Method and system for estimating the total mass of a vehicle

Abstract

The present invention relates to a system (12) for estimating a total mass (Ms) of a vehicle (1) when stationary, comprising: - receiving means (13, 14) for receiving at least one datum (h) from a suspension travel sensor, and a datum (α) from an inclination sensor, - estimation means for estimating the total mass (Ms) from the data (h, α), the estimation means comprising a neural network (15) trained to estimate the total mass (Ms) of the vehicle (1, 2) from the data received by the receiving means (13, 14). The invention also relates to a vehicle equipped with the estimation system, and to the corresponding method for estimating the mass of a vehicle when stationary.

Description

[0001] DESCRIPTION

[0002] Method and system for estimating the total mass of a vehicle

[0003] The present invention relates to the field of the automotive industry, and more specifically concerns a method for estimating the total mass of a vehicle, carried out in the vehicle when it is stationary.

[0004] The total mass of a vehicle, including its sprung and unsprung mass, is used by the Advanced Driver Assistance Systems (ADAS) implemented in the vehicle. Unsprung mass refers to the mass of the vehicle's components that follow the undulations of the road surface, such as the wheels, rims, tires, and shock absorbers. Sprung mass, on the other hand, refers to the mass of the vehicle's components located above the shock absorbers. Unsprung mass remains relatively constant throughout the vehicle's operation, while sprung mass depends on factors such as the vehicle's load and the number of occupants, and can therefore vary significantly.

[0005] Currently, advanced driver assistance systems typically use a fixed mass, entered into the vehicle's main computer, for their calculations when the vehicle starts, regardless of the load. The vehicle's chassis actuators, particularly the braking and stability control systems, therefore do not know the exact total mass of the vehicle. Consequently, the commands these systems apply to the chassis and wheels deviate from the intended braking or stability target, potentially impacting stability and passenger comfort.

[0006] An accurate estimation of the vehicle's total mass is therefore required to design optimal vehicle control, implemented by advanced driver assistance systems as well as by an engine control system capable of managing the vehicle's deceleration and acceleration. Such an estimation is also useful for predictive maintenance of vehicle components, particularly suspensions, whose wear due to damage caused by poor road conditions is a function of the vehicle's unsprung mass. Similarly, road infrastructure managers need to know the masses of vehicles using these infrastructures to model road wear, and on-board road safety systems also need to know the vehicle's total mass to detect the risk of rollover when the vehicle is overloaded.

[0007] Finally, knowing the exact total mass of a vehicle from the moment it starts allows for improved planning of vehicle journeys, for example when it is making deliveries and / or when the vehicle is electric, its range being then greatly impacted by its mass.

[0008] This estimation of the total mass of the vehicle must therefore be able to be carried out even before the vehicle starts.

[0009] Furthermore, the European Commission requires vehicle manufacturers to determine the total mass of the vehicle in real time.

[0010] Solutions exist for determining the mass of a vehicle while it is in motion. These solutions use measured or estimated data concerning the vehicle's longitudinal dynamics, including its longitudinal acceleration, longitudinal velocity, and rolling resistance coefficients. It is also possible to use extended Kalman filters to obtain mass estimates from parameters such as the vehicle's longitudinal and / or lateral accelerations, yaw rate, steering wheel angle, etc.

[0011] However, when the vehicle is stationary, it is not possible to use the equations of longitudinal vehicle dynamics. One solution is to use, for example, direct force sensors, mounted on the wheels, struts, or even in the wheel bearings, to approximate the mass, but these sensors are very expensive. It is also possible to use position information provided by suspension travel sensors used by vehicle lighting systems, or COSLAD (Control in Site of Discharge Lamps) systems, to adapt the direction of generated light beams according to changes in the vehicle's attitude. But the information provided by these sensors is difficult to use because the suspension travel they measure is not linear.Indeed, particularly on commercial vehicles, suspension friction, especially in leaf spring suspensions, is very significant and creates hysteresis. Furthermore, the suspensions have bump stops with non-linear behavior and very short linear travel. Finally, the characteristics of the suspension travel vary depending on the wear of the suspension.

[0012] These non-linearities do not allow for an accurate estimation of the total mass of the vehicle at rest, so that to estimate the total mass of the vehicle at rest, the use of very expensive force sensors is preferred to the use of suspension stroke sensors, although the latter are already present in the vehicle and therefore do not represent an additional cost related to this estimation.

[0013] The present invention aims to remedy at least in part the aforementioned drawbacks by providing a method and system for estimating the total mass of a stationary vehicle, which is inexpensive while being more accurate than using only the vehicle's suspension travel sensors.

[0014] To this end, the invention proposes a method for estimating the total mass of a stationary vehicle, comprising the following steps:

[0015] - Receiving at least one data point from a suspension travel sensor, and one data point from a tilt sensor,

[0016] - estimation of the total mass of the vehicle from the data received, the estimation process being characterized in that the estimation step involves an injection of the received data into a neural network trained to estimate the total mass of the vehicle from the injected data.

[0017] The suspension travel sensor, for example, is a vehicle suspension travel sensor, possibly the same one used by a vehicle lighting system, that provides data from which it is possible to calculate the variation in the vehicle's height relative to the ground. The tilt sensor calculates the angle of inclination of the vehicle's floor relative to a horizontal plane.

[0018] Thanks to the invention, the problems of non-linearity in suspension travel sensors are overcome by using a trained neural network. This network, trained on a vehicle of the same type as the one in which the estimation method according to the invention is implemented, is capable of providing an estimate of the vehicle's mass with acceptable accuracy, using only data from one or more suspension travel sensors and one or more tilt sensors as inputs. Furthermore, since the estimation method according to the invention relies on sensors already present in the vehicle, its implementation requires only software modifications to the vehicle and is therefore inexpensive.

[0019] However, the neural network may have other inputs, receiving data from sensors that vary according to the mass of the vehicle and can be used when the vehicle is stopped, in order to refine the estimate provided by the neural network.

[0020] Thus, in one embodiment of the invention, the reception step further includes, for example, receiving data from other sensors representative of the vehicle's total mass (any type of sensor that varies according to the vehicle's mass and provides usable data when the vehicle is stationary) and / or vehicle-specific data and / or driver-specific data. This additional data is used in addition to the data from the suspension travel and tilt sensors. The vehicle-specific and / or driver-specific data corresponds to contextual information such as the driver's weight, the vehicle's age, or its mileage, so as to further refine the estimate provided by the neural network.In particular, when the age of the vehicle or its mileage are used as input to the neural network and have therefore been taken into account during its training, the neural network is able to capture the effect of the aging of the vehicle's components on the estimated mass.

[0021] In one embodiment of the invention, the neural network is time-series based. Such a neural network, called a "Time Series Deep Forecasting Model" in English, uses knowledge of its inputs over a time window to provide a low-noise output value corresponding to the total mass of the vehicle. The neural network can use this knowledge to also provide an output with an accuracy estimate of the total mass of the vehicle, for example, by using the variances and covariances of its input data.

[0022] Alternatively, the estimation method according to the invention includes a temporal filtering step of the value read from the output of the neural network, corresponding to the total mass of the vehicle, so that this value is less dependent on measurement noise from the sensors. This temporal filtering uses, for example, a Kalman filter or a recursive least squares method. Furthermore, the estimation method may include a step for estimating the accuracy of the total mass estimate performed by the method, using, for example, the covariance matrix of the Kalman filter errors.

[0023] The accuracy of the estimation is used, for example, by the vehicle to inhibit the process of estimating the total mass of the vehicle when stationary, when this accuracy is characteristic of a defective sensor.

[0024] On the other hand, in one embodiment of the invention, the estimation method according to the invention includes a neural network training step, in which at least one training iteration includes

[0025] - the application of at least one data point from the suspension travel sensor, and one data point from the tilt sensor, as input to the neural network,

[0026] - updating the neural network by modifying the synaptic weights of neurons in the neural network, so as to obtain at the output of the neural network a value corresponding to a total mass determined during a previous step of determining the total mass of the vehicle in motion.

[0027] This learning iteration is, for example, performed on-board in the vehicle, when the vehicle goes into sleep mode after a driving phase. The vehicle goes into sleep mode when it stops immediately following the driving phase. The estimation of the vehicle's total mass at rest, according to the estimation method of the invention, then takes place, for example, when the vehicle subsequently wakes up. This embodiment of the invention allows the neural network used to estimate the vehicle's total mass at rest to be updated, for example, after each driving phase throughout the vehicle's lifespan, or only after a certain number of driving phases. Thus, the parameters of the neural network used to estimate the vehicle's total mass evolve during the vehicle's lifespan to account for its aging.

[0028] Determining the total mass of a moving vehicle is generally very precise; for example, it allows the total mass to be known to within 5%, which is less true for estimating the vehicle's static mass when stationary. Preferably, when the accuracy of the dynamic mass is assessed by the vehicle and falls below a predetermined threshold, the total mass determined while moving is not subsequently used to update the neural network used for estimating the vehicle's total mass when stationary.

[0029] The initial training of the neural network is performed once using a reference vehicle and can be implemented in all vehicles of the same model / type / range; it does not need to be done for each individual vehicle. This initial training enables the implementation of the method for estimating the total mass of the vehicle at rest according to the invention. During the vehicle's lifecycle, the training can be updated with new data, either by downloading it or by updating the model using data collected from the vehicle during a driving cycle.

[0030] The neural network is trained using the same types of data as used during the inference phase of the neural network, which is implemented by the vehicle to estimate its total mass when stationary.

[0031] The initial learning and update phases of the neural network model involve iterations in which the synaptic weights of the neurons are modified to best estimate the mass of the stationary vehicle, using as ground truth measured values ​​of the vehicle mass, or vehicle mass values ​​obtained while driving by a vehicle mass estimation process using vehicle dynamics data.

[0032] In one embodiment of the invention, the step of determining the total mass of the vehicle in motion uses, for example, at least one variable of the vehicle's longitudinal dynamics. This dynamic variable is, for example, the vehicle's longitudinal acceleration.

[0033] In a variant of this embodiment, the step of determining the total mass of the vehicle in motion uses, for example, a kinematic model of the vehicle and a model of its longitudinal dynamics, estimates the slope of the road and the rolling resistance coefficients of the tires, to estimate the total mass of the vehicle in real time.

[0034] In another variation of this embodiment, the step for determining the total mass of the moving vehicle uses a trained neuromimetic network that receives as input a value representing the torque applied to the vehicle's wheels and a value representing the vehicle's longitudinal acceleration, and outputs a value representing the vehicle's longitudinal force. The total mass is then determined by dividing the longitudinal force by the longitudinal acceleration. The trained neuromimetic network used in the step for determining the total mass of the moving vehicle is, of course, separate from the neural network used for estimating the total mass of the stationary vehicle. This separate neuromimetic network can receive additional input data, such as the vehicle's lateral acceleration, wind speed, or the gradient of the road on which the vehicle is traveling.

[0035] Furthermore, the step of determining the total mass of the vehicle preferably includes a step of smoothing the total mass determined over time during the driving phase, in order to eliminate noise related to sudden variations in speed or acceleration of the vehicle. Optionally, the determination of the total mass of the vehicle while driving is inhibited when unusual driving conditions are encountered, for example, when the road gradient is very steep, or when the vehicle brakes suddenly or when it makes a very sharp turn.

[0036] Furthermore, in another embodiment of the invention, the learning iteration is performed remotely from the vehicle. The learning step involves, firstly, transmitting data from the suspension travel and tilt sensors, as well as the vehicle's total mass (determined while driving), to a remote server configured to update the neural network. Secondly, the vehicle receives the updated neural network from the remote server. This remote learning of the neural network allows it to be enriched with data from different vehicles of the same type and with the same aging characteristics. It also saves computing resources within the vehicle, thus preventing it from being delayed in its shutdown.

[0037] Optionally, the learning phase includes, for each of the vehicle's missions (within a plurality of vehicle missions), a storage phase on a remote server or within the vehicle itself. This storage phase involves a predetermined total mass during the mission, along with data from the suspension travel sensor and the tilt sensor at the beginning or end of the mission (i.e., when the vehicle is stationary). The storage phase is followed by learning iterations, each corresponding to a separate update of the neural network using the data stored during one of the missions. A vehicle mission is a vehicle lifecycle comprising only a wake-up phase, a driving phase, and a sleep phase.

[0038] In other words, with this option, the neural network is only updated after a predetermined number of driving phases. This limits the number of software updates required for the vehicle, as well as the data exchange between the vehicle and the remote server in the event that the neural network update is deployed.

[0039] The invention also relates to a system for estimating the total mass of a stationary vehicle, comprising:

[0040] - means for receiving at least one data point from a suspension travel sensor, and one data point from a tilt sensor,

[0041] - means for estimating the total mass of the vehicle from said data, the estimation system being characterized in that the estimation means include a neural network trained to estimate the total mass of the vehicle from the data received by the receiving means.

[0042] The invention also relates to a vehicle comprising a system for estimating the total mass of the vehicle when stationary, according to the invention. The vehicle is, of course, a motor vehicle.

[0043] The system for estimating the total mass of a stationary vehicle according to the invention and the vehicle according to the invention have advantages similar to those of the method for estimating the total mass of a stationary vehicle according to the invention.

[0044] Other features and advantages of the invention will become apparent from the following description on the one hand, and from several illustrative and non-limiting examples given by reference to the attached schematic drawings on the other hand, in which: [fig 1] represents a first system for estimating the total mass of a stationary vehicle according to the invention, in one embodiment of the invention,

[0045] [fig 2] represents steps of a first method for estimating the total mass of a stationary vehicle according to the invention, implemented by the first estimation system of figure 1, in the embodiment of the invention of figure 1,

[0046] [Fig. 3] represents a second system for estimating the total mass of a stationary vehicle according to the invention, in a variant of the embodiment of Figure 1, and

[0047] [fig 4] represents steps of a second method for estimating the total mass of a stationary vehicle according to the invention, implemented by the second estimation system of figure 3, in the embodiment variant of figure 3.

[0048] According to one embodiment of the invention illustrated in figures 1 and 2, a stationary vehicle 1 according to the invention comprises a set 10 of sensors and / or estimators whose outputs are sent on a computer bus of the vehicle 1, for example the CAN bus (for the English "Controller Area Network").

[0049] This set 10 includes, in particular:

[0050] - at least one suspension stroke sensor, capable of providing at least one variation h, for example in centimeters, of a distance between the floor of vehicle 1 and the ground, relative to a default distance corresponding for example to a distance between the floor of vehicle 1 and the ground when vehicle 1 is not loaded and empty of any passengers; several suspension stroke sensors are preferably installed on separate axles of the vehicle;

[0051] - a tilt sensor, capable of providing an angle of inclination a in degrees, of the floor of the vehicle 1 with respect to a horizontal plane.

[0052] In order to allow estimation of the vehicle's dynamic mass in driving conditions, assembly 10 may also include:

[0053] - a sensor for longitudinal acceleration x of vehicle 1, in meters per second squared; and

[0054] - a sensor of a longitudinal torque Cx applied to the wheels of vehicle 2, in Newton-meters.

[0055] Of course, other embodiments are possible in which other sensors are used to estimate the dynamic mass of the vehicle, or in which this dynamic mass is not estimated.

[0056] Of course, the sensors in set 10 can contain a software component and therefore implement estimators. Furthermore, as an alternative, they provide representative values ​​of a variation in height, an angle of inclination, a longitudinal acceleration, or a longitudinal torque, expressed in units different from those mentioned above.

[0057] Vehicle 1 also includes a first estimation system 12 for the total mass Ms of vehicle 1 when it is stationary, particularly during a wake-up phase. This first estimation system 12, which is, for example, a main computer of vehicle 1, is configured to implement a first estimation method 100, shown in Figure 2, for the total mass Ms of vehicle 1 when it is stationary; this first estimation method 100 is therefore primarily implemented in software.

[0058] The first estimation system 12 includes, in particular:

[0059] - means for receiving 13 of the values ​​supplied on the CAN bus by the suspension stroke sensor(s) and by the tilt sensor, and

[0060] - means of receiving 14 the values ​​provided on the CAN bus by the longitudinal acceleration sensor and by the longitudinal torque sensor applied to the wheels of vehicle 1.

[0061] The first estimation system 12 also includes at least one processor, a random access memory and a read-only memory in which is stored a computer program whose instructions, when executed on the processor, enable the implementation of the first estimation process 100. The read-only memory stores in particular the data of a neural network 15 trained to estimate the total mass of the vehicle 1 at rest, from the values ​​received by the receiving means 13, and possibly other types of data, such as for example the mileage of the vehicle 1.

[0062] In this embodiment of the invention, neural network 15 is a time-series neural network, for example, an LSTM (Long Short-Term Memory) neural network. This type of neural network has the advantage of capturing the effect of aging of vehicle components on the estimated total mass. Of course, other types of neural networks can be used instead of an LSTM.

[0063] The neural network 15 is therefore configured, for example, to receive as input values ​​of height variations h, an angle of inclination a and the mileage of the vehicle, and to deliver as output a total mass Ms of the vehicle in kilograms.

[0064] The first estimation system 12 also includes in memory a module 16 for determining the total mass Md of the vehicle 1 while in motion. For example, in this embodiment of the invention, without being exhaustive of the means implemented to estimate the total mass Md of the vehicle while in motion, the determination module 16 may be a software module implementing a neuromimetic network (i.e., a neural network) trained to receive as input a longitudinal acceleration value a x and a longitudinal torque value Cx provided by the receiving means 14, and to deliver at output a longitudinal force value Fx. This neuromimetic network comprises, for example, four neuronal layers.

[0065] Module 16 for determination can also implement a divider capable of dividing this value of longitudinal force Fx by the value of longitudinal acceleration a xto provide a raw value for the total mass, which can then be filtered by a Kalman filter to deliver the total mass Md of vehicle 1 in motion. It should be noted that other embodiments are possible in which the determination module 16 uses, for example, the equations of the vehicle's kinematics and in particular its longitudinal dynamics, or Kalman filters using, for example, information from the lateral dynamics such as wheel steering angles, lateral acceleration and yaw rate, and not a trained neural network, to estimate the mass of the vehicle in motion.

[0066] The neural network 15 and the neuromimetic network are trained during a supervised learning phase, for example carried out on a test bench and using as ground truths real masses, i.e. accurately measured, of a test vehicle of the same type as vehicle 1. Alternatively, the supervised learning phase uses a campaign of driving with one or more test vehicles of the same type as vehicle 1.

[0067] Once the learning phase is complete, the neural network 15 is trained, and can be deployed across all vehicles in the range of the test vehicle.

[0068] The weights of the neural network 15 are, however, likely to be modified again after this learning phase, by means of a module 17 for updating the neural network 15, implemented in the first estimation system 12. The module 17 for updating the neural network 15 uses for this purpose as ground truth the total mass Md determined by the module 16 for determining the total mass of the vehicle 1 in motion, during a motion phase immediately preceding a sleep phase of the vehicle 1 during which this update is carried out.

[0069] A more precise operation of the first estimation system 12 is now described in relation to Figure 2, which presents the steps of the first estimation process 100 of the total mass Ms of the vehicle when it is at rest.

[0070] The first estimation method 100 includes a preliminary learning phase 102, which comprises, for example, a supervised learning phase performed using data obtained from a test vehicle, for example, on a test bench. During this supervised learning phase, the test vehicle is loaded with different load levels, allowing a range of total vehicle mass values ​​to be covered, and the test bench's inclination is modified for each mass value to cover a range of possible inclinations. For each load level and inclination, the data from the tilt and suspension travel sensors corresponding to that load level are fed into the neural network 15, and the synaptic weights of the neurons in the neural network 15 are modified so that the output of the neural network 5 is equal to the actual mass of the test vehicle with that load level.

[0071] Following this supervised learning phase, vehicle 1 includes a first phase <p1 de réveil, le véhicule 1 étant stationné à l’arrêt.

[0072] During this first phase <p1 de réveil, le premier procédé d’estimation 100 comporte une étape 104 de réception des données issues du ou des capteurs de course de suspension, et du capteur d’inclinaison, et éventuellement de réception d’une ou plusieurs données contextuelles, par exemple le kilométrage du véhicule, et / ou de données délivrées par tout type de capteur délivrant des données exploitables lorsque le véhicule est à l’arrêt et qui varient en fonction de la masse du véhicule. Lors de cette étape 104 de réception, les données reçues sont stockées dans un historique d’utilisation du réseau de neurones 15.The reception step 104 is followed by a step 106 of estimation of the total mass Ms of the vehicle 1 at rest, during which the data received in the previous step 104, and where appropriate contextual data or data from other sensors, are injected into the neural network 15, the total mass Ms of the vehicle 1 being read at the output of the neural network 15 then recorded and made available to applications requiring to know the total mass Ms of the vehicle at rest.

[0073] Following the awakening of vehicle 1, it enters a second phase <p2 de roulage, lors de laquelle le premier procédé d’estimation 100 comporte une étape 108 de détermination d’une masse totale Md du véhicule 1 en utilisant le module 16 de détermination de la masse totale Md du véhicule en roulage. Cette masse totale Md déterminée en roulage est enregistrée dans la mémoire vive du calculateur implémentant le premier système d’estimation 12, au regard des informations du ou des capteurs de course de suspension, et du capteur d’inclinaison acquises lors de la phase de réveil.

[0074] Following the operation of vehicle 1, it enters a third phase <p3 d’endormissement. Avant d’endormir les calculateurs du véhicule 1 , le réseau de neurones 15 est mis à jour par le module 17 de mise à jour du réseau de neurones 15, pour perfectionner son entrainement et tenir compte du vieillissement du véhicule. Le premier procédé de détermination 100 comporte pour cela une itération 110 d’apprentissage, mise en œuvre par le module 17 de mise à jour, préalablement à une nouvelle utilisation du premier procédé d’estimation 100 de la masse totale du véhicule 1 lors du prochain arrêt de celui-ci.

[0075] During this iteration 110, the data recorded in the usage history of the neural network 15, i.e. the values ​​of height variation h of the suspensions, the value of inclination a of the vehicle (and where applicable the mileage of the vehicle) stored in the history, are injected into the neural network 15 to update its learning, with regard to the values ​​of total mass Md determined during driving in the previous step 108. Thus, the mass values ​​of the vehicle estimated during the driving phase serve as ground truth to enrich the learning of the neural network used to estimate the static mass of the vehicle.

[0076] Alternatively, the update of the neural network 15 is not performed each time the vehicle is put to sleep, but only after recording at least a predetermined number of estimates of the total mass Ms of vehicle 1 at rest, followed by an equal number of determinations of a total mass Md while in motion. The update of the neural network for estimating the total mass Ms of the vehicle at rest is then performed using all the recorded data. An alternative embodiment of the invention, and more particularly a variant of the total mass Ms estimation system at rest, implemented in another vehicle 2, is now described in relation to Figure 3. In this variant, the main computer of vehicle 2 is equipped with a system for estimating the total mass Ms of the vehicle at rest, hereinafter referred to as the second estimation system 22.The second estimation system 22 of the total mass Ms of vehicle 2 at rest is implemented in a main computer of vehicle 2. This includes the same set 10 of sensors and / or estimators as vehicle 1 in the main embodiment of the invention.

[0077] The main computer of vehicle 2 also includes receiving means 13 and 14, identical to those of the first estimation system 12, of values ​​provided by the suspension travel sensor(s) of vehicle 2, the tilt sensor of vehicle 2, and the sensors enabling estimation of the dynamic mass of the vehicle, such as the longitudinal acceleration sensor and the longitudinal torque sensor applied to the wheels of vehicle 2.

[0078] The second estimation system 22 also implements a neural network 15 identical to that of the first estimation system 12, to estimate the total mass Ms of the vehicle 2 at rest, and a module 16 for determining the total mass Md of the vehicle 2 in motion, identical to that of the first estimation system 12.

[0079] The second estimation system 22 differs from the first estimation system 12 only in the way the neural network 15 is updated following one or more driving phases of the vehicle 2. Indeed, the second estimation system 22 does not perform this update itself, but delegates it to a remote server 26 of the vehicle 2 manufacturer. For this purpose, it includes a module 24 for transmitting data to the remote server 26, and for receiving data from the remote server 26. The communication between the transmission and reception module 24 and the remote server 26 is a radio communication, using, for example, a mobile telecommunications cellular network.

[0080] The precise operation of the second estimation system 22 is described in relation to Figure 4, which presents the steps of a second estimation process 200 of the total mass Ms of the vehicle when it is at rest, implemented by the second estimation system 22.

[0081] The second estimation method 200 includes a prior learning phase 202 which includes, for example, a supervised learning phase, identically to the prior learning phase 102 of the first estimation method 100.

[0082] Following this supervised learning phase, vehicle 2 includes a first phase <p1 de réveil, le véhicule 2 étant à l’arrêt.

[0083] During this first phase <p1 de réveil, le deuxième procédé d’estimation 200 comporte une étape 204 de réception des données issues du ou des capteurs de course de suspension, et du capteur d’inclinaison, et éventuellement de réception de données contextuelles telles que le kilométrage du véhicule, ces données reçues étant ensuite stockées dans un historique d’utilisation du réseau de neurones 15. L’étape 204 de réception est suivie d’une étape 206 d’estimation de la masse totale Ms du véhicule 2 à l’arrêt, lors de laquelle les données reçues à l’étape 204 précédente sont injectées dans le réseau de neurones 15, la masse totale Ms du véhicule 2 étant lue en sortie du réseau de neurones 15 puis enregistrée et rendue disponible aux applications nécessitant de connaître la masse totale Ms du véhicule à l’arrêt.

[0084] Following the awakening of vehicle 2, it enters a second phase <p2 de roulage, lors de laquelle le deuxième procédé d’estimation 200 comporte une étape 208 de détermination d’une masse totale Md du véhicule 2 en utilisant le module 16 de détermination de la masse totale Md du véhicule en roulage, qui par nature est plus précise que la masse totale Ms estimée à l’arrêt du véhicule. Cette masse totale Md déterminée en roulage est enregistrée dans la mémoire du calculateur implémentant le deuxième système d’estimation 22. N’importe quelle autre méthode d’estimation de la masse du véhicule en roulage peut être utilisée à la place du réseau neuromimétique.

[0085] Following the driving of vehicle 2, it enters a third phase <p3 d’endormissement. Avant d’endormir les calculateurs du véhicule 2, le deuxième système d’estimation 22 vérifie que l’estimation 206 de la masse totale Ms du véhicule 2 à l’arrêt correspond à une i-ème estimation de quantième strictement inférieur à un nombre prédéterminé K d’estimations de la masse totale du véhicule 2 à l’arrêt.

[0086] If this is the case (branch N), the second estimation system 22 stores 210 in its memory, before the vehicle 2 completely falls asleep, the total mass Md determined during the second driving phase <p2 au regard des informations délivrées par les capteurs du véhicule à l’arrêt contenues dans l’historique d’utilisation du réseau de neurones 15, afin de permettre une mise à jour ultérieure de celui-ci.

[0087] If on the contrary the estimate 206 corresponds to an i-th estimate with i equal to the predetermined number K (branch Y), then the second estimation system 22 transmits 212 to the remote server 26, via the transmission and reception module 24 of the second estimation system 22, before the complete sleep of the vehicle 2, the set of data of the usage history of the neural network 15, and the corresponding total masses Md determined while driving.

[0088] The remote server 26 then performs the update of the neural network 15, of which it has the latest version in memory, using the data transmitted to it, in a manner similar to the update performed by the module 17 of the update of the first estimation system 12.

[0089] Then, during the next wake-up of vehicle 2, the remote server 26 transmits the updated neural network to the transmission and reception module 24 of the second estimation system 22, which receives it and replaces the version of the neural network 15 in its memory with the neural network 15 received from the remote server 26. This update uses, for example, FOTA (Firmware Over The Air) technology. The second estimation system 22 then clears the usage history of the neural network 15 to avoid transmitting data already transmitted from this history during a subsequent transmission to the remote server 26. Of course, the invention is not limited to the examples just described, and many modifications can be made to these examples without departing from the scope of the invention.In particular, the characteristics of the different variants of embodiment of the invention envisaged in this application can be combined to carry out the invention, insofar as these variants are not incompatible with each other.

Claims

DEMANDS 1- Method for estimating (100, 200) the total mass (Ms) of a stationary vehicle (1, 2), comprising the following steps: - reception (104, 204) of at least one data point (h) from a suspension travel sensor, and one data point (a) from a tilt sensor, - estimation (106, 206) of the total mass (Ms) of the vehicle from the data (h, a) received, the estimation process (100, 200) being characterized in that the estimation step (106, 206) includes an injection of the received data into a neural network (15) trained to estimate the total mass (Ms) of the vehicle (1, 2) from the injected data. 2- Estimation method (100, 200) according to claim 1, wherein the neural network (15) is based on time series. 3- Estimation method (100, 200) according to claim 1 or 2, wherein the reception step (104, 204) further comprises the reception of data from other sensors representative of the total mass of the vehicle and / or vehicle-specific data (1, 2) and / or driver-specific data. 4- Estimation method (100, 200) according to any one of claims 1 to 3, comprising a neural network training step (15), in which at least one training iteration (110) comprises: - the application of at least one data point (h) from the suspension travel sensor, and one data point (a) from the tilt sensor, as input to the neural network (15), - the updating of the neural network (15) by modifying the synaptic weights of neurons in the neural network (15), so as to obtain at the output of the neural network (15) a value corresponding to a total mass (Md) determined during a previous step of determination (108, 208) of the total mass (Md) of the vehicle (1, 2) in motion. 5- Estimation method (100, 200) according to claim 4, wherein the step of determining (108, 208) the total mass (Md) of the vehicle (1, 2) rolling uses at least one variable of longitudinal dynamics (a x , Cx) of the vehicle (1 , 2). 6- Estimation method (100, 200) according to claim 4 or 5, wherein the step of determining (108, 208) the total mass (Md) of the vehicle (1, 2) in motion uses a trained neuromimetic network receiving as input a value representative of a torque (Cx) applied to the wheels of the vehicle (1, 2) and a value representative of a longitudinal acceleration (a x ) of the vehicle (1, 2), and providing as output a value of a longitudinal force (F x ) of the vehicle (1, 2). 7- Estimation method (100) according to any one of claims 4 to 6, wherein the learning iteration (110) is carried out in an on-board manner in the vehicle (1, 2), when the vehicle (1, 2) is put to sleep after the driving phase (cp2). 8- Estimation method (200) according to any one of claims 4 to 6, wherein the learning iteration (110) is carried out in a manner off the vehicle (1, 2), the learning step comprising on the one hand a transmission (212) of the data (h, a) from the suspension travel and tilt sensors, and of the total mass (Md) of the vehicle (1, 2) determined in motion, to a remote server (26) configured to update the neural network (15), and on the other hand a reception (214), by the vehicle (1, 2) of the updated neural network (15), sent by the remote server (26). 9- Estimation system (12, 22) for the total mass (Ms) of a stationary vehicle (1, 2), comprising: - means for receiving (13, 14) at least one data point (h) from a suspension travel sensor, and one data point (a) from a tilt sensor, - means for estimating the total mass (Ms) of the vehicle from said data (h, a), the estimation system (12, 22) being characterized in that the estimation means comprise a neural network (15) trained to estimate the total mass (Ms) of the vehicle (1, 2) from the data received by the receiving means (13, 14). 10- Vehicle (1, 2), characterized in that it comprises a system for estimating (12, 22) its total mass (Ms) at rest according to claim 9.

Images