Prioritizing repairs of motor vehicles

Abstract

An exemplar of a specified component (110) at risk of failure is installed in each of a specified number (100) of motor vehicles (105). A method (200) for prioritizing repairs of the motor vehicles (105) has the steps of determining (210) technical functional conditions of an exemplar of the component (110) on a respective motor vehicle (105); examining (230) said exemplar ; determining (230) a risk of failure of the exemplar under the determined functional conditions; determining (220), on the basis of previously determined functional conditions and risks of failure of examined exemplars, technical functional conditions under which the risk of failure of the component (110) is increased; and selecting (225) another motor vehicle (105) having technical functional conditions which indicate the highest possible risk of failure for the component (110).

Description

[0001] Prioritizing motor vehicle repairs

[0002] The present invention relates to the repair of a large number of motor vehicles, for example, as part of a recall campaign. In particular, the invention relates to the prioritization of motor vehicle repairs.

[0003] A subset of a series-produced motor vehicle may contain a component that later proves to be potentially defective or potentially non-functional. Under certain circumstances, a combination of components or a specific operating mode can also lead to defects. For safety reasons, in such a case, all affected motor vehicles are recalled and repaired by the manufacturer. Which motor vehicles are affected can usually be determined based on manufacturing records.

[0004] If the number of affected vehicles is large, it is logistically impossible to repair all of them simultaneously or within a very short period of time. One object underlying the present invention is to provide an improved technique for determining an advantageous order in which the vehicles are repaired. The invention solves this problem by means of the subject matter of the independent claims. Subordinate claims specify preferred embodiments.

[0005] A predetermined number of motor vehicles are each fitted with an example of a predetermined component at risk of failure. According to a first aspect of the present invention, a method for prioritizing repairs to motor vehicles comprises the steps of determining technical functional conditions of an example of the component on an assigned motor vehicle; examining this example; determining a failure risk of this example under the determined functional conditions; determining technical functional conditions under which the failure risk of the component is increased, based on previously determined functional conditions and failure risks of examined examples; and selecting another motor vehicle with technical functional conditions that indicate the highest possible failure risk for the component.

[0006] Using this method, a database can be improved for each vehicle examined, on the basis of which a subsequent vehicle is selected. The vehicle can be repaired during or after the inspection. The component can be replaced or repaired. The method can be distributed so that vehicles can be examined and repaired preferentially at different locations, with findings from the examinations being pooled. This allows the database to grow rapidly and allow for a further improved selection of subsequent vehicles.

[0007] The procedure can be used to prioritize the rapid repair of vehicles with an increased risk of failure as part of a vehicle recall campaign. The higher the risk of failure, the higher the priority a vehicle can be given. This can increase overall operational safety when using the vehicle. The overall risk of damage or consequential damage can be quickly and effectively reduced. Preference is given to selecting a vehicle with the highest possible specific component failure risk.

[0008] Naturally, untreated motor vehicles exhibit increasingly lower average failure risks as the fleet's repairs progress. In one embodiment, treatment of remaining motor vehicles can be omitted if it can be determined with a sufficiently high probability that these vehicles are not affected by the problem or that their failure risk is sufficiently low. It is further preferred that a relationship between the failure risk and the technical functional conditions be determined using a machine learning method. The method can comprise training a model. Advantageously, the fact that such a method can map a large number of input parameters to a small number of output parameters can be exploited.Especially when it's initially unclear which functional conditions correlate with a failure risk, training can quickly provide useful results. The further the vehicle testing progresses, the better the model can be trained, and the better it can predict which vehicles are potentially at high risk of failure.

[0009] The functional conditions may include a configuration of the motor vehicle. For example, the component may be particularly problematic if it is installed in combination with one or more other components in a motor vehicle. For example, it may turn out that a defect in the component occurs more frequently at high operating temperatures. Motor vehicles in which the component is installed near another component that dissipates a lot of heat may therefore have a higher risk of failure. Other motor vehicles that do not have such an additional component, in contrast, may be subject to a lower risk of failure. The configuration may be determinable from the motor vehicle's production documentation.

[0010] The functional conditions may also include a usage parameter of the motor vehicle. The usage parameter may include an operating parameter, for example, an electrical voltage, an electrical current, a temperature, a force, or a torque.

[0011] The usage parameter can also include a boundary condition such as vibration or humidity. The usage parameter can also include an operating frequency or operating duration. Ultimately, the operating parameter preferably depends on the manner in which the motor vehicle is used or stored. A usage parameter can occur individually on the finished motor vehicle and be measured or determined there. Some usage parameters are recorded on board a motor vehicle as standard, for example, driving speed or drive torque.

[0012] In a preferred embodiment, the additional motor vehicle is determined with respect to a subset of considered functional conditions. For example, from a multitude of potentially suitable functional conditions, a small number can be selected that correlate particularly strongly with the failure risk. Under certain circumstances, a combination of functional conditions can also correlate with the failure risk, for example, an increased temperature in conjunction with increased relative humidity. By isolating those functional conditions that demonstrably correlate strongly with the failure risk, consideration of other functional conditions that correlate only weakly or not at all with the failure risk can be omitted. The selection of an additional motor vehicle can be faster and with less effort. Discrimination can be improved due to reduced noise in the model's input data.In addition, the model can be trained more quickly with fewer parameters and the determination of such a streamlined model can have increased confidence.

[0013] The other motor vehicle can be identified, in particular with regard to predetermined functional conditions, using the Random Forest algorithm. A random forest is a machine learning technique in which several uncorrelated decision trees are generated during training. Each decision tree is trained with a different, randomly selected sample of the training data. Preferably, each tree only considers a randomly selected subset of all features at each node when allocating the objects from its sample. All trees are then combined to form an ensemble, the Random Forest. The result of the Random Forest can be generated from the results of all trees using an aggregate function. For classification tasks, the result corresponds to the class selected by the most trees. For regression tasks, the result is calculated as the average of the results of all trees.The use of random forests can correct deviations in the results of individual decision trees due to overfitting.

[0014] Further preferably, a functional condition is determined in which the density distribution of failed units is as different as possible from the density distribution of functional units. The determined functional condition can be used as the basis for the selection of the additional motor vehicle.

[0015] In one embodiment, for a parameter, the relative proportion of the examined motor vehicles exhibiting a defect related to the component (“sick” vehicles) is plotted over its value range, creating a first density distribution. A second density distribution can be created for the same parameter for motor vehicles that do not have the defect (“healthy” vehicles). If the density distributions differ only slightly, the parameter has only a minor influence on the presence or absence of the defect. Any condition or event can be defined as a defect, for example the occurrence of a predetermined symptom or the failure of a component.

[0016] The failure risk can be determined in different ways. In a first embodiment, the failure risk relates to the probability of a failure. The probability can be determined under the assumption that the component continues to operate under the specific functional conditions. The failure can affect the component under investigation or another component functionally or causally related to it. A predetermined duration or operating period can be assumed for the failure probability. The higher the probability that a failure will occur within the assumed time period, the greater the failure risk can be.

[0017] In a second embodiment, the failure risk may relate to the shortness of the component's expected remaining service life. Here, too, it can be assumed that the component will continue to operate under the specific operating conditions. In other words, the risk of failure may be higher the shorter the expected duration or service life until failure.

[0018] In a third embodiment, the failure risk can relate to the potential severity of a failure. The severity can be assumed to be possible, expected, or probable. The severity can relate to a danger to material or people. With regard to material, a severity that can be compensated for with little financial outlay can be classified as low, and a severity that requires considerable financial outlay for compensation can be determined as high. With regard to a person, an injury, permanent injury, or even death of a person can be assumed to have a high severity of consequences, whereas only a minor injury or merely a disturbance to the person can be assumed to have a low severity of consequences.

[0019] It should be noted that other types of default risk can also be applied and that several different default risks can be determined and combined into an overall value.

[0020] The functional conditions of the specimens can be collected in the motor vehicles during operation. For example, logging data, operating data, error data, or statistical data can be collected and forwarded in the motor vehicle. Preferably, the data is already provided before the motor vehicle is serviced with regard to the component. Optionally, it is possible to select which data is collected and to what extent it is collected. This allows targeted logging and aggregation of data that may be related to the component. The data can be made available externally via a preferably wireless interface. More preferably, the data is transmitted to a central control device and processed there.

[0021] In general, it is preferable that the components be repaired on the vehicles in the specified order. Minor deviations between the order and the specified failure risk can be tolerated. The process can be discontinued if it can be determined that the remaining vehicles have a sufficiently low failure risk.

[0022] It is further preferred that examples of the components are initially examined on a predetermined proportion of motor vehicles. Good results have been achieved with an initial proportion of approximately 5% of all motor vehicles. A smaller proportion is also possible, but may lead to less favorable selection results at the beginning of the method. The predetermined proportion can be randomly determined in the manner of a sample. Alternatively, the proportion can be selected using any heuristic, for example, based on expert knowledge or experience with other motor vehicles that may have been analyzed by another manufacturer.

[0023] According to a second aspect of the present invention, a control device for prioritizing repairs of motor vehicles described herein comprises: a first interface for recording technical functional conditions of an example of the component on an associated motor vehicle; a second interface for recording an examination result of this example; wherein the examination result includes a failure risk of this example under the determined functional conditions; and a processing device. The processing device is configured to determine, based on previously determined functional conditions and failure risks of examined examples, technical functional conditions under which the failure risk of the component is increased.Preferably, an output device is also provided for providing an indication of another motor vehicle with technical functional conditions that indicate the highest possible risk of failure for the component.

[0024] The processing device is preferably configured to partially or completely execute a method described herein. For this purpose, the processing device can be implemented electronically and, for example, comprise an integrated circuit, a programmable logic module, or a programmable microcomputer. The method can be implemented in the form of a configuration or as a computer program product with program code means for the processing device. The configuration or the computer program product can be stored on a computer-readable data carrier. Features or advantages of the method can be transferred to the device, or vice versa.

[0025] In a further development of the invention, the control device may comprise a device for examining a specimen of the component. The device may comprise a physical inspection or testing device into which the component can be inserted for examination. In another embodiment, the device may be connected to the specimen to perform the examination. The component may be removed from the motor vehicle for this purpose or examined in situ in the motor vehicle. The invention will now be described in more detail with reference to the accompanying drawings, in which:

[0026] Figure 1 a system;

[0027] Figure 2 shows a flow diagram of a method;

[0028] Figure 3 Representations of precision curves in the determination of motor vehicles;

[0029] Figure 4 shows exemplary density distributions with respect to different parameters; and

[0030] Figure 5 illustrates certain correlations of different parameters.

[0031] Figure 1 shows a group 100 of motor vehicles 105, with at least one example of a predetermined component 110 installed in each motor vehicle 105. One of the motor vehicles 105 is shown enlarged as an example. On this motor vehicle, the component 110 is connected to a control unit 115, for example, via a data bus on board the motor vehicle 105. The control unit 115 is preferably connected to an interface 120, which can be either wired or wireless.

[0032] The control unit 115 is configured to determine one or preferably several functional conditions of the component 110 on board the motor vehicle 105 and to provide them externally. The functional conditions can in particular comprise a configuration, an operating parameter, or an accompanying influence. Preferably, a type or scope of collected data can be controlled. More preferably, the control unit 115 specifically collects data related to the operation, function, or presence of the component 110. The component 110 can comprise any desired component of the motor vehicle 105, for example, a structural, mechanical, electrical, or processing component 110. Optionally, a component 110 that is not primarily technically dependent can also be included, for example, a design element.Component 110 may be defective, which may pose a danger to motor vehicle 105 or a person or object in the vicinity of motor vehicle 105. For example, component 110 may comprise a safety component, such as a seat belt buckle or an airbag, and the safety of a person on board motor vehicle 105 may not be guaranteed in the event of an accident if a defect occurs in the installed component 110.

[0033] It is known that such a defect can at least occur in component 110, especially if it has already been observed in a number of specimens. However, it may initially be unknown which influences could promote or prevent the occurrence or a negative consequence of the defect.

[0034] A control device 125 comprises a processing device 130, which is preferably connected to a data memory 135. A first interface 140 is configured to determine the technical functional conditions of an example of component 110 on an associated motor vehicle. A second interface 145 is configured to record an examination result of this example of component 110.

[0035] By way of example, the first interface 140 leads to a communication device 150 for communication with one of the motor vehicles 105. In another embodiment, the first interface 140 can also be configured, for example, to read a data carrier on which the corresponding data is stored. Also by way of example, the second interface 145 is connected to a device 155 with which an example of the component 110 can be examined. In particular, the device 155 can be used to determine whether the component has failed ("sick") or is functional ("healthy"). The device 155 can be configured to examine the example of the component 110 in isolation or in an installed state in the motor vehicle 105.

[0036] An optional third interface 160 is configured to output a determination result. The third interface 160 can be connected to an output device, such as a screen.

[0037] It is proposed that the processing device 130 comprise a model using machine learning methods that allows a failure risk of an instance of a component 110 to be determined based on prevailing or previously encountered technical functional conditions. The model can be trained based on technical functional conditions and associated test results of instances of the component 110 on various motor vehicles 105.

[0038] From the group of 100 motor vehicles 105 and the applicable technical functional conditions, a motor vehicle 105 can then be selected for which the failure risk of the associated component 110 is as high as possible. The selected motor vehicle 105 can be taken to a service appointment, during which the component 110 can be repaired or replaced. Furthermore, the vehicle can be examined, and its condition can be determined. The collected information can be used to further train the model.

[0039] The collected information can also be stored in data storage 135. After a certain runtime, the model can be retrained based on the stored data. In doing so, a selection of parameters of the technical functional conditions underlying the training can be changed.

[0040] In particular, the increased dataset can be used to better identify a parameter relevant to default risk, and the model can be configured to consider this parameter. A different parameter with low relevance to default risk can, however, no longer be included in the model, or its weight in making a decision using the model can be reduced.

[0041] Figure 2 shows a flowchart of a method 200 for prioritizing repairs of motor vehicles 105 in a group 100. The method 200 can be applied to improve a recall of the group 100 of motor vehicles 105. The group 100 can be narrowed down by determining motor vehicles 105 for which the risk of failure is negligible. Among the motor vehicles 105 that are serviced, those with a high risk of failure can be treated promptly or with priority, and those with a low risk of failure can be treated with a lower priority or less urgency.

[0042] It should be noted that the steps of the method shown can also be carried out in a different order than the one given, as a person skilled in the art will immediately recognize.

[0043] In a step 205, it can initially be determined which motor vehicles 105 belong to group 100. This determination can be made, for example, based on production documents. In particular, all motor vehicles 105 in which an example of a predetermined component 110 is installed can belong to group 100. In a step 210, the technical functional conditions under which the examples of component 110 perform their function in the individual motor vehicles can be determined. This step, in particular, can also be performed at a later time.

[0044] In a step 215, a sample of motor vehicles 105 can be formed that are scheduled for a service appointment. The sample can be determined based on the specific functional conditions. An attempt can also be made to determine the most representative sample possible by initially ignoring the functional conditions. Examples of component 110 of the motor vehicles 105 in the sample can be examined for their failure or risk of failure.

[0045] A correlation between functional conditions and a failure risk can be determined in a step 220. A machine learning method can be applied for this purpose. The method can initially be trained based on data determined for the motor vehicles 105 in the sample.

[0046] Based on the model, a motor vehicle 105 can be determined in a step 225 in which functional conditions prevail that favor or increase the risk of failure. In particular, a motor vehicle 105 can be selected for which the risk of failure is maximized among the motor vehicles 105 of group 100 not yet considered.

[0047] Such a motor vehicle 105 can be called in for a service appointment in a step 230, and the associated example of component 110 can be examined. The result of the examination, together with associated technical functional conditions, can be used to further train the model, as symbolized by the dashed line. The method 200 can then return to step 225, in which another motor vehicle 105 can be selected. Of course, several motor vehicles 105 can also be processed concurrently in steps 225 and 230. For this purpose, different locations can be provided, each of which can be used for a service. Information obtained from the service is preferably collected and processed centrally. The determination of further motor vehicles 105 is further preferably carried out on the basis of the collected data.

[0048] In a step 235, a motor vehicle 105 whose component 110 has already been examined can be repaired. This step can run concurrently with the other steps of method 200.

[0049] Figure 3 shows representations of precision curves in the determination of motor vehicles 105. From left to right, a first precision curve 305, a second precision curve 310, a third precision curve 315 and a fourth precision curve 320 are shown.

[0050] Each precision curve 305-320 represents, in the horizontal direction, a progress in the processing of the motor vehicles 105 of group 100. From left to right, the proportion increases from 0 (no motor vehicle 105 examined or treated) to 1 (all motor vehicles 105 examined or treated). In the vertical direction, a precision is plotted that indicates how many of the respectively examined motor vehicles 105 were problematic with respect to the assigned instance of component 110. A value of 1 reflects that every examined motor vehicle 105 was problematic, and a value of 0 that none of the examined motor vehicles 105 were problematic.

[0051] An ideal model would identify only motor vehicles 105 that are problematic and then determine that none of the motor vehicles 105 not yet recalled are problematic. An operating point of such an ideal model could be located at the coordinates (1 , 1 ). In practice, a compromise must be found between a (pessimistic) determination that yields many correct positive results, but also many incorrect positive results, and an (optimistic) determination that yields many correct negative results, but also many incorrect negative results. A relationship between correct and incorrect positive and negative determinations can be specified in a truth matrix (also: contingency table, confusion matrix).

[0052] The first precision curve 305 shows that, starting from a sample of feature carriers, a fairly good prediction can already be made, but the precision decreases with increasing recall.

[0053] If the model is retrained before training is completed using previously collected data, parameters that correlate more strongly with the risk of failure can be found more effectively. This allows the model to make a better determination. The subsequent precision curves 310 to 320 show models retrained in this way, with the training using increasingly more collected data from left to right.

[0054] Figure 4 shows exemplary qualitative density distributions 405-420 for various exemplary parameters. A value range for each parameter is plotted horizontally. A density is plotted vertically, indicating how frequently an observed or unobserved symptom occurs at the respective parameter value. A solid line represents a density distribution for an observed symptom, and a dashed line represents a density distribution for an unobserved symptom.

[0055] A parameter where the positive and negative density distributions are very similar—that is, where there is a large overlap of the areas under the density curves—is of little use for predicting the symptom based on a value for the associated parameter. Such a parameter forms the basis of the third density distribution, 415. This parameter can only provide an indication of the absence of the symptom if the respective value lies within one of two relatively narrow value ranges and is very high. However, a useful indication of the presence of the symptom cannot be determined based on this parameter.

[0056] A parameter with a strong difference between the positive and negative density distributions, however, is well suited to determining whether a value observed on a vehicle indicates a symptom or not. Density distributions 405, 410, or 420 can be helpful here.

[0057] If the symptom cannot be directly observed, an assessment can be made based on a parameter value as to whether the symptom is likely to be present or not. Ideal density distributions with respect to a parameter are as disjoint as possible.

[0058] Figure 5 shows correlations between various parameters and a symptom. A first exemplary parameter set 505 comprises four parameters, each of which is weighted and graphically represented as a bar graph.

[0059] A second exemplary parameter set 510 comprises eleven parameters in a similar representation. The more discriminatory a parameter is (see Figure 4 above), the greater its weight in the determination can be. The aim is to consider only a small number of parameters, each with the highest possible discriminatory power. Considering a larger number of parameters may be acceptable, even if the parameters have only moderate discriminatory power. Parameters with low discriminatory power, however, should be avoided. Reference symbols

[0060] 100 Group

[0061] 105 Motor vehicle

[0062] 110 Component

[0063] 115 Control unit

[0064] 120 interface

[0065] 125 Control device

[0066] 130 processing facility

[0067] 135 data storage

[0068] 140 first interface

[0069] 145 second interface

[0070] 150 communication device

[0071] 155 Device for examination

[0072] 160 third interface

[0073] 200 procedures

[0074] Record 205 vehicles

[0075] 210 Determine functional conditions

[0076] 215 Examine sample

[0077] 220 Determine correlation

[0078] Select 225 copies

[0079] Examine 230 specimens

[0080] Repair 235 copies

[0081] 305 first precision curve

[0082] 310 second precision curve

[0083] 315 third precision curve

[0084] 320 fourth precision curve

[0085] 405 first density distribution

[0086] 410 second density distribution

[0087] 415 third density distribution fourth density distribution first parameter set second parameter set

Claims

Claims 1 . A method (200) for prioritizing repairs of motor vehicles (105), wherein each of the motor vehicles (105) has an example of a predetermined component (110) at risk of failure installed; wherein the method (200) comprises the following steps: Determining (210) technical functional conditions of an example of the component (110) on an associated motor vehicle (105); Examine (230) this specimen; Determining (230) a risk of failure of this specimen under the specified operating conditions; Determining (220) technical functional conditions under which the failure risk of the component (110) is increased, on the basis of previously determined functional conditions and failure risks of examined specimens; and Selections (230) of another motor vehicle (105) with technical functional conditions that indicate the highest possible risk of failure for the component (110).

2. Method (200) according to claim 1, wherein a relationship between the failure risk and the technical functional conditions is determined (220) by means of a machine learning method.

3. The method (200) according to claim 1 or 2, wherein the functional conditions comprise a configuration of the motor vehicle (105).

4. Method (200) according to one of the preceding claims, wherein the functional conditions comprise a usage parameter of the motor vehicle (105).

5. Method (200) according to one of the preceding claims, wherein the further motor vehicle (105) is determined with respect to a subset of considered functional conditions.

6. The method (200) according to claim 5, wherein the further motor vehicle (105) is determined (220) with respect to predetermined functional conditions according to the Random Forest algorithm.

7. The method (200) according to claim 5 or 6, wherein a functional condition is determined (210) in which a density distribution of failed specimens is as different as possible from a density distribution of functional specimens.

8. The method (200) according to any one of the preceding claims, wherein the risk of failure relates to the probability of failure.

9. The method (200) according to any one of the preceding claims, wherein the failure risk relates to the brevity of an expected remaining functional life of the component (110).

10. The method (200) according to any one of the preceding claims, wherein the risk of failure relates to a possible severity of consequences of a failure.

11. Method (200) according to one of the preceding claims, wherein functional conditions of the specimens in the motor vehicles (105) are collected (210) during operation.

12. The method (200) according to any one of the preceding claims, wherein the component (110) is repaired (235) in the specified order on the motor vehicles (105).

13. Method (200) according to one of the preceding claims, wherein examples of the components (110) on a predetermined proportion of the motor vehicles (105) are initially examined (210).

14. A control device (125) for prioritizing repairs of motor vehicles (105), wherein an example of a predetermined component (110) is installed in each of the motor vehicles (105); wherein the device comprises the following elements: a first interface (140) for detecting technical functional conditions of an example of the component (110) on an associated motor vehicle (105); a second interface (145) for detecting an examination result of this example; wherein the examination result includes a failure risk of this example under the determined functional conditions; a processing device (130) for determining technical functional conditions under which the failure risk of the component (110) is increased, based on previously determined functional conditions and failure risks of examined examples;and an output device for providing an indication of another motor vehicle (105) with technical functional conditions that indicate the highest possible risk of failure for the component (110); 15. The control device (125) of claim 14, further comprising a device (155) for examining an instance of the component (110).

Images