Method for characterizing the friction coefficient that can be mobilized by the wheels of a motor vehicle using the information regarding its longitudinal dynamics

Abstract

The present invention relates to a method for characterizing the friction coefficient that can be mobilized by one or more wheels of a motor vehicle drivetrain. It comprises: - a step (501) of estimating a longitudinal stiffness Cx of the tires for each of the wheels, - a step (502) of estimating an initial slope μSlope of a ratio curve between a mobilized friction coefficient μ <sp / > and a longitudinal slip rate Sx for each of the wheels, - a step (503) of classifying the mobilizable friction coefficient among a finite set of classes, by a machine-learning classification algorithm trained for this purpose on the basis of the longitudinal stiffnesses Cx and slopes μSlope, - a step (504) of using the class of mobilizable friction coefficient calculated by a driving assistance system of the motor vehicle.

Description

Description Title of the invention: Method for characterizing the coefficient of friction available to the wheels of a motor vehicle using information from its longitudinal dynamics. Technical field

[0001] The invention relates to the field of motor vehicles in the broadest sense (i.e., passenger cars, vans, trucks, sports cars, electric or not, autonomous or not), and more specifically to a method for characterizing the coefficient of friction available to the wheels of such a motor vehicle, based on information from its longitudinal dynamics. The estimated coefficient of friction available can then be used by the various driver assistance systems implemented on the vehicle. Previous technique

[0002] Modern vehicles are equipped with increasingly sophisticated Advanced Driver Assistance Systems (ADAS). These ADAS systems aim to improve safety and driving efficiency by controlling various vehicle actuators.

[0003] Indeed, modern vehicles can be viewed as multi-actuator systems (or "overactuated systems") in which different actuators can be used to perform the same action, with the aim of improving comfort or safety. Each actuator has its operating range, within which it is optimal for achieving the best performance. For example, for lateral vehicle control, differential braking actuators are very effective on low-grip surfaces because the self-alignment of the wheels (which characterizes the tendency of the wheels to return to the vehicle's longitudinal axis on their own) is low. Conversely, steering actuators are more effective on high-grip surfaces because the self-alignment of the wheels is high.

[0004] To use the most appropriate actuators for each situation, ADAS systems need as much information as possible about the vehicle's operation. One such piece of information is the maximum tire-road friction coefficient that the vehicle can mobilize. This represents the maximum longitudinal or lateral force that the tires can generate on the road without exceeding their grip limits. This coefficient is also called the tire friction ellipse. The mobilizable friction coefficient depends on road surface grip and intrinsic vehicle characteristics, such as tire type and condition, as well as other factors like weather conditions, tire pressure, and driving dynamics.

[0005] Knowledge of the mobilizable (or potential, or maximum) tire-road friction coefficient can be used, for example, to select the most suitable actuators for lateral or longitudinal vehicle control, to adjust safety distances, etc.

[0006] In the current state of the art, most work aimed at estimating the friction coefficient uses two main approaches: - the so-called "direct" approaches, which require the use of additional sensors, for example acoustic sensors attached to the chassis near the wheels to measure the noise produced by the contact between the tire and the road, or vision sensors (cameras) that capture images of the road for classification purposes; - the so-called "indirect" approaches, which are based on observing the effects of tire-road adhesion on vehicle dynamics. They primarily use information relating to the longitudinal dynamics of the vehicle and wheel slippage.

[0007] Direct approaches involve an additional cost related to the installation of dedicated sensors. It should be noted that today, vehicles are increasingly equipped with sensors that classify road grip for noise control or ADAS functions. However, these solutions do not take into account vehicle dynamics (i.e., how the vehicle moves on the road and how it responds to the forces and movements acting upon it) or tire condition.

[0008] Indirect approaches mitigate the problem of additional cost by using only the information available on the CAN bus (Controller Area Network, a serial system bus used in modern vehicles to allow the various vehicle controllers and sensors to exchange data). However, these approaches require significant stress on the system dynamics to accurately estimate the friction coefficient. Therefore, their response time is highly dependent on the vehicle's driving dynamics.

[0009] Among indirect approaches, one example is US patent 7751961 B2, which forces the vehicle to perform intentional accelerations and decelerations to generate sufficient dynamics to estimate the coefficient. These accelerations and decelerations provide the longitudinal excitations or dynamics of the vehicle necessary to estimate the friction coefficient. This method is advantageous because the engine is used to create longitudinal dynamics even in the absence of driver input, particularly in low-dynamic driving conditions. However, the level of such excitation demand, if not properly controlled, can be intrusive or disruptive for the driver and passengers.This method also leads to an increase in the computational load of the control laws used to regulate, for example, the speed to a given reference, particularly in the case where speed regulation (ACC for "Adaptive Cruise Control") is activated, and an increase in energy consumption.

[0010] One aim of the invention is therefore to define a method for characterizing the tire-road friction coefficient that can be mobilized by the motor vehicle, usable for controlling the vehicle's control laws, according to an indirect approach exploiting the longitudinal dynamic information of the vehicle even when this dynamic is weak (i.e., functioning even in the absence of strong acceleration or braking). Summary of the invention

[0011] To this end, the present invention describes a method for characterizing the coefficient of friction that can be mobilized by one or more wheels of a motor vehicle's drive train, the method comprising: - a step of estimating a longitudinal stiffness C x tires for each of the wheels, - a step of estimating an initial slope of a ratio curve between a mobilized friction coefficient and a longitudinal slip rate S x for each of the wheels, - a classification step of the mobilizable friction coefficient among a finite set of classes of mobilizable friction coefficients, by implementing a machine learning classification algorithm trained for this purpose on the longitudinal stiffness(es) C x and the said slope(s) -siope - a step of using the class of mobilizable friction coefficient calculated during the previous step by a driving assistance system of the motor vehicle.

[0012] Advantageously, the step of classifying the friction coefficient that can be mobilized by the motor vehicle may further include providing the machine learning classification algorithm with at least one type of data from among: - data on the nature and condition of a road surface, - meteorological data, - data relating to the vehicle's tires, - data relating to the vehicle's dynamics.

[0013] Depending on different embodiments, the step of estimating a longitudinal stiffness C x may include smoothing of the calculated longitudinal stiffness value and / or the slope estimation step -siope may include smoothing of the calculated siope slope value.

[0014] Advantageously, the method according to the invention may include a preliminary step of validating the longitudinal dynamic conditions of the motor vehicle, the steps of estimating a longitudinal stiffness C x tire and slope estimation -siope are only carried out when the longitudinal dynamic conditions are reached.

[0015] According to a more precise embodiment, the longitudinal dynamic conditions of the motor vehicle are reached when the longitudinal dynamic of the motor vehicle is greater than its lateral dynamic.

[0016] According to an advantageous embodiment of the invention, the finite set of classes of friction coefficients that can be mobilized by the vehicle comprises two classes (strong / weak), and the machine learning classification algorithm is a logistic regression algorithm.

[0017] According to an advantageous embodiment of the invention, the step of estimating a longitudinal stiffness C x Tires include: - the calculation of a longitudinal force F x practicing on the wheel, - the calculation of a longitudinal slip rate S x of the wheel, - the calculation of the longitudinal stiffness C x such as : r dF * ds x

[0018] when S x is close to zero, typically in an interval [1% - 3%]-

[0019] According to an advantageous embodiment of the invention, the step of estimating an initial slope of the ratio curve between the mobilized friction coefficient and the longitudinal slip rate S x understand : - the calculation of the coefficient of friction mobilized by the wheel, - the calculation of the longitudinal slip rate S x of the wheel, - the calculation of the slope -siope such that: — ^ Slope dS x

[0020] when S x is close to zero, typically in an interval [1% - 3%]-

[0021] According to a particular embodiment of the invention, the supervised learning classification algorithm is configured to provide a probability associated with the class of the mobilizable friction coefficient, the probability being taken into account during the step of using the class of mobilizable friction coefficient by a vehicle driving assistance system.

[0022] The method according to the invention may include a preliminary step of training the machine learning classification algorithm, the training comprising: - the provision of a plurality of input data including longitudinal stiffness pairs C xof tires and initial slope of the curve of the ratio between the mobilized friction coefficient and the longitudinal slip rate S x for one or more wheels of the vehicle, - the association (602) of a class of mobilizable friction coefficients corresponding to each supplied torque, and - the determination (603) of a classification algorithm model allowing the best possible matching of pairs with corresponding classes.

[0023] Finally, the invention also relates to a motor vehicle comprising an electronic data processing device configured to implement a method for characterizing the mobilizable friction coefficient as described above. Brief description of the drawings

[0024] The invention will be better understood and other features, details and advantages will become clearer from the following description, given by way of non-limiting example, and from the accompanying figures, given by way of example.

[0025] [Fig. 1] Figure 1 shows the evolution of the longitudinal force F x exerted on the tires of a vehicle as a function of the longitudinal slip rate S x wheels for two classes of support.

[0026] [Fig. 2] Figure 2 shows measurements of longitudinal stiffness C x tires of a vehicle traveling on two classes of road surface.

[0027] [Fig. 3] Figure 3 shows the evolution of the mobilized friction coefficient as a function of the longitudinal slip rate S x of the wheel for two classes of rolling.

[0028] [Fig. 4] Figure 4 shows measurements of the initial V-slope of the curve of the ratio between the coefficient of friction mobilized by the wheel of a vehicle and the longitudinal slip rate S x of this wheel.

[0029] [Fig.5] Fig.5 schematically represents the steps of a process for characterizing the mobilizable friction coefficient of a motor vehicle according to the invention.

[0030] [Fig.6] Fig.6 schematically represents steps to carry out the learning of the classification algorithm used in the process of characterizing the coefficient of friction according to the invention. Description of the implementation methods

[0031] In the rest of the description, and unless otherwise stated, the term "friction coefficient" will be used to refer to the "friction coefficient that can be mobilized by the vehicle".

[0032] The invention relates to a method for characterizing the coefficient of friction available to a motor vehicle. The purpose of the method is not to precisely estimate this coefficient of friction, but to quickly determine a class representative of its value, possibly associated with a confidence level. Any number of classes can be used, for example, two classes of values ​​(high / low coefficient of friction), or three classes of values ​​(high / medium / low coefficient of friction).

[0033] To achieve this, the method according to the invention relies on processing data relating to the longitudinal dynamics of the vehicle using a supervised machine learning classification algorithm. A classification algorithm is an algorithm that classifies elements or data into predefined categories based on certain common characteristics and provides the probabilities of belonging to each class. These algorithms are commonly used in fields such as image recognition, data analysis, natural language processing, document classification, etc.

[0034] Classification algorithms are commonly used today for ADAS (Advanced Driver Assistance Systems), but primarily for identifying contextual elements of the vehicle's environment from camera images. The invention proposes using such a classification algorithm to identify the tire-road friction coefficient from data other than optical flow. One of the challenges lies in selecting the data used as input for these classification algorithms, which must be relevant to the target classes for the classification results to be meaningful.

[0035] To this end, the method according to the invention proposes to use a supervised learning classification algorithm to determine the class of the friction coefficient that can be mobilized by the vehicle, taking as inputs at least two parameters representative of its longitudinal dynamics: - longitudinal stiffness C x tires of one or more wheels of a steering wheel assembly of the vehicle, and the slope, 9 u i is the a initial slope (or slope at origin) of the ratio between the mobilized friction coefficient and the longitudinal slip rate S x for this wheel or these wheels.

[0036] Longitudinal stiffness C x A tire's longitudinal stiffness refers to its ability to resist longitudinal deformation when subjected to tensile or braking forces during driving. The higher the longitudinal stiffness, the better the tire maintains its shape and stability while in motion, contributing to improved road grip, handling, and braking performance.

[0037] Figure 1 shows the evolution of the longitudinal force F xexerted on the tires as a function of the longitudinal slip rate S x Wheels for two surface classes: one class with low grip (e.g., snow) and a class I2 with high grip (e.g., dry asphalt). These curves show a linear zone towards the origin, followed by a pseudo-linear zone, and then a zone where they reach a plateau (saturation zone) corresponding to the maximum longitudinal force that the vehicle can exert on the corresponding surface. The longitudinal stiffness C x corresponds to the absolute value of the slope of the curve of the ratio between the force F x and the slip rate S x wheels when the values ​​of S x are very small, that is to say in the initial linear zone of the curve.

[0038] This figure highlights the fact that the longitudinal stiffness C xThe stiffness obtained for a low-adhesion support is less than the longitudinal stiffness C x2 obtained for a high-adhesion substrate. Therefore, there is a link between the longitudinal stiffness C x tires and the coefficient of friction that can be mobilized by the vehicle. This is why the invention uses this information to characterize the class of the coefficient of friction that can be mobilized.

[0039] Figure 2 shows measurements of longitudinal stiffness C x Tires of a vehicle traveling on two road surface classes were analyzed. The results are represented as a density, with the dispersion along the x-axis proportional to this density. It is clear that for a tire, the better the contact with the road surface, the higher the longitudinal stiffness value C. x is high.

[0040] Figure 3 shows the evolution of the mobilized friction coefficient as a function of the longitudinal slip rate S x wheels, here too for both support classes and J2 - Just as in Figure 1, these curves reach a plateau, respectively j and J2, corresponding to the maximum friction coefficient that can be mobilized by the vehicle on the corresponding support. The siop value is l a slope of the curve, ratio between the coefficient of friction mobilized (i.e., used) by a vehicle wheel and the longitudinal slip rate S x of this wheel in the initial zone of the curve, that is to say for low values ​​of S x .

[0041] This figure highlights the fact that, for mobilizable adhesion coefficients j and avec < ^2' P enteThe slope corresponding to j is less than the slope Psiope2 corresponding to P^. This clearly shows that there is a link between the slope siope et I e mobilizable friction coefficient of the vehicle, which is why the invention also uses this information for the characterization of the friction coefficient class.

[0042] Figure 4 shows P slope measurements obtained by driving a vehicle on two surface classes: I1 (low grip) and I2 (high grip). The results are represented as a density, with the dispersion along the x-axis representing this density.

[0043] It also appears here that for a tire, the better the contact with the road surface, the higher the slope values. sonthigh. The measurement of this slope, near the origin of the curve, is therefore a good representation of the coefficient of friction that can be mobilized by the vehicle.

[0044] The method according to the invention consists of using a classification algorithm to define a class of the friction coefficient that can be mobilized by the vehicle. The inputs to this classification algorithm are values ​​of the vehicle's longitudinal dynamics that are representative of this friction coefficient, namely at least: - the value of the longitudinal stiffness coefficient C x tires for one or more wheels of a vehicle's drive axle and - the Psiope value of the slope of the ratio between the coefficient of friction mobilized by the wheel(s) and the longitudinal slip rate S x of this wheel or these wheels, for values ​​of S x weak.

[0045] One of the advantages of the invention is that these values ​​are available or directly derivable from vehicle data transmitted over the CAN bus. Therefore, implementing the invention does not require the installation of dedicated sensors.

[0046] Figure 5 schematically represents the steps of a process for characterizing the coefficient of friction that can be mobilized by a motor vehicle according to an embodiment of the invention.

[0047] The process includes a first step 501 of estimating a longitudinal stiffness C x Tires on at least one wheel of a set of drive wheels of the vehicle. Longitudinal stiffness C x corresponds to the slope at the origin of the curve relating the longitudinal force F x exerted on the tires at the slip rate S x wheels, so that:c x = ^s x =o

[0048] The formula given above is the theoretical formula. In practice, the longitudinal stiffness C will be calculated. x in the initial linear zone of the curve, that is, for values ​​of the slip rate S x very small. For example, the calculation can be performed when S x E [ 1%- 3%] , in order to avoid both noise related to vibratory phenomena which can disrupt the calculation around the origin of the curve, and values ​​from non-linear areas of the curve.

[0049] The longitudinal force F x The force exerted on a vehicle's wheel can be measured or estimated using equations characterizing the vehicle's longitudinal dynamics: for example, by dividing the engine torque applied to a set of wheels by the wheel radius to obtain the longitudinal force applied to the set of wheels, and then dividing by the number of wheels. The longitudinal force F xcan also be obtained using a nonlinear Kalman filter-type estimator that takes into account the vehicle's longitudinal velocity, longitudinal accelerations, yaw rate, etc. The dF value x represents the variation of F x over a given period.

[0050] The value S x The wheel slip ratio can be measured or estimated, for example, from the steering wheel angle or yaw rate. For example, the SAE (Society of Automotive Engineers) standard gives the following definition for estimating the wheel slip ratio:

[0051] ç _ (VX vR x M)

[0052] with : - V x the linear speed at the center of the wheel, - R the radius of the wheel, and - M the rotational speed of the wheel.

[0053] Other methods for calculating S xcan be considered, for example by distinguishing the acceleration and braking phases of the vehicle according to the following equation:

[0054] V x if braking (RCÜ-V X ) Rai otherwise

[0055] In order to characterize the longitudinal stiffness C x , several sets of simultaneous values ​​of F x and S x must be acquired or calculated.

[0056] Advantageously, the longitudinal stiffness C x The calculated value can be smoothed, for example by a Kalman filter, a moving average, a recursive least squares algorithm, or any other similar technique. Such smoothing helps to avoid abrupt jumps in C x related to point-by-point variations not representative of a change in tire-road contact adhesion, particularly when adhesion conditions are on the border between two classes.

[0057] The method for characterizing the coefficient of friction mobilized by a motor vehicle according to the invention also includes a second step 502 of estimating the slope between the coefficient of friction mobilized by one or more wheels of the vehicle and the longitudinal slip rate S x of these wheels in the initial part of the curve, that is to say for values ​​of S x weak. This step can be performed before, after or at the same time as step 501. It necessarily relates to the same wheel(s).

[0058] The coefficient of friction mobilized by a vehicle's wheel can be calculated in various ways known to those skilled in the art, using models of the vehicle's tires, friction force estimators, vehicle accelerations, or other factors. For example, the following equation can be used:

[0059] ^( t )= __ [ 0060] where: - F represents the lateral force exerted on the vehicle's wheel, and - F z represents the vertical force exerted on the vehicle's wheel.

[0061] The F forces x , F y and F z can be measured or estimated. For example, in a manner comparable to F x The force F can be obtained from a nonlinear Kalman filter using information from the vehicle's lateral dynamics (lateral velocity and acceleration, yaw rate, etc.). z can be directly derived from the mass of the vehicle.

[0062]

[0063] The slope value at the origin of the ratio between the coefficient of friction mobilized and the longitudinal slip rate S x of one or more wheels of a set of drive wheels of the vehicle is then equal to:

[0064] "siope'- l ' U

[0065] with d]l and dS x the respective variations of the mobilized friction coefficient and the longitudinal slip rate S x the wheels of the vehicle.

[0066] The formula given above is the theoretical formula. In practice, the Psiope slope will be calculated in the initial linear zone of the curve, that is, for very low slip values. For example, the calculation can be performed when S X E [ 1 % - 3 % ] , in order to avoid vibrational phenomena that could disrupt the calculation at the origin of the curve and values ​​coming from non-linear areas of the curve.

[0067] In order to characterize the slope P-siope, several sets of simultaneous values ​​of and S x must be acquired or calculated.

[0068] Advantageously, the calculated Psiope slope value can be smoothed, for example, by a Kalman filter, a moving average, a recursive least squares algorithm, or any other similar technique. Such smoothing avoids abrupt jumps in Psiope due to point variations that do not accurately reflect a change in tire-road contact adhesion, particularly when the adhesion conditions are at the boundary between two classes.

[0069] The method according to the invention includes a third step 503 of classifying the mobilizable friction coefficient of the vehicle for the wheel(s) concerned from among a finite set of classes of mobilizable friction coefficients. This step is performed by a machine learning classification algorithm previously trained for this purpose, using the longitudinal dynamic values ​​of the vehicle that represent the mobilizable friction coefficient, namely the longitudinal stiffness C x calculated during step 501 and the siope slope calculated during step 502.

[0070] When the process is implemented on a single wheel, the inputs to the classification algorithm are therefore the longitudinal stiffness C xand the slope P slope - The algorithm's output is the mobilizable friction coefficient for that wheel. When the process is implemented on both wheels of the same drive wheel set of the vehicle, a longitudinal stiffness torque C x / slope P slope est calculated for each wheel. Two different implementation methods are then possible: - the inputs of the classification algorithm are the two pairs of values, and the output is a mobilizable friction coefficient for the wheels of the wheel set; - The inputs to the classification algorithm are the average of the longitudinal stiffness values ​​C x and the average of the slope values ​​- The output is a mobilizable friction coefficient for the wheels of the wheel train.

[0071] When the vehicle has several sets of driven wheels, the method according to the invention can be implemented for each of these sets of wheels, in order to obtain a mobilizable friction coefficient specific to each set of wheels of the vehicle.

[0072] In this way, the calculated mobilizable friction coefficient class can be specific to one drive wheel of the vehicle, to each drive wheel of the vehicle, to one set of drive wheels of the vehicle or to several sets of drive wheels of the vehicle, as required.

[0073] The machine learning classification algorithm used can, for example, be a neural network, logistic regression, decision tree, etc. In the case of classifying the mobilizable friction coefficient into only two classes, the machine learning classification algorithm used is advantageously a logistic regression, as it is one of the best methods for binary classification, with a smooth transition approximated by a sigmoid function between the two classes.

[0074] The use of a machine learning classification algorithm allows us to rely on multiple inputs to determine the class of the friction coefficient that can be mobilized by the vehicle. It takes as inputs at least two values ​​of the vehicle's longitudinal dynamics that are representative of the mobilizable friction coefficient, called primary indicators, namely the longitudinal stiffness C xand the initial slope of the ratio curve between the mobilized friction coefficient and the longitudinal slip rate S x , for values ​​of S x weak, for one or more wheels of a set of drive wheels.

[0075] The inputs to the classification algorithm can also include secondary indicators as long as they are also representative of this friction coefficient, such as: - data on the nature and condition of the road surface, obtained for example from a camera or provided by vehicles preceding the ego vehicle, - meteorological data, obtained for example by a thermometer, a rain sensor, and / or a weather station, - data relating to tires, such as tire type, pressure, age and / or wear condition, - data relating to vehicle dynamics, such as speed, acceleration, yaw rate and / or slip rate.

[0076] The list of secondary indicators is not exhaustive. These secondary indicators help to improve the accuracy of the classification performed by the supervised learning classification algorithm.

[0077] The method according to the invention finally includes a fourth step 504 of using the class of available friction coefficient calculated in step 503 by a vehicle driver assistance system, for example, for an ADAS system mobilizing the braking, acceleration, or steering actuators for trajectory tracking or lane centering, for adaptive ACC to adjust safety distances from the vehicle ahead, or for collision detection. It could also be an engine control system that adjusts the torque setting or the maximum engine power demand according to the grip class, etc.

[0078] The classification algorithm is trained offline (i.e., outside the vehicle) and in a supervised manner. This involves providing the declassification algorithm with a set of known input and output data, called training data, so that it can learn to associate the inputs with their corresponding outputs by creating a model or mathematical function usable during the inference phase. The training data can come from measurements taken on a test vehicle or be extracted from a driving database.

[0079] Figure 6 schematically represents the steps of an embodiment of the learning of a classification algorithm usable in the characterization process according to the invention. The learning is carried out by considering input / output data of the algorithm of the same type as those used in the classification step 503.

[0080] Learning can then include the following steps: - provision of input data 601, this input data comprising at a minimum pairs of concurrent measurements of longitudinal stiffness C x tires and the V-slope of one or more wheels of the vehicle. The collected input data must be sufficiently numerous and representative of various application cases to allow the classification algorithm to offer satisfactory performance regardless of its implementation conditions. For example, a minimum of one million sets of measurements taken under various conditions of tire-road contact adhesion, weather conditions, tire condition, road surface condition and vehicle dynamics will be used; - Association 602 of a class of friction coefficients that can be mobilized by the vehicle to each pair of input samples. This association (or labeling) consists of assigning a class from among the N desired output classes to each set of measurements. The output class can be determined, for example: • at the driver's discretion, • by direct measurements of the road surface friction coefficient, • by approximation based on climatic conditions, the condition of the road surface, and / or the condition of the vehicle's tires; • etc... - Training 603 of the supervised machine learning classification algorithm. This step consists of providing it with the input data from step 601 and the corresponding classes from step 602, which constitute the training data, so that it learns to best associate the input data with the different classes sought; validation 604 of the trained classification algorithm obtained, by introducing new input data into the model and verifying whether the algorithm's outputs correspond to the corresponding classes. A prediction accuracy rate per input class can then be measured. If this accuracy rate is greater than a given threshold for each of the classes considered, for example 95% accuracy, then the trained algorithm can be deployed for implementation. Otherwise, the algorithm's accuracy is insufficient, and training must be restarted with new training data.

[0081] The trained classification algorithm can be used to implement the method for characterizing the friction coefficient available to the motor vehicle according to the invention on all vehicles in the range of the vehicle used to collect the training data, as well as on comparable vehicles. Comparable means that the characteristics of the vehicles, in particular their mass, dimensions, and tire size, are of the same order of magnitude.

[0082] According to a particular embodiment of the method according to the invention, the supervised learning classification algorithm implemented in step 503, which classifies the vehicle's available friction coefficient from a finite set of available friction coefficient classes, is configured to provide a probability associated with each classification. This probability provides an indication of the reliability of the determination, which can be taken into account to adapt step 504, which uses the calculated available friction coefficient class from a vehicle driver assistance system.

[0083] According to a particular embodiment of the method for characterizing the coefficient of friction available to the motor vehicle according to the invention, the steps of the method are carried out only when the vehicle is in conditions where longitudinal dynamics prevail over lateral dynamics, conditions in which the longitudinal stiffness estimates C x and slope are the es pl us precise. Thus, this particular embodiment of the method according to the invention includes a preliminary step 505 of validating the longitudinal dynamic conditions of the motor vehicle. This validation can be done in different ways, for example from longitudinal acceleration signals a x and lateral ây of the vehicle, taking: Longitudinal dynamics^ t) = kl

[0084] In this embodiment, steps 501 to 504 of the process are performed only when Longitudinal Dynamics^) is greater than or equal to a predefined threshold, for example 0.5 or 0.9. Setting the threshold to 0.5 ensures that the longitudinal stiffness C estimates x Tire and slope tests are carried out under conditions where the longitudinal dynamics of the vehicle are more stressed than the vertical dynamics.

[0085] In another embodiment, steps 501 and 502 are performed only when the vehicle is in conditions where sufficient longitudinal dynamics are exerted, i.e. when Longitudinal dynamics^ t) is greater than or equal to the predefined threshold. When this is not the case, the outputs of these two estimation steps correspond to the latest longitudinal stiffness values ​​C xtires and slope calculated under satisfactory longitudinal dynamic conditions.

[0086] The invention relates to a method for characterizing the coefficient of friction available to a motor vehicle, implementing a supervised machine learning classification algorithm that takes as minimum input an estimate of the longitudinal stiffness C x tires of one or more wheels of the vehicle, and an estimate of the initial slope of the curve relating the coefficient of friction mobilized by the vehicle on this or these wheels to the longitudinal slip rate S x wheels. This process is intended to be implemented on board a vehicle to allow it, for example at regular time intervals, to perform an estimation of the mobilizable friction coefficient.

[0087] When steps 501 and 502 include smoothing of longitudinal stiffness measurements Cx and / or Psiope slope they can be carried out at higher frequencies than the 503 classification step, in order to increase the number of readings to converge towards very precise measurements.

[0088] The rate at which step 504 of using the class of friction coefficient mobilizable by the motor vehicle by a driving assistance system is carried out is independent of the rate of the preceding steps.

[0089] The main advantages of the classification method according to the invention are: - it allows for the rapid detection of intervals of the mobilizable friction coefficient, based on the longitudinal dynamics of the vehicle. Unlike prior art methods, it does not require pushing the vehicle to the limits of its adhesion to quantify the mobilizable friction coefficient: the indicators used by the declassification algorithm correspond to slopes at the origin, which are therefore available as soon as a minimum of longitudinal dynamics is present; - It can be implemented using signals available on the vehicle's CAN bus: therefore, there is no need to install specific sensors; - It improves vehicle safety, whether it is an autonomous vehicle or a vehicle equipped with driver assistance systems; - It allows for the selection of the most suitable vehicle actuators for the context, and for better management of the motor power of these actuators, thanks to knowledge of the class of friction coefficient that can be mobilized.

[0090] The invention also relates to a vehicle comprising an electronic data processing device, for example an on-board computer or a microprocessor, configured to implement the method of characterizing the coefficient of friction that can be mobilized by a motor vehicle described above, from data acquired by sensors of the vehicle.

Claims

Demands

1. A method for characterizing the coefficient of friction that can be mobilized by one or more wheels of a motor vehicle's drive train, the method comprising: - a step (501) for estimating a longitudinal stiffness C x tires for each of said one or more wheels, - a step (502) of estimating an initial Psiope slope of a ratio curve between a friction coefficient mobilized and a longitudinal slip rate S x for each of the said one or more wheels, - a step (503) of classifying the mobilizable friction coefficient among a finite set of classes of mobilizable friction coefficients, by implementing a machine learning classification algorithm trained for this purpose on said longitudinal stiffness(es) C x and the said slope(s)S Iope, - a step (504) of using the class of mobilizable friction coefficient calculated during the previous step by a motor vehicle driving assistance system.

2. A method for characterizing the mobilizable friction coefficient according to claim 1, wherein the step (503) of classifying the mobilizable friction coefficient of the motor vehicle further comprises providing the machine learning classification algorithm with at least one type of data from: - data on the nature and condition of a road surface, - meteorological data, - data relating to the vehicle's tires, - data relating to the vehicle's dynamics.

3. A method for characterizing the mobilizable friction coefficient according to any one of the preceding claims, wherein step (501) of estimating a longitudinal stiffness C x includes smoothing of a calculated longitudinal stiffness value and / or step (502) of estimating a P slope includes smoothing of a calculated P slope value.

4. A method for characterizing the mobilizable friction coefficient according to any one of the preceding claims, further comprising a preliminary step (505) for validating the longitudinal dynamic conditions of the motor vehicle, and steps (501, 502) for estimating a longitudinal stiffness C x tires and P slope estimation are only performed when longitudinal dynamic conditions are met.

5. A method for characterizing the mobilizable friction coefficient according to the preceding claim, wherein the longitudinal dynamic conditions of the motor vehicle are reached when the longitudinal dynamics of the motor vehicle are greater than its lateral dynamics.

6. A method for characterizing the mobilizable friction coefficient according to any one of the preceding claims, wherein the finite set of classes of friction coefficients mobilizable by the vehicle comprises two classes, and wherein the machine learning classification algorithm is a logistic regression algorithm.

7. A method for characterizing the mobilizable friction coefficient according to any one of the preceding claims, wherein step (501) of estimating a longitudinal stiffness C x Tires include: - the calculation of a longitudinal force F xpracticing on the wheel, - the calculation of a longitudinal slip rate S x of the wheel, - the calculation of the longitudinal stiffness C x such as : r dF * dS x when S x is close to zero, typically in an interval [1%-3%]-

8. Method for characterizing the mobilizable friction coefficient according to any one of the preceding claims, wherein step (502) of estimating an initial slope of the ratio curve between the mobilized friction coefficient and the longitudinal slip rate S x includes: - calculating the coefficient of friction mobilized by the wheel, - the calculation of the longitudinal slip rate S x of the wheel, - the calculation of the Psiope slope te l 9 ue : — dp ^ Slope dS x when S x is close to zero, typically in an interval [1%-3%]-

9. Method for characterizing the mobilizable friction coefficient according to any one of the preceding claims, wherein the supervised learning classification algorithm is configured to provide a probability associated with the class of the mobilizable friction coefficient, said probability being taken into account during step (504) of using the class of mobilizable friction coefficient by a vehicle driving assistance system.

10. A method for characterizing the mobilizable friction coefficient according to any one of the preceding claims, further comprising a preliminary step of training the machine learning classification algorithm, the training comprising: - the provision (601) of a plurality of input data including longitudinal stiffness pairs C xof tires and initial slope of the curve of the ratio between the coefficient of friction mobilized and the longitudinal slip rate S x for one or more wheels of the vehicle, - the association (602) of a class of mobilizable friction coefficient corresponding to each supplied couple, and - the determination (603) of a classification algorithm model allowing to match the couples with the corresponding classes as closely as possible.

11. Motor vehicle comprising an electronic data processing device configured to implement a method for characterizing the mobilizable friction coefficient according to any one of claims 1 to 10.

Images