Methods for context-dependent evaluation of a vehicle's condition and vehicle

Abstract

Method of a vehicle (1) for evaluating a vehicle condition, the method comprising the following steps: Determine (S1) a component-specific metric for monitoring a vehicle component (60) and a current context (K1, K2, K3) of the vehicle (1); Determine (S2) a diagnostic value (D) of the vehicle component (60) depending on the determined context (K1, K2, K3) within the determined metric; and Determining (S3) a vehicle condition using a computer-based model and based on the determined diagnostic value (D), where weights of the model are determined using a neural network and where training data (T) of the neural network assign diagnostic values ​​(D) of vehicle components (60) to vehicle faults (F).

Description

[0001] The invention relates to a method for a vehicle to evaluate a vehicle's state in a context-dependent manner. Furthermore, the invention relates to a vehicle comprising at least one vehicle component and a control unit configured to perform the method according to the invention.

[0002] Modern vehicles are complex technical systems with a multitude of interacting components. Detecting and addressing faults in these components is crucial for safety, as even a single failure in one component can have critical consequences for the entire system. Especially in passenger vehicles, precise monitoring and analysis of the vehicle's condition is essential to reliably identify and correct malfunctions.

[0003] Especially in the field of autonomous driving, fault detection is of great importance. Automated vehicles offer numerous opportunities to improve driving safety and the driving experience for the driver. However, the driver's autonomy regarding driving decisions is increasingly being transferred to the vehicle or the control units operating within it. Ultimately, these developments will lead to an automated vehicle that can maneuver completely without human intervention.

[0004] SAE J3016 categorizes automated vehicles into different levels. In SAE Level 4 and 5 autonomous vehicles, the systems responsible for autonomous driving operate partially or completely autonomously. This also means that driver intervention is either partially or not at all possible. In these cases, careful monitoring of all systems is essential to detect errors and malfunctions quickly and accurately.

[0005] DE 10 2019 115 092 A1 describes a method for determining the object recognition rate of an artificial neural network for object recognition in an automated motor vehicle. In the method, data with a specific condition are selected from a test dataset. The object recognition rate of the artificial neural network is then determined for the selected data with the specified condition.

[0006] DE 10 2019 135 608 A1 describes a method, a device, and a system for detecting anomalous operating states of a device. Model data (M), representative of an expected operating state of at least one component of the device, are provided to the device. The device acquires measurement data representative of an actual operating state of the at least one component. The device then determines comparative data based on the model data and the measurement data representative of an expected operating state. Based on the comparative data and the measurement data, it is checked whether there is a deviation of the actual operating state from the expected operating state. The final step involves assigning an anomalous operating state to the at least one component corresponding to a time of acquisition of the measurement data, depending on the deviation.

[0007] Conventional fault detection methods typically compare the behavior of vehicle components under test with predefined static limit values. However, the use of static limit values, particularly in automated vehicles, fails to adequately address the diverse and varied requirements of different driving scenarios. For example, a sensor's performance in terms of driving safety during automated driving on a highway may differ from its performance in urban driving. Current fault detection methods either fail to adequately represent this diversity or do so only partially.

[0008] The invention is based on the objective of enriching the state of the art and overcoming or at least reducing the disadvantages arising from it, and of providing an improved method for evaluating a vehicle's condition.

[0009] The problem according to the invention is solved by a method and vehicle according to the claims. Preferred embodiments are the subject of the dependent claims.

[0010] A first aspect of the present disclosure relates to a method for evaluating a vehicle's state. For the purposes of this disclosure, a vehicle is preferably a means of transportation designed to transport persons and / or cargo on Earth, in the air, and / or in space. Preferably, the vehicle is a passenger car with an internal combustion engine, an electric motor, or a hybrid engine. Particularly preferably, the vehicle is a vehicle trained for autonomous driving and, for example, trained for operation in accordance with SAE Level 4 or 5.

[0011] The method according to the invention comprises, as a process step, the determination of a component-specific metric for monitoring a vehicle component. For the purposes of this disclosure, a metric is preferably a reference system for measuring quantifiable quantities. Preferably, at least one suitable metric for monitoring is assigned to each vehicle component. Preferably, this assignment takes place during the manufacturing of the respective vehicle component. The metrics assigned to each vehicle component are preferably stored in a first lookup table (LUT). Preferably, the component-specific metric is determined by looking it up in the LUT.The metric is preferably a scale for plotting a measured value from a sensor, an estimated data volume in a line and / or memory, expected outputs from a software component, RAM and / or CPU usage, vehicle speed and / or acceleration, current, voltage, and / or temperature. In other words, the metric provides a space for each vehicle component in which the sample values ​​relating to the component can be plotted against each other and compared. Particularly preferably, each metric is defined by a standard, especially a distance standard, for comparing sample values.

[0012] Furthermore, the inventive method determines the current context of the vehicle. This context preferably defines the vehicle's environment and / or driving situation. For example, the context reflects whether the vehicle is driving on a highway or in a city, whether it is day or night, and / or whether it is dry or raining. The context can also reflect whether the vehicle is driving manually, semi-automated, or fully automated. The level of detail in the representation of the vehicle's environment and / or driving situation by the context can vary. The context can be determined using data acquired within the vehicle and / or externally, as will be explained in more detail below.

[0013] In a further step, a diagnostic value for the vehicle component is determined based on the identified context within the determined metric. Preferably, a sample value and a context-dependent set of limit values ​​are defined within the determined metric. The diagnostic value is preferably the result of comparing the sample value with the set of limit values.

[0014] As a further step, the method according to the invention includes determining a vehicle state using an AI-based model and based on the determined diagnostic value. In other words, a vehicle state is determined by processing the determined diagnostic value through a model whose weights were generated by artificial intelligence. For the purposes of this disclosure, a vehicle state is an operating state of the vehicle, in particular a safe or an unsafe or critical operating state. The weights of the AI-based model are determined using a neural network. Training data for the neural network assigns diagnostic values ​​of vehicle components to vehicle faults. In other words, the diagnostic values ​​are provided to the neural network as input, and the vehicle faults as output to be learned.The neural network is preferably based on algorithms known to those skilled in the art, such as those that can be created or used with software packages like CNTK, Deeplearning4j, Keras, Mallet, mlpack, Neural Lab, and similar machine learning algorithms. For those skilled in the art, training neural networks using given training data is a simple programming task nowadays, which will not be discussed in detail here. Preferably, the neural network uses the input diagnostic values ​​to determine at least one simulated state variable until the deviation between the simulated state variable and the given training values ​​falls below a predetermined threshold (backpropagation). If this condition is met, the weighting factors of the neural network, for example, the gradients between internal layers of the network, are output to the AI-based model.

[0015] The method according to the invention advantageously enables a more precise and application-specific evaluation of vehicle states by using context-dependent diagnostic values ​​within a component-specific metric. This advantageously leads to increased vehicle safety, as unsafe states can be detected more accurately. In prior art methods for state evaluation by checking a vehicle component, the situation in which the vehicle is at the time of the check was disregarded. However, it is possible that the behavior of a vehicle component is error-free or even desirable in a certain context, while the same behavior would be dangerous in another context. For example, if a vehicle control unit sets a speed of 130 km / h on the highway, this is perfectly normal behavior. However, if this occurs in a pedestrian zone, a fault must be assumed.According to the present invention, this circumstance is taken into account, and an improved method for evaluating the vehicle condition based on vehicle components is provided. The use of the AI-based model for determining the vehicle condition is a further advantage of the method, since the model's weights are updated and improved by the learning neural network. Thus, the vehicle's safety is increased by the method according to the invention by incorporating the vehicle's current context and using the AI-based model to determine the vehicle condition.

[0016] In a preferred embodiment of the method according to the invention, a plurality of metrics for monitoring a plurality of vehicle components are determined, and at least one diagnostic value is determined for each of the plurality of vehicle components. Preferably, the method according to the invention is carried out using each of the plurality of metrics and for each of the plurality of vehicle components. Preferably, a plurality of diagnostic values ​​are determined for at least one of the vehicle components, wherein the diagnostic values ​​are determined within a plurality of metrics for monitoring the at least one vehicle component. For example, a LIDAR sensor can be monitored by evaluating its distance measurement, its temperature, and its current consumption. This advantageously increases the statistical relevance of fault detection for the vehicle component.For example, if all three diagnostic values ​​from the LIDAR sensor are abnormal, then a fault is highly likely. If only two of the diagnostic values ​​are abnormal, then the sensor may be faulty but not yet defective. By using multiple metrics per vehicle component, a multidimensional definition of context-dependent fault states is possible, which further enhances the application-specificity of the vehicle condition evaluation.

[0017] In a further preferred embodiment, determining a diagnostic value includes, as a first step, determining a sample value of the vehicle component within the determined metric. Preferably, the sample value is determined as an output of the vehicle component during its operation. Alternatively, and more preferably, the vehicle component is stimulated with an input to determine the sample value. More preferably, the sample value is determined by means of a sensor on the vehicle component. Furthermore, in the method, a set of limit values ​​within the determined metric is determined depending on the determined context. Preferably, a lower and an upper context-dependent limit value are determined within the determined metric. The context-dependent limit values ​​are preferably stored in a second LUT. Preferably, the first LUT and the second LUT are parts of a single LUT.Preferably, the set of limit values ​​is determined by looking them up in the LUT. The LUT is preferably stored locally in the vehicle's memory or retrieved from a server in a network. The server is preferably configured to manage and update the LUT. Preferably, the server communicates with a plurality of vehicles performing the method according to the invention and updates the LUT based on feedback from the plurality of vehicles. Managing the LUT by a server in a network thus advantageously leads to a continuous adaptation and improvement of the LUT and therefore of the limit values ​​used.

[0018] The diagnostic value is then preferably determined as the quotient of the difference between the determined sample value and the lower limit value, and the difference between the upper limit value and the lower limit value. In other words, the diagnostic value is calculated as the result of the following formula: Diagnostic value = Sample value − lower limit value / upper limit value − lower limit value

[0019] In other words, the diagnostic value is formed by normalizing the sample value to a set of limit values. Preferably, a diagnostic value outside the range of 0 to 1 indicates a fault in the vehicle component.

[0020] By determining the diagnostic value according to the preferred embodiment described above, advantageously scaled diagnostic values ​​for all vehicle components can be determined. The diagnostic value thus advantageously enables a comparison of vehicle components that are monitored using different metrics.

[0021] In a further preferred embodiment, the training data is generated by the vehicle and / or at least one other vehicle when a vehicle fault is detected. Preferably, the training data includes at least one diagnostic value from at least one vehicle component. More preferably, the training data includes a plurality of diagnostic values ​​from a plurality of vehicle components. In other words, the vehicle and / or at least one other vehicle generate the training data each time a vehicle fault occurs. Preferably, the training data is generated by a plurality of vehicles.

[0022] Furthermore, the training data is preferably stored in a database. This database is preferably stored on a server within a network that communicates with a multitude of vehicles. If a vehicle fault occurs in one of the vehicles, this vehicle preferably transmits the fault, along with at least one diagnostic value from a vehicle component, to the server that manages and updates the database. The vehicle component whose diagnostic value is transmitted is preferably the component that caused the fault. Furthermore, preferably, all diagnostic values ​​from all vehicle components are transmitted to the server along with the fault. Using training data generated in this way, the neural network can be advantageously trained to recognize faults caused by more than one vehicle component.Furthermore, diagnostic values ​​from vehicle components, obtained before the vehicle fault occurred, are preferably transmitted along with the vehicle fault information. Training data generated in this way can advantageously train the neural network to predict vehicle faults even before they occur.

[0023] Another preferred embodiment of the method according to the present disclosure provides that the vehicle state indicates a fault-free state of the vehicle, a fault in at least one vehicle component, and / or an impending fault in at least one vehicle component. Preferably, the vehicle state indicates a fault and / or impending fault of a plurality of vehicle components. More preferably, the vehicle state indicates whether the vehicle is in a critical state and / or whether action is required in response to the vehicle's state. Preferably, the indication of whether a critical state exists and / or whether action is required depends on the number and / or type of faulty and / or imminently faulty vehicle components.

[0024] Each vehicle component is preferably assigned to a category. For example, Category 1 vehicle components are safety-relevant and not redundantly installed in the vehicle, Category 2 vehicle components are safety-relevant with redundancy, and Category 3 vehicle components are not safety-relevant. In this example, if a Category 1 vehicle component were defective, the vehicle status would indicate a critical condition and a required action, such as emergency operation of the vehicle and / or a visit to a workshop. If a Category 2 component is defective, but the redundant component is functional, a workshop visit would be indicated as a required action; however, if the redundant component is also defective, emergency operation of the vehicle is also required.For example, a defective vehicle component of category 3 only requires a warning light as a measure; if a majority of category 3 vehicle components are faulty, a visit to a workshop may be necessary, for example depending on whether the number of defective vehicle components exceeds a previously defined limit.

[0025] The indication of whether a critical condition exists and / or whether a measure is required is also preferably dependent on the type of vehicle and / or the type of use of the vehicle, and furthermore preferably context-dependent.

[0026] A further preferred embodiment of the method according to the invention also includes determining whether a measure is required in response to the determined vehicle condition. Preferably, the determined vehicle condition, as described above, indicates whether a measure is required. Also preferably, at least one necessary measure can be determined based on the vehicle condition. A measure in response to the determined vehicle condition preferably includes emergency operation of the vehicle, a visit to a workshop, and / or activation of a warning light. A measure also preferably includes an emergency maneuver such as emergency braking and / or activation of the hazard warning lights, a safety precaution such as reducing the speed, and / or activating additional redundant hardware and software components as replacements for faulty vehicle components and / or vehicle components expected to fail soon.

[0027] If a measure is required, it will be implemented as a priority. If no measure is required, the preceding procedural steps will be repeated as a priority.

[0028] In a preferred embodiment of the method according to the invention, the context is defined as a function of a plurality of parameters characterizing the vehicle's environment and / or driving situation. The vehicle's environment is preferably characterized by parameters relating to the vehicle's surroundings, such as geographical data, weather data, traffic data, and legal requirements. A driving situation is preferably characterized by parameters relating to the vehicle itself, such as the state of the assistance systems, an SAE level, the number of passengers, the range, or the distance remaining. The parameters are preferably determined by means of vehicle sensors and / or based on external inputs. The context can be determined by the vehicle itself and / or by an external server. The context preferably consists of a plurality of subcontexts.A subcontext preferably includes a single feature of the vehicle's environment and / or driving situation, such as the weather or road type. Preferably, the context corresponds to an operational design domain (ODD) determined by the vehicle. According to the SAE J3016 standard, an ODD comprises the conditions under which a specific automated driving system or a function thereof is specifically designed to operate.

[0029] In a further preferred embodiment of the method according to the invention, the diagnostic value of the vehicle component is determined at a context-dependent frequency. Preferably, a sample value and a context-dependent set of limit values ​​are defined within the determined metric. The diagnostic value is preferably the result of a comparison of the sample value with the set of limit values. Preferably, the method according to the invention is carried out with each determined diagnostic value. Preferably, the sample value is determined as a periodic subset of the output values ​​of the vehicle component during the vehicle component's operation. Alternatively, preferably, the vehicle component is periodically excited with an input to determine the sample value. Furthermore, preferably, the sample value is determined periodically by means of a sensor on the vehicle component. The frequency of the periodic determination is preferably context-dependent.Preferably, the frequency is increased if the vehicle component is safety-relevant in the determined context. Preferably, the frequency is decreased if the vehicle component is not safety-relevant in the determined context and / or is not used.

[0030] Another aspect of the invention relates to a vehicle, in particular a passenger car with an internal combustion, electric, or hybrid engine. The vehicle is particularly preferably a vehicle designed for automated driving and, for example, configured for operation according to SAE Level 4 or 5. The vehicle preferably comprises a vehicle component. Furthermore, the vehicle comprises a control unit configured to perform the previously described inventive method for evaluating a vehicle state.

[0031] A preferred embodiment of the vehicle according to the invention comprises a plurality of vehicle components. The control unit is preferably configured to perform a method as described above for each of the plurality of vehicle components.

[0032] Further preferred embodiments of the vehicle according to the invention correspond to the preferred embodiments of the method according to the invention described above.

[0033] Another aspect of the invention relates to a computer program comprising instructions which, when the program is executed by a computer, such as a control unit of a vehicle, cause it to execute the method according to the invention as described above.

[0034] Another aspect of the invention relates to a computer-readable storage medium comprising instructions which, when executed by a computer, such as a control unit of a vehicle, cause it to execute the method according to the invention as described above.

[0035] Further preferred embodiments of the invention result from the other features mentioned in the dependent claims.

[0036] Unless otherwise stated in individual cases, the various embodiments of the invention mentioned in this application can be advantageously combined with one another.

[0037] The invention is explained below using exemplary embodiments with reference to the accompanying drawings. These show: Fig. 1 a schematic flowchart of a method according to the invention; Fig. 2a a graphical representation of an exemplary sample value determined over a period of time and static limit values; Fig. 2b a graphical representation of an exemplary sample value determined over a period of time and context-dependent limit values; Fig. 3 a schematic flowchart of a method according to the invention in accordance with an exemplary implementation form; Fig. 4 a schematic flowchart of part of a method according to the invention in accordance with a further exemplary implementation form; Fig. 5 a graphical representation of the training of a neural network and adjustments of AI-based models according to one embodiment; and Fig. 6 a schematic representation of a vehicle according to the invention in one embodiment.

[0038] Fig. Figure 1 shows a schematic flowchart of a method according to the invention, in particular a method of a vehicle for evaluating a vehicle condition.

[0039] The procedure includes, as a process step S1, the determination of a component-specific metric for monitoring the vehicle component. Additionally, in process step S1, the current context of the vehicle is determined. This context defines the vehicle's environment and / or driving situation.

[0040] In a second process step S2, a diagnostic value for the vehicle component is determined within the determined metric, depending on the context established in the first process step S1. Specifically, a sample value and a context-dependent set of limit values ​​are determined within the established metric. The diagnostic value is the result of comparing the sample value with the set of limit values.

[0041] In the third process step S3, a vehicle condition is determined using a computer-based model and based on the diagnostic value determined in the second process step S2. Weights for the model are determined using a neural network, and training data for the neural network assigns diagnostic values ​​from vehicle components to vehicle faults.

[0042] Fig. 2a and Fig. Section 2b illustrates, using graphical examples, how a sample value P and limit values ​​G1, G2, G1', G2' are determined and compared over a period of time. This involves... Fig. 2a static limit values ​​G1', G2' and in Fig. 2b Context-dependent limit values ​​G1, G2 are shown.

[0043] The sample value P is determined within a defined metric. In the inventive method, this metric is determined specifically for the vehicle component being monitored. Sensors, for example, typically have defined operating ranges. This operating range can be used to indicate whether the data supplied by the sensor is correct. Switches are equipped with message buffers to process multiple messages. The maximum number of messages in the buffer is known. Therefore, monitoring the number of messages in the buffer can detect a potential overload of the switch. Applications required for vehicle operation perform actions on a set of input parameters and provide an output, which can be used as a metric to monitor the respective application. It is also possible to monitor a vehicle component using multiple metrics.For example, both utilization and throughput can be used as metrics for a software component. In such a case, the method according to the invention is preferably carried out with the majority of metrics.

[0044] In the Fig. 2a and Fig. In example 2b, a vehicle control unit is to be monitored. The control unit's output includes, for example, instructions that control the vehicle's longitudinal and lateral movement. Therefore, an output for a set vehicle speed was determined as an example metric.

[0045] The sample value P determined within the metric is represented as a solid line. It initially rises rapidly, then decreases over time until it experiences a brief, sharp increase. This increase in the determined sample value P is marked with a lightning bolt, as it indicates an error F in the control unit.

[0046] In Fig. Figure 2a shows how the sample value P is compared to a set of static limit values ​​G1', G2', which are represented as dashed lines. The lower limit value G2' has the value 0 km / h and the upper limit value G1' has the value of the vehicle's maximum speed. The sample value P lies consistently between the static limit values ​​G1', G2', which means that the error F is not detected.

[0047] In Fig. In step 2b, the sample value P is compared to a set of context-dependent limits G1 and G2. The lower limit G2 is set to 0 km / h. The upper limit G1 takes on different values ​​depending on the context K1, K2, and K3. In the first context K1, the vehicle is on a highway without a speed limit. Therefore, the upper limit G1 is set to the vehicle's maximum speed. In the second context K2, the vehicle is in a city, so the upper limit G1 is set to 50 km / h. In the third context K3, the vehicle is in a parking lot, so the upper limit G1 is set to walking speed. Now, in the third context K3, the sample value P lies outside the limits G1 and G2. The error F is detected and can be addressed.

[0048] The in Fig. 2a and Fig. The example shown in Figure 2b illustrates the advantage of the context-dependent fault detection of the method according to the invention. The method enables more precise detection of faults F in vehicle components and therefore advantageously leads to increased vehicle safety. To enable more precise fault detection, a context K1, K2, K3 consists, in particular, of a plurality of subcontexts. If, for example, the context K1 were to include, in addition to the subcontext "highway," also the subcontext "rain" or "traffic jam," the speed of the upper limit value G1 would be determined to be lower.

[0049] Fig. Figure 3 shows a schematic flowchart of a method according to the invention in an exemplary embodiment. In the first method step S1, as described above, a component-specific metric and a current context K1, K2, K3 of the vehicle are determined.

[0050] In this exemplary embodiment, the second process step S2 is divided. In the first part of the second process step S2a, the sample value P is determined within the established metric. In the second part S2b, the limit values ​​G1 and G2 are then determined based on the context established in the first process step S1. The third part of the second process step S2c follows, in which the diagnostic value is determined. Specifically, the diagnostic value is calculated as the quotient of the difference between the determined sample value P and the lower limit value G2, and the difference between the upper limit value G1 and the lower limit value G2. The diagnostic value is thus determined as the result of the following formula: Diagnostic value = P−G2G1−G2

[0051] In the third process step S3, a vehicle condition is determined using the AI-based model and based on the diagnostic value determined in the third part of the second process step S2c. Specifically, in this exemplary embodiment, the third process step S3 also determines whether a measure is required in response to the determined vehicle condition. If a measure is required, the process follows the first path W1a. In the fourth process step S4, the measure is then implemented. If no measure is required, the process follows the second path W1b. In this case, a critical vehicle condition has been detected. The process then restarts with the first process step S1.

[0052] Fig. Figure 4 shows a schematic flowchart of part of a method according to the invention in a further exemplary embodiment. It illustrates how a plurality of diagnostic values ​​D are determined for each of a plurality of vehicle components 60. In particular, a plurality of diagnostic values ​​Da1 to Dan are determined for a first vehicle component 60a, the diagnostic values ​​Db1 to Dbn for a second vehicle component 60b, and the diagnostic values ​​Dc1 to Dcn for a third vehicle component 60c. The diagnostic values ​​are determined, in particular, within a plurality of metrics and / or at different times. All diagnostic values ​​Da1 to Dan, Db1 to Dbn, and Dc1 to Dcn are then incorporated in the first part of the third process step S3a into the determination of the vehicle state using the AI-based model, which is graphically represented by the arrows.In the second part of the third procedure step S3b, it is then determined whether a measure is required in response to the detected vehicle condition. If a measure is required, the procedure follows the first path W1a, as described previously. In the fourth procedure step S4, the measure is then implemented. If no measure is required, the procedure follows the second path W1b. In this case, no critical vehicle condition was detected. The procedure then starts again with the first procedure step S1 and repeats all preceding procedure steps.

[0053] For example, the first diagnostic value of the third vehicle component, Dc1, shows an irregularity. However, since the third vehicle component is not safety-relevant and the other diagnostic values, Dc2 to Dcn, are unremarkable, a safe vehicle state is determined in the first part of the third procedure step, S3a. Accordingly, no required action is determined in the second part of the third procedure step; the procedure follows the second path, W1b, and starts again. During the subsequent procedure run, all diagnostic values, Da1 to Dan, of the first vehicle component, 60a, are now abnormal. This component is indeed safety-relevant, but it has a redundant second vehicle component, 60b. The diagnostic values, Db1 to Dbn, of the second vehicle component indicate that it, too, could soon become faulty.

[0054] In step S3a, a vehicle state is determined based on the diagnostic values, indicating a faulty first vehicle component 60a and a faulty vehicle component 60b that will soon follow. Based on this, the next step S3b determines that a measure is required, whereupon the procedure follows the first path W1a and implements this measure in the fourth procedure step S4. For example, the vehicle could be instructed to reduce its speed, pull onto the shoulder, and stop there with its hazard warning lights activated. For pulling onto the shoulder, the redundant, not yet faulty second vehicle component 60b is used instead of the first vehicle component 60a.

[0055] Fig. Figure 5 shows a graphical representation of the training of a neural network and the adaptation of AI-based models according to one embodiment. In this exemplary embodiment, the neural network and a database containing training data T are implemented in a server of a network 53. A plurality of vehicles 1 to 52 are configured to communicate with the server of the network 53. The training data T in the database includes, in particular, diagnostic values ​​D of vehicle components 60, which are assigned to vehicle faults F. In a fifth process step S5, the training data T is used, in particular, to train the neural network. Specifically, in the fifth process step S5, the diagnostic values ​​are provided to the neural network as input and the vehicle faults as output to be learned.The neural network then uses the input diagnostic values ​​to determine at least one simulated state variable until the deviation between the simulated state variable and the specified training values ​​falls below a predetermined threshold (backpropagation). If this condition is met, the weighting factors of the neural network, for example, the gradients between internal layers of the network, are output in a sixth process step S6 to the AI-based models implemented within vehicles 1 to 52.

[0056] If vehicles 1 to 52 detect a vehicle fault F, they transmit the fault F, along with at least one diagnostic value D of a vehicle component 60, to the server of the network 53 in a seventh process step S7. This server manages and updates the database. The neural network is then trained with the updated training data in step S5, and the weights of the models for vehicles 1 to 52 are updated in step S6. This allows the models, and thus the method according to the invention, to be continuously updated and improved.

[0057] Fig. Figure 6 shows a schematic representation, in particular a block diagram, of an exemplary vehicle 1, especially a two-track motor vehicle with an internal combustion, electric or hybrid engine. The vehicle 1 is specifically designed to perform autonomous driving at SAE Level 4 or 5. The vehicle 1 comprises a plurality of vehicle components 60.

[0058] A number of first sensors comprise a portion of the vehicle components 60, specifically a first sensor 11, a second sensor 12, and a third sensor 13. The first sensors 11, 12, and 13 are configured to acquire environmental data from the vehicle 1 and include, for example, a camera for capturing an image of the roadway, traffic signs, and / or lane markings in front of the vehicle 1; distance sensors, such as ultrasonic sensors, for detecting distances to objects surrounding the vehicle 1; thermometers for detecting the vehicle's ambient temperature; and / or a rain sensor for detecting weather data. The first sensors 11, 12, and 13 transmit the environmental signals they acquire to a control unit 40 of the vehicle 1.

[0059] The vehicle 1 also has, as further vehicle components 60, a plurality of secondary sensors, in particular a fourth sensor 21, a fifth sensor 22, and a sixth sensor 23. The secondary sensors 21, 22, and 23 are sensors for determining vehicle data relating to the vehicle 1 itself, in particular current position and movement information of the vehicle 1. The secondary sensors are therefore, for example, speed sensors, acceleration sensors, tilt sensors, or the like. The secondary sensors 21, 22, and 23 transmit the status signals they detect to the control unit 40 of the vehicle 1.

[0060] Another vehicle component is a communication module 30 with a memory 31 and one or more transponders or receivers 32. The transponder 32 is a radio, WLAN, GPS, or Bluetooth receiver, or the like, in particular a transponder configured for communication in a communication network. The transponder communicates with the internal memory 31 of the communication module 30, for example, via a suitable data bus. Using the transponder 32, the current position of the vehicle 1 can, for example, be determined by communication with a GPS satellite 51 and stored in the internal memory 31. Furthermore, the communication module 30 is configured to communicate with another vehicle 52 via V2V communication, preferably via a communication network 53.Furthermore, the communication module 30 is also configured to communicate with a server of the communication network 53. The communication module 30 also communicates with the control unit 40. In particular, it transmits received data to the control unit and / or receives data to be sent from it.

[0061] The communication network 53 is preferably a network conforming to the 3GPP standard, for example, an LTE, LTE-A (4G), or 5G communication network. The communication network may also be designed for the following operations or conform to the following standards: High Speed ​​Packet Access (HSPA), a Universal Mobile Telecommunication System (UMTS), UMTS Terrestrial Radio Access Network (UTRAN), evolved-UTRAN (e-UTRAN), Global System for Mobile Communication (GSM), Enhanced Data Rates for GSM Evolution (EDGE), or GSM / EDGE Radio Access Network (GERAN). Alternatively or additionally, the communication network may also be designed according to one of the following standards: Worldwide Inter-operability for Microwave Access (WiMAX) network IEEE 802.16 or Wireless Local Area Network (WLAN) IEEE 802.11.The communication network also preferably uses one of the following encoding methods: Orthogonal Frequency Division Multiple Access (OFDMA), Time Division Multiple Access (TDMA), Code Division Multiple Access (CDMA), Wideband-CDMA (WCDMA), Frequency Division Multiple Access (FDMA) or Spatial Division Multiple Access (SDMA), etc.

[0062] The vehicle 1 further comprises a control unit 40 according to the invention, which is configured for fully automated driving operation, in particular for longitudinal and lateral control of the vehicle 1. For this purpose, the control unit 40 has an internal memory 41 and a CPU 42, which communicate with each other, for example via a suitable data bus. In addition, the control unit 40 is in communication connection with at least the first sensors 11, 12, 13, the second sensors 21, 22, 23 and the communication module 30, for example via one or more respective CAN connections, one or more respective SPI connections or other suitable data connections. The control unit 40 is also specifically configured to carry out the method according to the invention, described in detail above, for each of the plurality of vehicle components 60. Reference symbol list 1 vehicle 11 first sensor 12 second sensor 13 third sensor 21 fourth sensor 22 fifth sensor 23 sixth sensor 30 communication units 31 GB internal storage 32 transceivers 40 Control unit 41 GB internal storage 42 CPU 51 satellite 52 more vehicles 53 Network 60 vehicle components 60a first vehicle component 60b second vehicle component 60c third vehicle component Diagnostic value Da1 first diagnostic value of the first vehicle component Then nter diagnostic value of the first vehicle component Db1 first diagnostic value of the second vehicle component Dbn nter diagnostic value of the second vehicle component DC1 first diagnostic value of the third vehicle component Dcn nter diagnostic value of the third vehicle component F error G1 upper limit G1' upper limit G2 lower limit G2' lower limit K1 first context K2 second context K3 third context P Sample value S1 first process step S2 second procedure step S2a first part of the second procedure step S2b second part of the second procedure step S2c third part of the second procedure step S3 third process step S3a first part of the third procedure step S3b second part of the third procedure step S4 fourth process step S5 fifth process step S6 sixth procedure step S7 seventh process step T Training data W1a first way W1b second way

Claims

[1] Method of a vehicle (1) for evaluating a vehicle condition, the method comprising the following steps: Determine (S1) a component-specific metric for monitoring a vehicle component (60) and a current context (K1, K2, K3) of the vehicle (1); Determine (S2) a diagnostic value (D) of the vehicle component (60) depending on the determined context (K1, K2, K3) within the determined metric; and Determining (S3) a vehicle condition using a computer-based model and based on the determined diagnostic value (D), where weights of the model are determined using a neural network and where training data (T) of the neural network assign diagnostic values ​​(D) of vehicle components (60) to vehicle faults (F). [2] Method according to claim 1, wherein a plurality of metrics for monitoring a plurality of vehicle components (60) of the vehicle (1) is determined and at least one diagnostic value (D) is determined for each of the plurality of vehicle components (60). [3] Method according to one of claims 1 or 2, wherein determining (S2) a diagnostic value (D) comprises the following steps: Determine (S2a, S2b) a sample value (P) of the vehicle component (60) within the determined metric and, depending on the determined context (K1, K2, K3), a set of limit values ​​(G1, G2) within the determined metric; Determining (S2c) the diagnostic value (D) as the quotient of the difference between the determined sample value (P) and the lower limit (G2) and the difference between the upper limit (G1) and the lower limit (G2). [4] Method according to any of the preceding claims, wherein the training data (T) are generated by the vehicle (1) and / or at least one other vehicle (52) when a vehicle fault (F) is detected and include at least one diagnostic value (D) of at least one vehicle component (60). [5] Method according to any of the preceding claims, wherein the vehicle state indicates a fault-free state of the vehicle (1), a fault (F) of at least one vehicle component (60) of the vehicle (1) and / or an impending fault (F) of at least one vehicle component (60) of the vehicle (1). [6] Method according to any of the preceding claims, further comprising: Determine (S3b) whether an action is required in response to the determined vehicle condition; if a measure is required (W1a), carry out (S4) the measure; and If no action is required (W1b), repeat the preceding procedural steps (S1, S2, S3). [7] Method according to one of the preceding claims, wherein the context (K1, K2, K3) is defined depending on a plurality of parameters characterizing an environment and / or driving situation of the vehicle (1). [8] Method according to any of the preceding claims, wherein the diagnostic value (D) of the vehicle component (60) is determined with a context-dependent frequency. [9] Vehicle (1) comprising a vehicle component (60) and a control unit (40) configured to carry out a method according to any one of claims 1 to 8. [10] Vehicle (1) according to claim 9, comprising a plurality of vehicle components (60), wherein the control unit (40) is configured to perform a method according to any one of claims 1 to 8 with each of the plurality of vehicle components (60).

Images