interface of the control device or vehicle system.

[0014] The control device can include and / or be connected to at least one NLP model.

[0015] In other words, event data records are data records that document or encompass the (respective) event, e.g., containing an error code and / or timestamp and / or sensor values ​​and / or a natural language description of the event. In particular, it may be stipulated that the error code includes the natural language description and / or must first be decoded. The natural language description could, for example, be "Tire pressure too low, front right." Specifically, the event may be or include a natural language error message.

[0016] The event can be, for example, a situation or incident detected in the vehicle. In particular, the event can include or describe (at least partially) a sensor-detected irregularity in natural language.

[0017] The term "similar" can mean that several event log entries are similar or identical in content and / or technical aspects. Similarity can be determined, for example, using a cosine similarity test, where event log entries are marked as "similar" if they exceed a predefined similarity threshold. In particular, this can mean that the event log entries are related in content or semantics, even though they each contain different error codes or events. Additionally or alternatively, it can be stipulated that event log entries are considered similar if they refer to the same error-related event.

[0018] The term "remaining" or "discarded" event memory data can refer to (all) those event memory data that were grouped together as similar and / or not removed.

[0019] This simplifies troubleshooting in the workshop and / or makes issuing diagnostics more efficient and / or straightforward. Furthermore, it enables and / or performs automated, fast, and / or reliable prioritization of the event memory data of the vehicle or at least one of its control units. In particular, it allows for the removal of redundant event memory data.

[0020] According to the invention, the at least one NLP model comprises the removal of event memory data whose semantic similarity is below a predetermined similarity threshold and / or the grouping of event memory data whose semantic similarity is equal to or above the predetermined similarity threshold as similar event memory data. The similarity threshold can, for example, be between 0.3 and 0.8 when applying cosine similarity and / or the Jaccard index. In particular, it can be provided that the event memory data is in numerical form and / or represented by sentence embeddings.

[0021] Alternatively, according to the invention, event memory data, in particular the event(s) that occur in a predetermined temporal dependency on other event memory data (the event(s)), are marked as consequential errors and removed. The predetermined temporal dependency is determined by extracting at least one absolute and one relative occurrence time of the event memory data. The temporal dependency can, for example, range from 0.1 seconds to 30 seconds. The temporal dependency can be predetermined in such a way that an occurring event or its corresponding event memory data within the predetermined temporal dependency is considered or marked and / or thereby removed as a consequential error of a preceding event or its event memory data.In other words, it's possible that event data (the event(s)) doesn't occur independently, but is caused by other event data (the event(s)). This allows event data to be identified and removed as a consequence of an error.

[0022] The invention also includes further developments that result in additional advantages.

[0023] Further training stipulates that at least one NLP model performs vectorization (e.g., using TF-IDF and / or Word to Vector, Word2Vec) of the event memory data. This allows the event memory data to be represented numerically and / or enables mathematical similarity analysis.

[0024] A training course stipulates that at least one NLP model performs cluster formation. This can involve either clustering or grouping, for example, using K-means.

[0025] This allows event log data to be clustered or grouped, particularly without the need for predefined rules. Redundant event log entries can thus be removed and / or similar event log data can be grouped together. In particular, this allows event log data to be divided thematically and / or semantically, e.g., into event log groups.

[0026] Further training stipulates that at least one NLP model must include a similarity metric. This similarity metric (e.g., implemented as cosine similarity and / or the Jaccard index) can be used to calculate how similar the event memory data is to each other.

[0027] Further training stipulates that at least one NLP model must include a machine language model. Among other things, this allows for the determination of semantic relationships between event data. In particular, it enables the identification of not only word matches between event data, but also, and especially, between the respective natural language descriptions.

[0028] Further training stipulates that the machine language model includes a Bidirectional Encoder Representations from Transformers (BERT) and / or an artificial transformer network. In particular, BERT and / or the artificial transformer network ensure bidirectional contextual processing of event memory data. This allows event memory data to be evaluated, grouped, and / or classified based on its semantic content and / or independently of word order.

[0029] Further training stipulates that automated fault detection is performed using a server system or the offboard diagnostic system, without requiring any physical intervention on the vehicle(s). In other words, automated fault detection is performed remotely. This can be achieved, for example, via the communication interface. Specifically, this can be done using mobile networks, the internet, and / or Bluetooth. This reduces the workload for workshops, as the event data has already been pre-filtered and / or only the similar event data needs to be examined.

[0030] A further training program stipulates that at least one absolute occurrence time describes the time and date when the event took place, while the relative occurrence time describes a duration or time interval between at least two events. In other words, the absolute occurrence time can mark the beginning of the temporal dependency, while the relative occurrence time describes the duration between this beginning and the end of the temporal dependency. Using the relative occurrence time, a time interval between different events can thus be determined, whereby, by considering the predefined temporal dependency, i.e., a predefined temporal threshold, events and / or (therefore also) event memory data can be automatically identified and / or removed as consequential errors.

[0031] Further training stipulates that event data is evaluated based on its relative and / or absolute frequency of occurrence according to at least one predefined relevance criterion. This evaluation may include the exclusion of event data with, for example, an absolute and / or relative frequency exceeding a predefined threshold. In other words, only event data with a relative and / or absolute frequency below the predefined threshold can be considered and / or transmitted to the workshop and / or the off-board diagnostic system. The predefined threshold for absolute frequency could, for example, range from 5 to 50. The predefined threshold for relative frequency could include the aforementioned temporal dependency. This allows for the prioritization of event data.

[0032] Further training stipulates that it includes at least one predefined relevance criterion: only event log data with a relative and / or absolute frequency below a predefined threshold is considered. It may be possible to apply a combined threshold, which takes into account or includes the respective predefined thresholds for relative and absolute frequency. This allows the event log data to be filtered and / or reduced.

[0033] Further training involves receiving event data from at least one control unit of various vehicle models. In other words, event data can be determined across different models. This allows event data and / or fault detection to be considered not only for a single vehicle, but also for correlations between different model series. In particular, this enables the identification of general faults or events that occur not only in a specific vehicle, but across different models.

[0034] For use cases or application situations that may arise during the procedure and are not explicitly described here, it may be provided that, according to the procedure, an error message and / or a request for user feedback is issued and / or a default setting and / or a predetermined initial state is set.

[0035] The invention also includes the control device for the motor vehicle. The control device can comprise a data processing device or a processor circuit configured to carry out an embodiment of the method according to the invention. For this purpose, the processor circuit can comprise at least one microprocessor and / or at least one microcontroller and / or at least one FPGA (Field Programmable Gate Array) and / or at least one DSP (Digital Signal Processor). In particular, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or an NPU (Neural Processing Unit) can be used as the microprocessor. Furthermore, the processor circuit can comprise program code configured to carry out the embodiment of the method according to the invention when executed by the processor circuit.The program code can be stored in a data memory of the processor device. The processor device can be based, for example, on at least one circuit board and / or on at least one SoC (System on Chip).

[0036] The invention also includes a vehicle system comprising the control device. The control device can be installed in a motor vehicle or distributed between the motor vehicle and a server system of the motor vehicle.

[0037] The motor vehicle according to the invention is preferably designed as a motor vehicle, in particular as a passenger car or truck, or as a passenger bus or motorcycle.

[0038] As a further solution, the invention also includes a computer-readable storage medium comprising program code which, when executed by a computer or a computer network, causes it to execute an embodiment of the method according to the invention. The storage medium can be provided at least partially as a non-volatile data storage medium (e.g., as flash memory and / or as an SSD - solid state drive) and / or at least partially as a volatile data storage medium (e.g., as RAM - random access memory). The storage medium can be located within the computer or computer network. However, the storage medium can also be operated, for example, as an app store server and / or cloud server on the internet. The computer or computer network can provide a processor circuit with, for example, at least one microprocessor.The program code can be provided as binary code, assembly code, source code in a programming language (e.g., C), or a program script (e.g., Python). Alternatively, the computer-readable storage medium can be implemented as a signal containing computer-readable data, such as a time-varying voltage signal or a radio signal.

[0039] The invention also includes combinations of the features of the described embodiments. The invention therefore also includes realizations that each exhibit a combination of the features of several of the described embodiments, provided that the embodiments have not been described as mutually exclusive.

[0040] The following are exemplary embodiments of the invention described. This is illustrated by: Fig. a schematic representation according to one embodiment for reducing event storage data.

[0041] The exemplary embodiments described below are preferred embodiments of the invention. In these exemplary embodiments, the described components each represent individual features of the invention, which can be considered independently of one another and each further develops the invention independently. Therefore, the disclosure is intended to include combinations of features of the embodiments other than those shown. Furthermore, the described embodiments can also be supplemented by further features of the invention already described.

[0042] In the figure, identical reference symbols denote functionally equivalent elements.

[0043] The figure shows a motor vehicle 10 comprising a vehicle system 15 and at least one control unit 2. Together with a server system 20, the vehicle system 15 includes a control device 11. The server system 20 includes, by way of example, event memory data 7 that has been extracted and / or filtered from the at least one control unit 2. This event memory data 7 can additionally or alternatively be stored or stored in the vehicle system 15.

[0044] According to one embodiment of the idea, event memory data 7 can be reduced or filtered to support fault detection in one or more motor vehicles 10. The event memory data 7 can each represent a fault-related event. First, event memory data 7 can be received from at least one control unit 2 of the one motor vehicle 10 or the multiple motor vehicles 10. Subsequently, at least one NLP (Natural Language Processing) model can be applied to the event memory data 7, thereby reducing the event memory data 7 by removing redundant event memory data 7 and / or aggregating similar event memory data 7. The redundant event memory data 7 can comprise semantically similar fault-related events, at least partially.Finally, the remaining event memory data 7 can be transmitted to a workshop and / or an off-board diagnostic system to assist in fault detection.

[0045] The idea can therefore include an automated method using algorithms from machine learning and artificial intelligence. This method can be applied in the (higher-level) server system 20, without requiring any physical work on the (actual) vehicle or motor vehicle 10 to detect and / or analyze faults. Additionally, multiple event logs, containing event log data 7, from different vehicles and / or vehicle fleets, and in particular cross-model vehicle architectures, can be automatically analyzed and / or examined and / or filtered using the embodiment described above.

[0046] A key task of the automated method can be, or include, linking the (various) event memory entries or event memory data 7 from the (various) vehicles and the (individual) absolute and relative times of occurrence. The absolute time of occurrence can describe the actual time and date when an event took place, while the relative time of occurrence encompasses the time interval between two events. Based on the absolute and / or relative time of occurrence, subsequent errors in the (existing) data or event memory data 7 can be found and / or corresponding event memory data can be removed.

[0047] Within the framework of this concept, an NLP model (or at least one) based on natural language processing (NLP) can be developed using machine learning algorithms. This NLP model can be used to extract and / or filter event memory data from a large number of vehicles. This allows for the interpretation of individual event memory entries and a direct assignment to their absolute and / or relative frequency of occurrence. Subsequently, a frequency distribution of the event memory data can be generated across multiple event memory groups. The results of this frequency distribution can then be used to, for example, assign lower weight or lower weight to very frequently occurring events or events exceeding a predefined number, for instance, in relation to a vehicle problem, particularly a breakdown.Event logs with a lower absolute and / or relative frequency of occurrence or below the specified number can, on the other hand, be marked or labelled as more relevant and / or displayed or shown in a list of event logs relevant to the vehicle problem that occurred, in particular the breakdown.

[0048] Once an event memory extract (i.e., the event memory data of a vehicle) has been analyzed or retrieved via the NLP model (e.g., by a workshop or an offboard backend system), a reduced list of event memories can be provided or accessed. These memories are particularly relevant, from a statistical perspective, to a specific vehicle problem, especially a breakdown. The subsequent analysis of the vehicle can then be carried out more efficiently and / or quickly, for example, by the workshop. This can result in shorter downtimes during troubleshooting and / or lower costs for the user.

[0049] One possible example is the following: Take the event log entry from the previous example and perform the analysis or filtering using at least one NLP model from the concept. This can reduce the number of event log entries that are statistically relevant to the vehicle problem, especially the breakdown, from 120 to 10, for example. As a result, only about one-eighth of the analysis time might be required, since the workshop only needs to analyze 10 event log entries.

[0050] Overall, the examples demonstrate how automated AI event memory analysis can be provided for a wide variety of vehicles. Reference symbol list 2 Control unit 7 Event log data 10 motor vehicle 11 Control device 15 Vehicle system 20 server systems

Claims

[1] Method for reducing event memory data (7) to assist fault detection in one or more motor vehicles (10), wherein the event memory data (7) each represent a fault-related event, wherein the following steps are performed: - Receiving event memory data (7) from at least one control unit (2) of the one motor vehicle (10) or of the several motor vehicles (10), - Applying at least one NLP (Natural Language Processing) model to the event memory data (7), thereby reducing the event memory data (7) by removing redundant event memory entries and / or aggregating similar event memory data (7), wherein the redundant event memory entries cover semantically at least partially similar error-related events, - Transmitting the remaining event memory data (7) to a workshop and / or an off-board diagnostic system to assist in fault detection, wherein at least one NLP model includes removing event memory data (7) whose semantic similarity is below a specified similarity threshold and / or grouping event memory data (7) whose semantic similarity is equal to or above the specified similarity threshold as similar event memory data (7) or wherein event memory data (7) that occur in a predefined temporal dependency on other event memory data (7) are marked as consequential errors and removed, wherein the predefined temporal dependency is determined by extracting at least one absolute and one relative occurrence time of the event memory data (7). [2] Method according to claim 1, wherein the at least one NLP model performs a vectorization of the event storage data (7). [3] Method according to any of the preceding claims, wherein the at least one NLP model performs cluster formation. [4] Method according to any of the preceding claims, wherein the at least one NLP model includes a similarity metric. [5] Method according to any of the preceding claims, wherein the at least one NLP model comprises a machine language model. [6] Method according to claim 5, wherein the machine language model comprises a Bidirectional Encoder Representations from Transformers, BERT, and / or an artificial Transformer network. [7] Method according to one of the preceding claims, wherein the automated fault detection is carried out using a server system (20) without requiring any physical intervention on the motor vehicle (10) and / or the multiple motor vehicles (10). [8] Method according to one of the preceding claims, wherein, when the event memory data (7) which occur in a predetermined temporal dependency on other event memory data (7) are marked as consequential errors and removed, wherein the predetermined temporal dependency is determined by extracting at least one absolute and one relative occurrence time of the event memory data (7), the at least one absolute occurrence time describing the time and date when the event took place, wherein the relative occurrence time describing a time interval between at least two events. [9] Method according to one of the preceding claims, wherein, when the event memory data (7) which occur in a predetermined temporal dependency on other event memory data (7) are marked as consequential errors and removed, wherein the predetermined temporal dependency is determined by extracting at least one absolute and one relative occurrence time of the event memory data (7), the event memory data (7) are evaluated on the basis of their relative and / or absolute occurrence frequency according to at least one predetermined relevance criterion. [10] Method according to claim 9, wherein the at least one predefined relevance criterion comprises that only event storage data (7) are considered which include a relative and / or absolute frequency of occurrence below a predefined threshold. [11] Method according to one of the preceding claims, wherein event memory data (7) are received from at least one control unit (2) of different vehicle models of the motor vehicles (10). [12] Control device (11), wherein the control device (11) comprises a processor circuit which has program instructions which, when executed by the processor circuit, cause it to carry out a method according to one of the preceding method claims. [13] Vehicle system (15) comprising a control device (11) according to claim 12, wherein the control device (11) is installed in a motor vehicle (10) or divided between the motor vehicle (10) and a server system (20) of the motor vehicle (10).

Images