Method and device for reliability control of a vehicle component failure prediction model
The method improves vehicle component failure prediction reliability by assessing model variability, ensuring consistent predictions and reducing breakdown risks through proactive maintenance.
Patent Information
- Application Number
- FR2023005320
- Authority / Receiving Office
- FR · FR
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-30
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing vehicle maintenance methods, whether preventive or corrective, do not guarantee the reliability of failure predictions, leading to unnecessary costs and vehicle immobilization due to unforeseen breakdowns.
A method for controlling a vehicle component failure prediction model involves learning from data, collecting validation data over time, determining a sequence of prediction results, and assessing the reliability of the model based on the variability of these results to ensure consistent and reliable predictions.
This approach enhances the reliability of failure predictions, allowing for proactive maintenance and reducing the risk of unexpected breakdowns by validating the model's consistency before deployment, thereby saving time and resources.
Smart Images

Figure 00000000_0000_ABST
Abstract
Description
Title of the invention: Method and device for controlling the reliability of a failure prediction model of a vehicle component Technical field
[0001] The invention relates to methods and devices for controlling a failure prediction model of one or more components of a vehicle, in particular for controlling and verifying the reliability of the prediction model. The invention also relates to a method and a device for predictive maintenance of a vehicle. Technological background
[0002] Contemporary vehicles are made up of a large number of components or organs, each susceptible to breakdown or defect.
[0003] When the driver of a vehicle is faced with a defect or breakdown of one or more components of his vehicle, the latter must then contact a specialized center to determine whether the breakdown or defect observed requires a corrective maintenance operation to correct the breakdown or defect after the latter has appeared. Such corrective maintenance is most often accompanied by an intervention or repair in a garage for example. Such corrective maintenance is synonymous with costs generated by the corrective maintenance and inconvenience caused by the immobilization of the vehicle for a period sometimes lasting up to several days or even several weeks.
[0004] To limit the risks of breakdown or defect and prevent associated problems, preventive maintenance operations of vehicles are planned. However, these maintenance operations do not guarantee that a vehicle component will encounter a defect or suffer a breakdown after the fact. Summary of the invention
[0005] An object of the present invention is to solve at least one of the problems of the technological background described above.
[0006] Another object of the present invention is to improve the prediction of breakdowns or defects of organ(s) or component(s) of a vehicle.
[0007] Another object of the present invention is to improve the reliability of the prediction of failures of component(s) of a vehicle.
[0008] According to a first aspect, the invention relates to a method for controlling a failure prediction model of a vehicle component, the prediction model having been learned in a so-called learning phase from learning data obtained from a set of first vehicles, the method comprising the following steps: - collection of validation data representative of indicators associated with at least one second vehicle and measured by the at least one second vehicle for each day of a plurality of successive days in time; - determining a sequence of component failure prediction results by predicting, for each day of the plurality of days, a probability of component failure by feeding the failure prediction model with validation data associated with each day and with a set of days of the plurality of days temporally preceding this day; - determination of information representative of the reliability of the failure prediction model based on information representative of variability of the prediction results in the sequence.
[0009] Such a method makes it possible to test the reliability of a failure prediction model by using vehicle data collected over several days and the failure prediction obtained for each of these days. The reliability is tested by analyzing the variability of the results to detect any inconsistencies in the results.
[0010] Validating the reliability of a prediction model makes it possible to deploy a more secure model for use in production of this model. This makes it possible, for example, to avoid limiting or avoiding validating the model in production (saving time and money) and to improve vehicle safety with more reliable prediction models in production.
[0011] According to a variant, the prediction model is determined to be unreliable when the prediction results in the sequence are according to a non-monotonic curve and the prediction model is determined to be reliable when the prediction results in the sequence are according to an increasing monotonic curve.
[0012] According to another variant, each failure prediction result of the sequence belongs to a set of failure prediction results comprising: - a first result corresponding to a failure prediction for the component; and - a second result corresponding to a prediction of an absence of failure for the component.
[0013] According to a further variant, the prediction model is determined to be reliable if, when the prediction result for a determined day of the plurality corresponds to the first result, the prediction result for each day of the plurality following the determined day corresponds to the first result.
[0014] According to another variant, the prediction model is determined to be unreliable if, when the prediction result for a determined day of the plurality corresponds to the first result, the prediction result for at least one day of the plurality following the determined day corresponds to the second result.
[0015] According to another variant, the at least one second vehicle does not belong to the first set of vehicles.
[0016] According to an additional variant, the method further comprises a step of transmitting parameters representative of the prediction model to at least a third vehicle when the prediction model is determined to be reliable.
[0017] According to another variant, the component corresponding to a battery, the indicators belong to a set of indicators comprising: - indicator representing a fault code; - indicator representing a battery discharge; - indicator representing a journey time; - indicator representing a journey distance.
[0018] According to a second aspect, the invention relates to a device for controlling a failure prediction model of a component of a vehicle, the device comprising a memory associated with a processor configured for implementing the steps of the method according to the first aspect of the invention.
[0019] According to a third aspect, the invention relates to a computer program which comprises instructions adapted for executing the steps of the method according to the first aspect of the invention, in particular when the computer program is executed by at least one processor.
[0020] Such a computer program may use any programming language, and be in the form of source code, object code, or intermediate code between source code and object code, such as in a partially compiled form, or in any other desirable form.
[0021] According to a fourth aspect, the invention relates to a computer-readable recording medium on which is recorded a computer program comprising instructions for executing the steps of the method according to the first aspect of the invention.
[0022] On the one hand, the recording medium may be any entity or device capable of storing the program. For example, the medium may comprise a storage means, such as a ROM memory, a CD-ROM or a microelectronic circuit type ROM memory, or a magnetic recording means or a hard disk.
[0023] Furthermore, this recording medium may also be a transmissible medium such as an electrical or optical signal, such a signal being able to be conveyed via an electrical or optical cable, by conventional or hertzian radio or by self-directed laser beam or by other means. The computer program according to the invention may in particular be downloaded from an Internet-type network.
[0024] Alternatively, the recording medium may be an integrated circuit in which the computer program is incorporated, the integrated circuit being adapted to carry out or to be used in carrying out the method in question. Brief description of the figures
[0025] Other characteristics and advantages of the invention will emerge from the description of the non-limiting embodiments of the invention below, with reference to the appended figures 1 to 4, in which:
[0026] [Fig-1] schematically illustrates a communication environment of a vehicle, according to a particular embodiment of the present invention;
[0027] [Fig.2] schematically illustrates a process for verifying the reliability of a failure prediction model of a vehicle component of [Fig.l], according to a particular and non-limiting exemplary embodiment of the present invention.
[0028] [Fig.3] schematically illustrates a device configured to control a failure prediction model of a component of a vehicle of [Fig.l], according to a particular exemplary embodiment of the present invention;
[0029] [Fig.4] illustrates a flowchart of the different steps of a method for controlling a failure prediction model of a component of a vehicle of [Fig.l], according to a particular exemplary embodiment of the present invention. Description of the embodiments
[0030] A method and a device for controlling a model for predicting failure of a component of a vehicle will now be described in the following with joint reference to FIGS. 1 to 4. The same elements are identified with the same reference signs throughout the description which follows.
[0031] According to a particular and non-limiting example of embodiment of the invention, the control of a failure prediction model of one (or more) component(s) of a vehicle, for example a battery of a vehicle, comprises the collection, for example from a set of second vehicles or from a data storage device, of validation data representative of indicators associated with the second vehicles and measured by the second vehicles for each day of a time period comprising a plurality of temporally successive days. A sequence of failure prediction results of the component is determined or obtained by predicting, for each day of the plurality of days, a probability of failure of the component by feeding the failure prediction model with the validation data associated with each day and with a set of days of the plurality of days temporally preceding this day.For the prediction of a probability of failure of the component on a day called “day D” of the plurality of days, the data obtained for this day D and for the previous days D are aggregated to feed the prediction model and determine the probability of failure (with as a result a probability . illustrating the risk that a failure occurs or simply as a result a first value indicating a risk of failure and a second value indicating an absence of risk of failure). Information representative of the reliability of the failure prediction model is determined based on information representative of a variability of the prediction results in the sequence of the prediction results obtained for each day of the plurality of days.
[0032] High variability in the prediction results is for example representative of a lack of reliability of the prediction model while low or zero variability in the sequence of results is representative of reliability of the prediction model.
[0033] A breakdown corresponds for example to a failure of a component or organ of the vehicle, that is to say to a malfunction resulting in a stoppage of the operation of the component or organ in question. A breakdown also corresponds to a degraded operation of the component or organ in question compared to normal operation, for example an operation or only part of the services that the component or organ provides in normal operation is provided.
[0034] The prediction model is for example learned according to any machine learning method known to those skilled in the art using a set of learning data obtained for example from a set of first vehicles.
[0035] The determination of the reliability of the learned prediction model is for example implemented by using a validation data set different from the training data set, the validation data set being for example obtained from a set of second vehicles different from the first vehicles.
[0036] The validation of the learned prediction model, for example the determination of its reliability, is for example implemented after the model learning phase. According to a variant, this validation is implemented during the learning phase, for example at the end of the learning phase.
[0037] The implementation of a machine learning method makes it possible to estimate more reliably and in real time the risk of failure associated with a component of a vehicle based on indicators received from this vehicle and representative of the actual use made of this vehicle and / or the environment of this vehicle, on the basis of a model learned from the same indicators of a set of other vehicles which have not experienced failures of this component, or at least which have experienced a low number of fault codes, and of a set of other vehicles which have experienced this failure, or at least which have experienced a higher number of fault codes for example.
[0038] Predicting the risk of failure makes it possible to anticipate the problem and potentially to correct it before it occurs, improving breakdown prevention.
[0039] Validating a prediction model makes it possible to ensure, for example before deploying the model in production, that the predictions obtained are reliable.
[0040] [Fig. 1] schematically illustrates a communication environment 1 for vehicles, according to a particular and non-limiting exemplary embodiment of the present invention.
[0041] The vehicles of the environment comprise for example a set of first vehicles 11, a set of second vehicles 12 and one or more third vehicles 13.
[0042] Each vehicle 11, 12, 13 corresponds for example to a vehicle with a thermal engine, with electric motor(s) or even a hybrid vehicle with a thermal engine and one or more electric motors. Each vehicle thus corresponds for example to a land vehicle, for example a car, a truck, a bus, a motorcycle.
[0043] Each vehicle 11, 12, 13 carries a communication system configured to communicate with one or more remote devices 101 via a wireless communication network infrastructure. The remote device 101 advantageously corresponds to a device configured to process data, for example data stored in the memory of the remote device 101 and / or data received from the vehicle. The remote device 101 corresponds for example to a server of the “cloud” 100.
[0044] The communication system of each vehicle 11, 12, 13 comprises for example one or more communication antennas connected to a telematic control unit, called TCU (from the English "Telematic Control Unit"), itself connected to one or more computers of the on-board system of the vehicle. The antenna(s), the TCU unit and the computer(s) form for example a multiplexed architecture for the realization of different services useful for the proper functioning of the vehicle and for assisting the driver and / or the passengers of the vehicle in the control of the vehicle and / or for establishing a diagnosis on the functioning of one or more components of the vehicle.The computer(s) and the TCU communicate and exchange data with each other via one or more computer buses, for example a communication bus of the CAN (Controller Area Network), CAN FD (Controller Area Network Flexible Data-Rate), FlexRay (according to the ISO 17458 standard) or Ethernet (according to the ISO / IEC 802-3 standard) data bus type.
[0045] The mobile communication infrastructure enabling wireless data communication between each vehicle 11, 12, 13 and the remote device 101 comprises, for example, one or more communication devices 110 of the relay antenna type. (cellular network) or roadside unit, called UBR. In a communication mode using such a network architecture, the data is for example transmitted by the vehicle 11, 12, 13 to the remote device 101 of the “cloud” 100 via a relay antenna 110 (the antenna 110 being for example connected to the “cloud” 100 via a wired link).
[0046] The wireless communication system allowing the exchange of data between the vehicles 11, 12, 13 and the remote device 101 corresponds for example to: - a vehicle-to-infrastructure V2I communication system (from English “vehicle-to-infrastructure”), for example based on the 3GPP LTE-V or IEEE 802.1 Ip standards of ITS G5; or - a cellular network type communication system, for example an LTE (Long-Term Evolution) or LTE-Advanced (LTE-Advanced) type network, also called LTE 3G, 4G or 5G; or - a Wifi type communication system according to IEEE 802.11, for example according to IEEE 802.lin or IEEE 802.1 lac.
[0047] The first vehicles 11 form for example a set of vehicles for obtaining training data used for training one or more prediction models, for example one or more vehicle component failure prediction models.
[0048] The second vehicles 12 form, for example, a set of vehicles for obtaining validation or test data for the learned prediction model(s).
[0049] The third vehicle(s) 13 correspond, for example, to vehicles on which the prediction model(s) learned and validated via the validation data set(s) are deployed or implemented.
[0050] A process for controlling a vehicle prediction model is advantageously implemented by a remote device 101 connected in communication with the first and second vehicles 11, 12 via a wireless connection or link.
[0051] The process of controlling a prediction model, for example a component failure prediction model, presented below is described with regard to a particular and non-limiting example relating to the prediction of a failure of a vehicle battery, for example a starter battery.
[0052] Of course, such a failure prediction process applies to any component or organ of a vehicle, for example a traction battery, an AD AS system (from the English “Advanced Driver-Assistance System” or in French “Advanced Driving Assistance System”), an alternator, a starter, one or more computers, etc.
[0053] The process of controlling the prediction model is implemented during a validation phase, such a validation phase being implemented from a temporal point of view after a so-called learning or training phase of the prediction model. prediction and before a so-called production, prediction or inference phase using the learned and validated prediction model.
[0054] According to an alternative embodiment, the validation phase corresponds to a sub-phase of the learning phase and is implemented during the learning phase.
[0055] The learning and production phases are known to those skilled in the art. An example of carrying out such phases is described below for illustrative purposes.
[0056] Learning phase
[0057] The learning implemented in the first phase advantageously corresponds to supervised learning from a set of data associated with a set or group of second vehicles comprising for example a few tens, a few hundreds, thousands or tens / hundreds of thousands of first vehicles 11. According to an alternative embodiment, the learning implemented in the learning phase corresponds to unsupervised learning from the set of data associated with the set of first vehicles 11.
[0058] According to a particular example, the first vehicles included in the set of first vehicles 11 all have the same configuration or a similar configuration, that is to say that they all correspond to the same type of vehicle (for example the same series of a particular model of vehicle) with identical or similar components or members (same version for example).
[0059] In a first operation of the learning phase, the remote device 101 collects a set of indicators emitted by each of the first vehicles 11. The indicators are for example transmitted by each first vehicle via the wireless network infrastructure described above to the remote device 101. The indicators correspond for example to some or all of the vehicle signals emitted by the sensors on board each first vehicle 11. These indicators are for example transmitted at each occurrence, or continuously at a determined frequency, of an event detected by a sensor or a computer on board each first vehicle 11.The indicators reported to the remote device 101 correspond for example to the indicators of a determined list stored in the memory of a computer of the on-board system of each first vehicle 11 which makes a selection of the indicators to be transmitted from among all the indicators that each first vehicle 11 detects or measures.
[0060] According to an alternative embodiment, the indicators transmitted to the remote device 101 by each first vehicle 11 correspond to all the events detected by each first vehicle 11, the remote device 101 subsequently making a selection from all the indicators received to use only those defined to be used in learning the starter battery failure prediction model.
[0061] The collection of indicators is for example implemented over a determined time interval corresponding for example to several weeks, several months or several years of use of the first vehicles, for example in real time.
[0062] These indicators received from the set of first vehicles during the determined time interval are advantageously stored in the memory of the remote device 101 or of a remote device dedicated to the collection and storage of these indicators and connected to the remote device 101, for example via a wired link.
[0063] According to an alternative embodiment, information representative of the time instant at which each indicator was detected, measured or transmitted is associated with said each indicator. In other words, each indicator is time-stamped.
[0064] The indicators belong for example to the following set of indicators, the selected indicators corresponding to all or part of the following indicators according to any possible combination: - indicator representing a fault code, also called DTC code (from the English “Diagnostic Trouble Code” or in French “Code faute de diagnostic”); and / or - indicator representing a battery discharge; and / or - indicator representing the duration of each journey made by the vehicle; and / or - indicator representing the distance of each journey made by the vehicle.
[0065] The above list is provided for illustration purposes and is not exhaustive; other indicators may be taken into account.
[0066] According to an alternative embodiment, one or more of the indicators listed above are used or processed to deduce one or more additional indicators representative of the state of the battery. This or these indicators are used, for example, to supply specific rules or formulas making it possible to determine the additional indicator(s).
[0067] In a second operation of the learning phase, the remote device 101 identifies the first vehicles of the set which have had one or more battery breakdowns / faults. These first vehicles form a first part of the set of first vehicles 11. These first vehicles are for example identified from data received from one or more servers 102 associated with after-sales services provided by one or more companies, one or more companies providing vehicle repairs (car garage for example) which allow the updating of a database relating to vehicles which have broken down or are experiencing a fault in one or more of their components / organs. The server(s) 102 transmit for example a set of data identifying each first vehicle 11 for which a battery breakdown (or fault) has been noted or reported.The data set advantageously includes for each first vehicle and for each identified failure on the date on which the failure was . identified or reported. The data set further advantageously comprises an identifier of each first vehicle 11 making it possible to determine one or more of the following information: - the type of vehicle; and / or - the model of the vehicle; and / or - the make of the vehicle; and / or - the hardware and software configuration of the vehicle.
[0068] The identifier of each first vehicle 11 corresponds for example to the so-called VIN number (from the English “Vehicle Identification Number” or in French “Numéro d'identification de véhicule”).
[0069] In a third operation of the learning phase, the remote device 101 identifies the first vehicles of the set which have not experienced a starter battery failure, i.e. the first vehicles for which no battery failure has been reported or identified by the after-sales services. These first vehicles form a second part of the set of first vehicles 11, the first part and the second part forming the set of first vehicles in a complementary manner.
[0070] The first vehicles of the second part are for example identified from data received from one or more servers 102 of a car manufacturer, the first vehicles of the second part being selected in that their configuration corresponds to that of the first vehicles of the first part.
[0071] The first vehicles of the second part are for example identified from their VIN number.
[0072] Thus and advantageously, the first part of the set of first vehicles and the second part of the set of first vehicles form parts which are coherent with each other in that the first vehicles which compose them have an identical or similar configuration (for example same make and same model, or even same make, same model and same level of equipment).
[0073] In a fourth operation of the learning phase, the indicators are grouped into two sets of indicators. A first set of indicators groups the indicators collected in the first operation and corresponding to the first vehicles of the first part of the set of first vehicles (i.e., the first vehicles having had a battery failure). A second set of indicators groups the indicators collected in the first operation and corresponding to the first vehicles of the second part of the set of first vehicles (i.e., the second vehicles having not experienced a battery failure).
[0074] In a fifth operation of the learning phase, the model(s) of battery failure prediction is / are generated or trained by classifying the indicators of the first set and the second set of second indicators.
[0075] When the prediction phase is implemented by a third vehicle 13, the parameters defining each prediction model obtained in the fifth operation are transmitted by the remote device 101 to the third vehicle 13 which stores them in memory associated with a computer of the on-board system for the implementation by this computer of the prediction phase.
[0076] The transmission of these parameters is only implemented once the validation phase is complete, when the learned prediction model is determined to be reliable at the end of this validation phase.
[0077] Such a learning phase thus makes it possible to cross-reference the data from the first vehicles (i.e. the indicators) with data obtained from after-sales services (data on the breakdowns / faults encountered by these first vehicles).
[0078] Validation phase
[0079] The validation phase is described according to an example of embodiment illustrated by the [Fig.2],
[0080] In a first operation of the validation phase, the device in charge of the validation phase, i.e. the remote device 101, collects a set of validation data representative of indicators obtained from the set of second vehicles 12, i.e. indicators measured or detected by the second vehicles 12. The indicators of the validation data correspond to the indicators used and described in the learning phase.
[0081] The validation data indicators are for example transmitted by the second vehicles 12 to the remote device 101 via the wireless network infrastructure over time when an event associated with such an indicator occurs and is detected or measured by a second vehicle 12.
[0082] The collection of indicators during the validation phase is similar to the collection of indicators in the learning phase and the elements relating to the indicators described with regard to the learning phase are taken up in the validation phase.
[0083] The collection of indicators is for example implemented over a determined time interval corresponding for example to several weeks, several months or several years of use of the second vehicles, for example in real time.
[0084] These indicators received from the set of second vehicles 12 during the determined time interval are advantageously stored in the memory of the remote device 101 or of a remote device dedicated to the collection and storage of these indicators and connected to the remote device 101, for example via a wired link.
[0085] According to an alternative embodiment, information representative of the time instant at which each indicator was detected, measured or transmitted is associated with said each indicator. In other words, each indicator is time-stamped.
[0086] The indicators are for example collected and / or grouped by day as illustrated in [Fig.2], the days being indicated by “D1” for the first day of the time interval, “D2” for the second day and so on up to for example “D151” according to the particular example of [Fig.2].
[0087] The days D1 to D151 form a temporal sequence of days succeeding one another, D1 being earlier than D2 from a temporal point of view, D3 being earlier than D2 and so on up to D151 which is earlier than D150. The days D1 to D151 correspond for example to consecutive days. According to another example, each day is for example representative of a week, D1 being associated with week 1, D2 with week 2 and so on up to D151 which is associated with week 151, the weeks being consecutive from a temporal point of view.
[0088] In a second operation of the validation phase, a sequence of component failure prediction results is determined by predicting, for each day of the plurality of days forming the sequence of days D1 to D151, a probability of component failure from the learned prediction model.
[0089] For this purpose, for each prediction associated with each day D1 to D151, the failure prediction model is fed with validation data 21 to 26 formed from the data of the day in question and the days temporally preceding this day in question.
[0090] This makes it possible to obtain a failure prediction result for each day of the plurality of days forming the time interval or sequence of days D1 to D151.
[0091] For example, for day Jl, the prediction is based on the validation data 21 obtained for this first day JL
[0092] For day D2, the prediction is based on the validation data obtained for day D2 and for the day preceding D2, i.e. DL The validation data 22 feeding the prediction model to obtain the result of the prediction for D2 correspond to the data obtained by aggregating the data obtained for day D1 and for day D2.
[0093] For day D3, the prediction is based on the validation data obtained for day D3 and for the days preceding D3, i.e. D1 and D2. The validation data 23 feeding the prediction model to obtain the result of the prediction for D3 correspond to the data obtained by aggregating the data obtained for day D1, for day D2 and for day D3.
[0094] The same reasoning is applied for each day D1 to D151. For day D151, the prediction is based on the validation data obtained for day D151 and for all the days preceding D151, i.e. each day from D1 to D150. The validation data 26 feeding the prediction model to obtain the result of the prediction for D151 correspond to the data obtained by aggregating the data obtained for days D1 to D151.
[0095] The result of the prediction for each day is indicated in column “P” 201. According to the example of [Fig.l], the result of the prediction for day D1 is equal to 0 (for example corresponding to an absence of predicted breakdown), the result of the prediction for day D2 is equal to 0, the result of the prediction for day D3 is equal to 0, the result of the prediction for day D149 is equal to 1 (for example corresponding to a predicted breakdown), the result of the prediction for day D150 is equal to 1 and the result of the prediction for day D151 is equal to 1.
[0096] According to another example, the prediction results correspond to a probability that a failure occurs, such a probability being for example indicated by a set of determined values between 0 and 1 (for example 0, 0.2, 0.4, 0.6, 0.8 and 1), the number of values corresponding to the number of classes associated with the prediction model (also called classification model).
[0097] The sequence of prediction results P 201 is for example represented in the form of a curve 202 marking the evolution or variation of the prediction results P (on the ordinate of the diagram) as a function of the days J (on the abscissa of the diagram).
[0098] In a third operation of the validation phase, the prediction results P 201 and in particular the variability of the prediction results P 201 over the time interval are analyzed to determine information representative of the reliability of the prediction model used to obtain the prediction results P 201.
[0099] The information representing reliability includes, for example, the following values: - 0 for a prediction model determined to be unreliable; and - 1 for a prediction model determined to be reliable.
[0100] For example, a prediction model is determined or judged to be unreliable when the prediction results in the sequence P 201 are according to a non-monotonic curve, for example when the results vary in a sawtooth pattern. According to another example, a prediction model is determined or judged to be unreliable when a prediction result equal to 1 for a day of the sequence is followed by a prediction result equal to 0 for a later day of the sequence. According to another example, a prediction model is determined or judged to be unreliable when a prediction result equal to 0 for a day of the sequence is followed by a prediction result equal to 1 for a preceding day of the sequence. Such examples are representative of variability in the prediction results 201, showing that the prediction model is unreliable.
[0101] In certain cases, when a breakdown is predicted for a given day D, for example day 149, the prediction results for the days after day 149 from a temporal point of view (days D150 and D151) must also be 1 for consistency. Indeed, if a failure is predicted on day D, the failure must be predicted for the following days, in particular because the component ages.
[0102] In other cases, when a breakdown is predicted for a given day D, for example day 149, the subsequent prediction results may return to the value 0 after remaining at 1 for one or more days (for example for days 160, 165, 200 and following), for example if the component has been changed or repaired or if the driver has changed behavior with his vehicle.
[0103] A prediction model is determined or judged to be reliable when the prediction results in the sequence P 201 are according to a monotonically increasing curve, such as the curve 202 illustrated in [Fig.2], the variability of the results being considered zero or low. Thus, the prediction model is determined to be reliable if, when the prediction result for a determined day of the sequence of days forming the time interval is equal to 1 (failure prediction), the prediction result for each day of the sequence following the determined day (or after the determined day) is also equal to 1. Otherwise, the prediction model is determined to be unreliable.
[0104] When the prediction model learned in the learning phase is validated during the validation phase, i.e. it has been determined to be reliable, then the prediction model can be deployed in production, for example on the third vehicle(s) 13.
[0105] For this purpose, the parameters of the prediction model are for example transmitted to every third vehicle 13.
[0106] Prediction phase
[0107] The prediction, production or inference phase is implemented by collecting production data corresponding to indicators measured or detected by the third vehicle 13 considered. These indicators are advantageously identical or similar to the indicators described with regard to the learning and validation phases.
[0108] The data thus obtained are provided as input to the prediction model to predict the probability that a breakdown will occur. The result is for example 0 or 1, or according to another example, a probability value is obtained and is compared to one or more threshold values to determine whether the risk of breakdown is proven or high. When the probability of breakdown reaches the threshold value (for example 60%, 70%, 80%, 90%), an alert is for example generated and for example transmitted: - to the third vehicle 13 to alert the driver of the risk of breakdown and / or inform him of the procedure to follow to prevent the breakdown; and / or - to a server associated with an after-sales service responsible for contacting the third vehicle 13 (or its owner) to assist it in the measures to be taken to prevent the breakdown or defect; and / or - to the driver or owner of the third vehicle 13 to alert him or her of the risk of breakdown and / or inform him or her of the procedure to follow to prevent the breakdown; this data is, for example, transmitted to a display device integrated into the third vehicle 13 and / or to a remote device such as a smartphone.
[0109] [Fig. 3] schematically illustrates a device 3 configured to control a prediction model learned during a learning phase, for example a vehicle component failure prediction model, according to a particular and non-limiting exemplary embodiment of the present invention. The device 3 corresponds for example to a computing or data processing device, for example the remote device 101. According to another example, the device 3 corresponds to a device embedded in the vehicle 10, for example a computer.
[0110] The device 3 is for example configured for the implementation of the operations described with regard to figures 1 and 2 and / or the steps of the method described with regard to [Fig.4]. Examples of such a device 3 include, but are not limited to, on-board electronic equipment such as an on-board computer of a vehicle, an electronic calculator such as an ECU (“Electronic Control Unit”), a telematic control unit, called TCU (from the English “Telematic Control Unit”), a smartphone, a tablet, a laptop, a server, or a combination of several of the devices listed above. The elements of the device 3, individually or in combination, can be integrated into a single integrated circuit, into several integrated circuits, and / or into discrete components.The device 3 can be produced in the form of electronic circuits or software (or computer) modules or even a combination of electronic circuits and software modules.
[0111] The device 3 comprises one (or more) processor(s) 30 configured to execute instructions for carrying out the steps of the method and / or for executing the instructions of the software(s) embedded in the device 3. The processor 30 may include integrated memory, an input / output interface, and various circuits known to those skilled in the art. The device 3 further comprises at least one memory 31 corresponding for example to a volatile and / or non-volatile memory and / or comprises a memory storage device which may comprise volatile and / or non-volatile memory, such as EEPROM, ROM, PROM, RAM, DRAM, SRAM, flash, magnetic or optical disk.
[0112] The computer code of the embedded software(s) comprising the instructions to be loaded and executed by the processor is for example stored in the memory 31.
[0113] According to various particular embodiments, the device 3 is coupled in communication with other similar devices or systems and / or with devices communication, for example a TCU (Telematic Control Unit), for example via a communication bus or through dedicated input / output ports.
[0114] According to a particular and non-limiting embodiment, the device 3 comprises a block 32 of interface elements for communicating with external devices, for example a remote server or the “cloud”, a TCU unit. The interface elements of the block 32 comprise one or more of the following interfaces: - RF radio frequency interface, for example Bluetooth® or Wi-Fi® type, LTE (Long-Term Evolution), LTE-Advanced; - USB interface (from the English “Universal Serial Bus” or “Universal Serial Bus” in French); - HDMI interface (from the English “High Definition Multimedia Interface” or “High Definition Multimedia Interface” in French); - LIN interface (from the English “Local Interconnect Network”).
[0115] According to another particular embodiment, the device 3 comprises a communication interface 33 which makes it possible to establish communication with other devices (such as other computers of the embedded system or other servers) via a communication channel 330. The communication interface 33 corresponds for example to a transmitter configured to transmit and receive information and / or data via the communication channel 330. The communication interface 33 corresponds for example to a wired network of the CAN (from the English “Controller Area Network” or in French “Réseau de contrôles”) type, CAN FD (from the English “Controller Area Network Flexible Data-Rate” or in French “Réseau de contrôles à débit de données flexible”), FlexRay (standardized by the ISO 17458 standard) or Ethernet (standardized by the ISO / IEC 802-3 standard).
[0116] According to an additional particular embodiment, the device 3 can provide output signals to one or more external devices, such as a display screen 340, one or more speakers 350 and / or other peripherals 360 (projection system) via respective adapted output interfaces 34, 35, 36. According to a variant, one or other of the external devices is integrated into the device 3. The display screen corresponds for example to a screen, touch-sensitive or not.
[0117] [Fig. 4] illustrates a flowchart of the different steps of a method for controlling a vehicle component failure prediction model, according to a particular and non-limiting exemplary embodiment of the present invention. The method is for example implemented by the remote device 101 or by a system comprising one or more second vehicles 12 connected in wireless communication to the device distant 101.
[0118] In a first step 41, validation data representative of indicators associated with at least one second vehicle and measured by the at least one second vehicle are collected for each day of a plurality of temporally successive days.
[0119] In a second step 42, a sequence of component failure prediction results is determined by predicting, for each day of the plurality of days, a probability of component failure by feeding the failure prediction model with validation data associated with each day and with a set of days of the plurality of days temporally preceding this day.
[0120] In a third step 43, information representative of the reliability of the failure prediction model is determined as a function of information representative of a variation or variability of the prediction results in the sequence.
[0121] According to an alternative embodiment, the variants and examples of the operations described in relation to figures 1 and 2 apply to the steps of the method of [Fig.4].
[0122] Of course, the invention is not limited to the embodiments described above but extends to a method of alerting of a breakdown before the breakdown occurs and the device configured for the implementation of such a method.
[0123] The invention also relates to a system comprising one or more vehicles and a remote device, for example the remote device 101, the vehicle(s) being connected in wireless communication with the remote device 101.
Claims
Demands
1. A method for controlling a failure prediction model for a vehicle component, said prediction model having been learned in a so-called learning phase from training data obtained from a set of first vehicles (11), said method comprising the following steps: - collecting (41) validation data representative of indicators associated with at least a second vehicle and measured by said at least a second vehicle for each day of a temporally successive plurality of days; - determining (42) a sequence of failure prediction results (201) for said component by predicting, for each day of said plurality of days, a probability of failure of said component by feeding said failure prediction model the validation data (21 to 26) associated with said each day and with a set of days of said plurality of days temporally preceding said each day;- determination (43) of information representative of the reliability of said failure prediction model as a function of information representative of the variability of prediction results in said sequence (201).;
2. A method according to claim 1, wherein said prediction model is determined to be unreliable when the prediction results in said sequence (201) are along a non-monotonic curve and said prediction model is determined to be reliable when the prediction results in said sequence (201) are along a monotonically increasing curve (202).
3. A method according to claim 1 or 2, wherein each failure prediction result of said sequence belongs to a set of failure prediction results comprising: - a first result corresponding to a failure prediction for said component; and - a second result corresponding to a prediction of no failure for said component.
4. A method according to claim 3, wherein said prediction model is determined to be reliable if, when the prediction result for a given day of said plurality corresponds to said first result, the prediction result for each day of said plurality following that determined day corresponds to that first result.
5. A method according to claim 3 or 4, wherein said prediction model is determined to be unreliable if, when the prediction result for a specified day of said plurality corresponds to said first result, the prediction result for at least one day of said plurality following said specified day corresponds to said second result.
6. A method according to any one of claims 1 to 5, wherein said at least one second vehicle does not belong to the set of first vehicles.
7. A method according to any one of claims 1 to 6, further comprising a step of transmitting representative parameters of said prediction model to at least a third vehicle when said prediction model is determined to be reliable.
8. A method according to any one of claims 1 to 7, wherein, said component corresponding to a battery, said indicators belong to a set of indicators comprising: - an indicator representing a fault code; - an indicator representing a battery discharge; - an indicator representing a journey time; - an indicator representing a journey distance.
9. Computer program comprising instructions for carrying out the method according to any one of the preceding claims, when such instructions are executed by a processor.
10. Device (3) for controlling a failure prediction model of a vehicle component, said device (3) comprising a memory (31) associated with at least one processor (30) configured for carrying out the steps of the method according to any one of claims 1 to 8.
Citation Information
Patent Citations
Method and device for predicting the failure of at least one component of a vehicle
FR3120724A1
Decomposed machine learning model evaluation system
US20210287115A1
