Coupling of slope and rolling radius estimators, improved vehicle mass estimator

Abstract

Recursive estimators rely on modeling the underlying physical system to determine a vehicle's mass from engine torque, longitudinal acceleration measurements, and wheel angular velocity measurements. To improve the estimates, the angular acceleration inherently generated by a polynomial smoother preprocessing the measurements is reused in the recursive estimation of the road slope and the vehicle's longitudinal speed. Furthermore, to obtain observable models without unnecessary simplifications, this recursive slope estimator is coupled with a recursive tire dynamic radius estimator so that their estimated longitudinal speed and estimated dynamic radius are used in the other estimator's estimation.Estimating the dynamic radius allows the use of more complex models; the accuracy of the vehicle mass estimation is improved by the accuracy of the road's actual slope estimation. See Figure 2 for the abbreviation.

Description

Title of the invention: Coupling of slope and rolling radius estimators, improved vehicle mass estimator. Technical field of disclosure.

[0001] The present invention relates to the general field of recursive estimators of wheeled vehicle parameters or states. Prior art

[0002] The "states" refer to vehicle data that vary over time, such as road slope or speed, while the "parameters" refer to vehicle data that do not vary over time, such as vehicle surface.

[0003] The term "recursive" indicates that the estimations are performed again as soon as new data are acquired and available. Recursion allows for "real-time" estimation, that is, continuous estimation at each new time of data acquisition.

[0004] Certain states and parameters can be measured directly using sensors. This is the case, for example, for the vehicle's engine torque, the vehicle's longitudinal acceleration, the engine's angular velocity, and the wheel's angular velocity.

[0005] Other states and parameters cannot be measured and are therefore estimated recursively from the measurements obtained. This is the case for the mass of the vehicle, the rolling resistance of the vehicle, the actual longitudinal speed of the vehicle, the effective rolling radius of a tire, the center of gravity of the vehicle, the aerodynamic coefficient of the vehicle, the grip of the tires, the wear of the tires, the load on the tires, the stiffness of the tires (Kx, Dz), etc.

[0006] These estimates are crucial for many applications.

[0007] For example, estimating the vehicle's mass makes it possible to determine the vertical load on each tire, which in turn allows for a precise estimation of tire wear, tire grip, and tire rolling resistance during usage phases, enabling appropriate measures to be taken. Typically, knowledge (estimation) of the vehicle's overall load allows for better prediction of the battery range of electric vehicles (EVs), whose transported load can vary considerably over time. Improved routing of EVs to charging stations can then be proposed, optimizing battery usage and wear.

[0008] To do this, recursive estimators or “virtual sensors” (the data are not directly measured by the sensors, but obtained indirectly, estimated at (starting from existing data without a dedicated sensor) are implemented which rely on a modeling of the underlying physical system.

[0009] To improve the quality of the estimated data, it is important to use accurate and representative physical models of the physical system for the estimation process. However, overly complex models are not observable; that is, they contain more unknown variables or parameters than known (i.e., measured) variables or parameters. Conversely, overly simplified models compromise the quality of the estimated data. "Quality" can refer to data reliability, data accuracy, data adaptability to different types of vehicles, and / or data robustness to different driving scenarios.

[0010] Today in particular, vehicle mass estimators rely on simplifying assumptions of a linear system (model) of the longitudinal dynamics of the vehicle (Newton's second law), to obtain observable models from measurements of the longitudinal acceleration of the vehicle, the angular velocity of the wheels and the engine torque of the vehicle.

[0011] A first simplification consists of neglecting the effects of the headwind, as well as the additional inertial masses due to the rotating components of the vehicle's powertrain.

[0012] A second simplification lies in assimilating the dynamic rolling radius to the nominal radius of the tire, thus neglecting variations in the tire radius over time. Indeed, the dynamic rolling radius is rather difficult to estimate in practice, or at the very least requires a dedicated estimator necessitating additional measurements.

[0013] A third simplification lies in assimilating the road slope as perceived by the accelerometer (measuring longitudinal acceleration) to the actual road slope, thus neglecting any errors (e.g., those arising from accelerator misalignment, the load distribution acting on the vehicle's suspension, and the vehicle's suspension dynamics). The road slope can be estimated using the same estimator as the vehicle mass or a dedicated estimator. However, existing estimators of the actual road slope remain, for the time being, of limited quality.

[0014] These simplifications nevertheless impair the final quality of the estimates, particularly of the vehicle mass.

[0015] There is therefore a need to remedy the aforementioned drawbacks, in particular with a view to improving the recursive estimation of the mass of a vehicle, with the same input measurements. Statement of Disclosure

[0016] The inventors have found that a coupling between the determination of the slope of the road and the dynamic rolling radius of the wheels makes it possible to restore the observability of a more complex model, and in particular to avoid the second simplification above.

[0017] In this context, the present disclosure proposes a device for monitoring a vehicle operating on a road, the device comprising: a communication interface for obtaining, from one or more sensors, measurements of the vehicle's longitudinal acceleration (ax>1MU) and the angular velocity of the vehicle's wheels (œx), one or more processors implementing: - a recursive slope estimator estimating a road slope and a longitudinal vehicle speed, from measurements of longitudinal acceleration and angular velocity of the wheels and a dynamic rolling radius of the tires, and - a recursive radius estimator estimating the dynamic rolling radius of the tires (rdyn) from the measurement of angular velocity of the wheels and the longitudinal speed estimated by the recursive slope estimator.

[0018] In addition to the observability of a more complex model, this dual estimation resulting from the coupling improves the quality of the estimates because both estimators now rely on a higher-quality estimate of the longitudinal velocity and the dynamic rolling radius. Indeed, the dynamic rolling radius is very sensitive to the longitudinal velocity, just as the road gradient is very sensitive to the rolling radius.

[0019] Typically, the estimate obtained by one of the estimators in the previous recursion can be used as input to the other estimator in the next recursion. Alternatively, the two estimators can be serialized, one using, as input, the estimate of the other obtained in the previous recursion, while the other (downstream in the series) uses, as input, the estimate of the first estimator obtained in the current recursion. In other words, the recursive slope estimator estimates the road slope at recursion k from the dynamic rolling radius of the tires estimated by the recursive radius estimator at recursion k-1, and / or the recursive radius estimator estimates the dynamic rolling radius of the tires at recursion k from the longitudinal speed estimated by the recursive slope estimator at recursion k-1.

[0020] Alternatively, simultaneous estimation is also possible where the two estimators perform their calculations in parallel, each estimator improving the accuracy of the other estimator during the same recursion, for example by exchanging intermediate estimates. In other words, the recursive slope estimator and the recursive radius estimator perform iterative estimations during a recursion. of estimation and exchange their previous iterative estimates (obtained at iteration i-1) to perform the next iterative estimation (iteration i).

[0021] The disclosure also relates to a system comprising a supervisory device as defined above and a vehicle control unit for controlling an on-board function based on the estimates from the supervisory device. As mentioned previously, the road gradient and longitudinal speed thus estimated can be used in estimating the vehicle's mass to try to improve vehicle safety and control. For example, it is then possible to accurately estimate the vertical load experienced by each tire, tire wear, tire grip, tire rolling resistance, predict the battery range of electric vehicles (EVs), and thus take appropriate measures (illuminate an overload indicator, indicate tire wear status, adjust a vehicle electric range indicator or route the vehicle to charging stations, etc.).

[0022] Optional features of embodiments are defined in the appended claims. Some of these features are explained below with reference to a device, while they can be transposed into process features.

[0023] In one embodiment, the recursive slope estimator estimates the road slope and the longitudinal speed of the vehicle, further using an angular acceleration of the wheels obtained from measurements of the angular velocity of the vehicle's wheels. This arrangement improves the quality of the estimates.

[0024] In a particular embodiment, the monitoring device includes a polynomial measurement smoother, generating the angular acceleration of the vehicle wheels by deriving the angular velocity of the wheels.

[0025] Indeed, existing recursive estimators of real slope are generally implemented using linear Kalman filters, which receive measurements of the vehicle's longitudinal acceleration and the angular velocity of the wheels from polynomial smoothers. Polynomial smoothing is typically a polynomial regression of a measurement signal in order to eliminate, locally at each measurement, the noise of the sensor used.

[0026] To be effective, the smoothing uses polynomials of multiple order (two or more). The above configuration thus takes advantage of the intrinsic calculations of polynomial smoothers—including that of the (smoothed and accurate) derivative of the input measurements—to improve existing real slope estimators by integrating, into the conventional inputs, the angular acceleration of the wheels obtained from a polynomial smoother by differentiating the measured angular velocity. The quality of the estimation is thereby finds improved, without requiring additional calculation module, additional measurement input.

[0027] In one embodiment, the polynomial smoother comprises a polynomial Kalman smoother. Alternatively, a simple Savitzky-Golay smoothing filter may be used.

[0028] Alternatively, the monitoring device may include an angular velocity differentiator calculating the angular acceleration of the vehicle's wheels by difference between successive measurements of the angular velocity of successive wheels.

[0029] In one embodiment, the recursive slope estimator includes a linear Kalman filter estimating the road slope (a), a longitudinal acceleration of the vehicle (ax), and the longitudinal speed of the vehicle (vx). The filter operates on input data consisting of measurements of the longitudinal acceleration and angular velocity of the wheels, optionally the angular acceleration of the wheels, and the rolling radius of the vehicle's tires.

[0030] In one embodiment, the underlying model of the recursive slope estimator includes the following state evolution system:

[0031] vx(t)= vx(t-1)+ax(t). At,

[0032] ax(t)= ax(tl)

[0033] g.sin(a(t))=g.sin(a(tl))

[0034] and the following measurement system:

[0035] vx(t)=œx(t).rdyn(t),

[0036] ax(t)= ( Z ) .rdyn(t),

[0037] ax>IMU(t)=ax(t)+g.sin(a(t)),

[0038] where ax is an estimated longitudinal acceleration of the vehicle, g the gravitational constant, At the time step between two estimation recursions, and an angular acceleration of the vehicle's wheels.

[0039] In one embodiment, the underlying model of the recursive radius estimator includes: vx(t)=cox(t).rdyn(t). Compared to a direct calculation of rdyn, the use of the recursive radius estimator allows for a noise-robust (noise from sensors, smoothing) and therefore accurate estimation.

[0040] In a particular embodiment, the recursive radius estimator comprises a recursive least squares filter whose parameter vector is _ r _ ]r, the regression vector is a = [ ] and the measures are y =

[0041] In one embodiment, the monitoring device further includes a recursive vehicle mass estimator estimating a vehicle mass (mv) from the estimated road slope (a), the estimated longitudinal speed (vx) and the estimated dynamic rolling radius of the tires (rdyn).

[0042] In a particular embodiment, the recursive mass estimator includes a recursive least squares estimator.

[0043] In a particular embodiment, the underlying model of the recursive mass estimator includes:

[0044] (O ax (t) + mv (t) gsin (a (t)) rdy^

[0045] + m v (t)g(Çq + c r [ v x (t) + ç4v x (t) 4 ) + 5 P air A^ d v x (f) 2

[0046] where Meng(t) is a motor torque, y(t) a transmission ratio, qtot(t) a mechanical or motor transmission efficiency, ax an estimated longitudinal acceleration of the vehicle, g the gravitational constant, cr0, cr[, cr4 rolling resistance coefficients, pair an ambient air density, Af a frontal area of ​​the vehicle and cd an aerodynamic coefficient of the vehicle.

[0047] Equivalent models may be used in which the term mv(t).g.(Cr+c ri.vx(t)+cr4.vx(t)4) is replaced by any model of the form f(vx(t), a(t)). By way of example, the expression mv(t).g.cr.cos(a(t)) may also be used.

[0048] The disclosure also relates to a method for supervising a vehicle operating on a road, the method comprising the following steps: to obtain, via a communication interface, measurements of longitudinal acceleration of the vehicle and angular velocity of the vehicle's wheels acquired by one or more sensors, to estimate, recursively, the road slope and the longitudinal speed of the vehicle, from measurements of longitudinal acceleration and angular velocity of the wheels and a dynamic rolling radius of the tires, and to estimate, recursively, the dynamic rolling radius of the tires from the measurement of the angular velocity of the wheels and the longitudinal velocity estimated by the recursive slope estimator.

[0049] At least part of the methods according to the invention can be implemented by computer. Accordingly, the present invention can take the form of an entirely hardware embodiment, an entirely software embodiment (comprising firmware, resident software, microcode, etc.), or an embodiment combining software and hardware aspects, all of which can be collectively referred to herein as a "circuit," "module," or "system." Furthermore, the present invention can take the form of a computer program product embedded in any non-transient recording medium comprising computer-readable program code for implementing the above method.

[0050] A tangible or non-transient medium may include a storage medium such as a hard disk drive, a magnetic tape device or a semiconductor memory device and the like. Brief description of the drawings

[0051] Other objects, features and advantages of the invention will become apparent from the following description, given solely by way of non-limiting example, and made with reference to the accompanying drawings in which:

[0052] [Fig-1] illustrates a motor vehicle-type physical system, of which the longitudinal behavior defined by Newton's second law (equilibrium of forces) can be reduced to a more or less complex model;

[0053] [Fig.2] illustrates a vehicle control system according to embodiments;

[0054] [Fig.2A] illustrates the use of the angular acceleration of the wheels wx(k) generated by a polynomial smoother in the recursive estimation of the slope of the road a(k) by a slope estimator;

[0055] [Fig.2B] illustrates a parallel coupling of a slope estimator with a radius estimator;

[0056] [Fig.2C] illustrates series couplings of a slope estimator and a radius estimator;

[0057] [Fig.3] illustrates, using a flowchart, the steps of a supervisory process implementing the control device, according to embodiments;

[0058] [Fig. 3A] illustrates, using a flowchart, the steps of a coupled estimation process for the slope of a road, the longitudinal speed of a vehicle, and a dynamic rolling radius of the vehicle; and

[0059] [Fig.4] illustrates a computer hardware architecture of an estimation system or device according to embodiments.

[0060] Detailed description of at least one embodiment

[0061] Certain states or parameters of a vehicle cannot be measured directly by sensors, but through recursive estimators.

[0062] The term “vehicle” means any type of wheeled motorized vehicle regardless of its energy source (thermal, electric or other), such as a light vehicle or car, a utility vehicle, a truck, a military vehicle, but also an airplane, an amphibious vehicle, etc.

[0063] Recursive estimators rely on a model of the underlying physical system, typically based on Newton's laws, to determine the mass of a vehicle from the engine torque, longitudinal acceleration measurements of the vehicle and angular velocity measurements of the vehicle's wheels.

[0064] To improve the estimates, the angular acceleration intrinsically generated by a polynomial smoother preprocessing the angular velocity measurements is reused in the recursive estimation of the road slope and the vehicle's longitudinal speed. Furthermore, to obtain observable models without unnecessary simplifications, this recursive slope estimator is coupled with a recursive tire dynamic radius estimator so that their estimated longitudinal speed and estimated dynamic radius are respectively used in the estimates of the other estimator. Estimating the dynamic radius allows the use of more complex models; the accuracy of the vehicle mass estimation is enhanced by the accuracy of the road slope estimation.

[0065] Figure 1 represents a physical system of the motor vehicle type, whose longitudinal behavior, defined by Newton's second law (equilibrium of forces), can be reduced to a more or less complex model. The forces involved include:

[0066] the inertial force Finertie(t) whose value is (mv(t)+mpt(t)).ax(t), with mv(t) the mass of the vehicle and mpt(t) the equivalent mass of the rotating parts whose value is mpt(t)=Jtot / rdyn(t)2, with Jtot the moment of inertial of the mass of the components of the powertrain and rdyn(t) the dynamic rolling radius of the tires,

[0067] the traction force Ftract(t) whose value is / \ , where Meng(t) ^T“ is the engine torque, y(t) is the transmission ratio, and qtot(t) is the mechanical or motor transmission efficiency (typically fixed between 0.95 and 0.99). The tractive force can be negative in the case of engine braking. In this case, it is a braking force.

[0068] the aerodynamic force or resistance Faero(t) whose value is 1 z / ,v2. , where even is the ambient air density, vx(t) is the velocity longitudinal axis of the vehicle, vwind(t) is the longitudinal wind speed, Af is the frontal area of ​​the vehicle and cd is the aerodynamic coefficient of the vehicle.

[0069] the rolling resistance Fron(t) whose value is mv(t).g.(cr 0+cri.vx(t)+cr4.vx(t)4), with g the gravitational constant, cr 0, crb cr4 rolling resistance coefficients. Other formulations of the form f(vx(t), a(t)) can be used alternatively, for example mv(t).g.cr.cos(a(t)), and

[0070] the gravitational force Fgrav(t) whose value is mv(t).g.sin(a(t)), with and a(t) the angle of the slope (between -90° and 90°).

[0071] The balance of forces leads to Finertie(t) = Ftract(t) + Faero(t) + Fron(t) + Fgrav(t), and thus to the following model (1) which however neglects all random disturbances and unaccounted losses:

[0072] J _ rdyJf) +m v (0£sin(«(0 )

[0073] +mv(t)g(c^ + Cr}Vx(J) +Cr4Vx(t)4)+^pa^)2

[0074] This complete longitudinal model, although offering an accurate representation of the system in Figure 1, is very difficult to use in practice, particularly because it is not observable (the number of unknown values ​​is greater than the known values ​​Meng{t), (AP^, Af). The random nature of the wind, for example, is very U1) difficult to measure, and some vehicle parameters, such as rotational inertias, are vehicle-specific and often unknown to the user.

[0075] A typical first simplification of the model consists of neglecting the effects of the headwind v M because of an impact estimated to be negligible on the inertia total vehicle mass, as well as the inertial masses of the rotating components of the powertrain relative to the total vehicle mass. Furthermore, since the longitudinal speed of the vehicle is generally unknown, it is usually calculated from the angular velocity of the vehicle's wheels, vx(t): vx(t) = Wx(t) Xvn(0 •

[0076] This leads to the following model (2):

[0077] ( / )^(0 + mv(t)gsin(a(t))

[0078] + m y (t)g(c r0 + c r । v x (t) + c r4 v x ( / ) 4 ) + j / 4h (t) 2 -

[0079] This model is not observable at first glance, since rdyn, mv, ax, a, vx and the coefficients c are unknown.

[0080] Some embodiments of the present disclosure make this model observable, whereas known techniques consider it preferable to simplify the model further.

[0081] For example, variations in tire radius are neglected: the dynamic rolling radius Δd is considered equal to the nominal (constant) tire radius rnom. Consequently, the longitudinal acceleration of the vehicle can be approximated as: ax(t) ~ wx(t)znom. This results in the following model (3):

[0082] ~ mv(t)(t)rnom + mv(t)gsin(a(t))

[0083] + m v (t)g(e) r0 +c r ] w x ( t ) r nom + c r4 (m x ( t ) r no J 4 ) +1 p aj( t ) r nom )“

[0084] This model is observable and therefore usable. Only mv, a and the coefficients c are unknown.

[0085] Another simplification consists of assimilating the road slope as perceived by the accelerometer `imu(O)` to the actual road slope `a(O)`, thus neglecting any error. Also, if the acceleration `xjmu(O)` measured by the accelerometer is related to the true acceleration of the vehicle by the following formula: `x,IMU(*)` `ax(O) + g-Sin(alMU(O))` with `sin(aiMU(*))` the gravitational acceleration which corrupts the accelerometer measurements and which is a function of the time-varying pitch angle of the accelerometer, then the simplification results in `(t)` ≈ `g.sin()`, hence the following model (4):

[0086] Fnom m? ( t ) JMU (t) + m v (t) g(c r0 + c d w x ( t )^name 3" (Df*name) )

[0087] +( t )rMm)“

[0088] This simpler model (only mv and the coefficients c are unknown) is also observable and allows simplified calculations to obtain the mass of the vehicle.

[0089] To solve these different models, recursive estimators are generally used, which evaluate the unknown parameters or states at times (or "recursions") that correspond approximately to the times of acquisition of observable data by sensors. These unknown parameters or states are thus considered to be estimated in real time.

[0090] One aspect of the present disclosure makes model (2) observable by coupling a recursive estimator of road slope and vehicle longitudinal speed with a recursive radius estimator estimating the dynamic rolling radius of the vehicle's tires. The coupling consists in the estimators reciprocally providing their estimates for the other estimator's estimates. Thus, the recursive slope estimator estimates the road slope and vehicle longitudinal speed from measurements of longitudinal acceleration and wheel angular velocity and the dynamic rolling radius of the tires estimated by the recursive radius estimator, while the recursive radius estimator estimates the dynamic rolling radius of the tires from the wheel angular velocity measurement and the longitudinal speed estimated by the recursive slope estimator.

[0091] Although it is already known in the literature to use recursive estimators of slope and longitudinal velocity, these are not coupled with a dynamic rolling radius estimator to achieve a simultaneous or "dual" estimation.

[0092] This simultaneous estimation from direct sensor measurements (only longitudinal acceleration and wheel angular velocity) allows us to obtain the following states: rdyn(t), ax(t), a(t), vx(t), as will be illustrated later. Therefore, model (2) becomes observable. A mass mv(t) of the vehicle can be estimated from the values ​​of these states.

[0093] Another aspect—whether separate or combined—of the present disclosure improves the estimation of the road slope—and consequently that of the vehicle mass mv(t) from model (2) or (3), for example—by providing that the recursive slope estimator estimates the road slope from measurements of longitudinal acceleration and wheel angular velocity. The wheel angular acceleration is intrinsically evaluated by a polynomial smoother that preprocesses the angular velocity measurements to remove sensor noise. Indeed, the polynomial nature of the smoother (generally a polynomial of degree >1) leads to the calculation of one or more derivatives of the smoothed measurements, in this case, the measured angular velocity.

[0094] Figure 2 illustrates a vehicle control system 200 which allows, from measurements taken on the vehicle, to estimate for example the mass mv of the vehicle and to take appropriate measures accordingly.

[0095] The system 200 includes one or more on-board sensors 210 and a control device 220.

[0096] The sensors 210 are configured to acquire, at time t, measurements or "observations" o(t) of the vehicle. In the example of [Fig. 1], an onboard torque meter measures the vehicle's motor torque Meng(t), an accelerometer measures the vehicle's longitudinal acceleration ax>IMU(t), and gyroscopes measure the angular velocities oex(t) of the wheels (which are considered identical for all wheels). Beyond this example, any type of sensor can be used.

[0097] The control device 220 can be wholly or partially integrated into the vehicle for real-time estimation, or external to the vehicle, in which case the acquired measurements (Meng(t), ax>IMU(t), cox(t)) and / or the estimations performed in the vehicle are transmitted in batches to one or more servers—for example, in the cloud—to perform subsequent estimation. Signals for implementing the appropriate measures can be transmitted back to the vehicle. Transmissions can be carried out via wired (e.g., Ethernet network) or wireless (e.g., Wi-Fi network or Bluetooth link—trade names) connections to / from the server. For example, the control device 220 can to be implemented in a remote processing server that manages a fleet of vehicles and transmits commands acting on them in return.

[0098] The control device 220 includes a communication interface 230, a monitoring device 240 and a control unit 250.

[0099] The communication interface 230 obtains the measurements Meng(t), ax, iMu(t), and œx(t) from the sensors 210. Typically, for an on-board control device 220, the interface is a serial interface on the vehicle's CAN (Controller Area Network) data bus, from which these measurements are retrieved. For an external control device 220, the interface can be any wired or wireless communication interface (e.g., a mobile phone modem).

[0100] The monitoring device 240 performs estimations of unknown states and parameters of the implemented model, for example, model (2) above. This monitoring device implements particular embodiments of this disclosure.

[0101] This may be one or more computer processors executing one or more computer programs for the implementation of the treatments and estimators described below.

[0102] In the scenario proposed as an example, the monitoring device 240 allows the estimation of the mass mv(t) of the vehicle.

[0103] The control unit 250 uses the state(s) and parameter(s) estimated by the monitoring device 240 to generate a signal to the vehicle for the implementation of one or more appropriate measures. In the example scenario, the estimated mass mv(t) of the vehicle allows for the evaluation of the vertical load on each tire, tire wear, tire grip, tire rolling resistance, or even the prediction of the battery range of electric vehicles (EVs). Appropriate measures can be taken, such as activating an overload indicator, indicating tire wear status, adjusting a vehicle electric range indicator, or adjusting the vehicle's routing to charging stations.

[0104] In detail, the monitoring device 240 of the figure includes a preprocessing unit 241, one or more polynomial smoothers 242, a recursive slope estimator 243, a recursive radius estimator 244 and a mass estimator 245.

[0105] The pre-processing unit 241 is optional. However, it performs processing on the measurements Meng(t), ax>1MU(t), and œx(t) to make them synchronous. Indeed, each sensor has its own acquisition frequency and the sensors are not synchronized with each other.

[0106] The preprocessing may include an interpolation function, so that the measurements are made synchronous: a measurement is interpolated (if necessary) at each interpolation time. A clock is embedded for this purpose. The interpolation step or frequency (when necessary) defines the recursion step or frequency, since the estimators will perform an estimation at each recursion from the measurements interpolated for that recursion. Hereafter, the index 'k' refers to the recursion 'k' and corresponds to an interpolation time tk. The recursion frequency can be defined by an operator of the device 220. In the example of the vehicle in [Fig. 1], a recursion period between 10 milliseconds (ms) and 500 ms is compatible with off-the-shelf sensors and real-time operation.

[0107] The preprocessing may also include a function for selecting acquired measurements, which serves to select measurements consistent with assumptions made in the choice of the applied model, and thus remove outliers. Typically, model (2) of the proposed scenario models longitudinal vehicle behavior. Measurements acquired on non-longitudinal behaviors can therefore be excluded, since the resulting mass mv estimate would not be reliable.

[0108] By way of example, all or part of the following criteria can be evaluated at time tk to retain the measurements Meng(k), ax>1MU(k), cox(k) corresponding to that time:

[0109] - the yaw rate of the vehicle (which can be calculated directly from the (wheel rotation speed or using a dedicated yaw rate sensor) must remain below a threshold defined by the operator,

[0110] - the acceleration aXjlMU(k) of the vehicle must be positive and greater than a defined threshold by the operator. This condition aims to guarantee full excitation of the input data to the system, an assumption for the proper functioning of RLS (Recursive Least Squares) estimators.

[0111] - the speed cox(k) of the vehicle must be greater than a threshold defined by the operator,

[0112] - the vehicle must not operate in four-wheel drive mode. Also, if the If the vehicle is equipped with all-wheel drive or four-wheel drive, the secondary torque (of the second set of wheels) must remain below a threshold defined by the operator.

[0113] A polynomial smoother 242 allows the noise of the sensor used to be removed from a measurement signal Meng(k), aXjlMU(k), or cox(k). It is therefore also a pre-processing of the measurement signals. In one embodiment, the polynomial smoothing of the measurement signals can be performed before the resynchronization 241. In practice, each measurement signal is smoothed by a dedicated polynomial smoother. Also, block 242 groups, in the example, three polynomial smoothers, one for each of the signals Meng(k), ax>IMU(k) or œx(k).

[0114] The polynomial smoothing of each measurement signal can include polynomial regression implemented using a polynomial Kalman smoother, or "PKS" (for "Polynomial Kalman Smoother"). The PKS is based on the Savitzky-Golay filter, as described in the publication 'Recursive Generalized Total Least Squares with Noise Covariance Estimation' (S. Rhode et al., 2014). The PKS works by sliding a fixed-size window over the measurements of the processed signal and fitting a polynomial to the points of this window. The value of the polynomial at the center point of the window is then considered the smoothed value. This process is repeated for each measurement, producing a smoothed signal. The degree of the polynomial is greater than 1, which allows, during calculations, to evaluate the derivative of the input measurements, and in particular the angular acceleration of the vehicle's wheels (üx(k) by differentiation of the angular velocity of the wheels during the smoothing of cox(k).

[0115] Algorithm 5 of the publication S. Rhode et al. can be used, where Bt are the measurements at the recursion 't' of the signal to be smoothed, and x represents the corrected (smoothed) output measurements. The width of the window 'w' is predefined, for example, between 4 and 100, with Wi and wr being the left and right half-windows composing the window w. The matrix C is defined by equation (13) of this publication. The forgetting factor X (between 0 and 1) and the dimension of the polynomial (polynomial order) used are prefixed. The noise covariance matrix P and the state transition matrix A are of the same dimension axa, where a is the polynomial order minus one. P is initialized with [3x1] where IeRaxa is the identity matrix of dimension a and [3] is a regularization parameter defined in the range [0; 100]. A is also defined in the publication S. Rhode et al. to equation (11).

[0116] Each smoother can be adjusted according to the noise level and frequency dynamics of the signal to be smoothed. For example, the acceleration signal aXjlMU(k) can be considered very noisy and dynamically strong, in which case the PKS smoother can have a polynomial order of four and a window size of 10. The angular velocity signal cox(k), on the other hand, can be considered less noisy and dynamically weaker, in which case the PKS smoother can have a polynomial order of two and a window size of 3.

[0117] As an alternative to using the 242 polynomial smoother to generate Mx(k), an angular velocity differentiator can be used to obtain the angular acceleration of the vehicle's wheels (dx(k)). The angular velocity differentiator can simply calculate the difference Δcox between two successive velocity measurements angular of successive wheels cox(t-1 ) and cox(t), and divide it by the time elapsed ôt between the two measurements: ôcox / ôt.

[0118] The recursive slope estimator 243 receives, as input, the smoothed signals ax>1MU(k), cox(k), and jjx(k). This estimator estimates the slope a(k) of the road (the estimate is denoted a(k)) from these measurements of longitudinal acceleration and angular wheel velocity, and from the angular wheel acceleration generated by the smoother 242. In particular, the recursive slope estimator 243 can be based on the following underlying models:

[0119] for the state evolution system:

[0120] vx(t)= vx(tl)+ax(t).At,

[0121] ax(t)= ax(tl)

[0122] g.sin(a(t))=g.sin(a(tl))

[0123] and for the measurement system:

[0124] vx(t)=œx(t).rdyn(t),

[0125] ax(t)=mx(O.rdyn(t),

[0126] ax>IMU(t)=ax(t)+g. sin(a(t)).

[0127] In a preferred mode, the recursive slope estimator estimates the slope of the road from further the dynamic rolling radius of the tires k) estimated by the recursive radius estimator 244 as described below.

[0128] Typically, the recursive slope estimator 243 implements a linear Kalman filter or LKF (for "Linear Kalman Filter") to estimate, in addition to the slope a(k) of the road, a real longitudinal acceleration ax(k) of the vehicle (the estimate is denoted âx(&)) and a real longitudinal speed vx(k) of the vehicle (the estimate is denoted

[0129] The state vector to be estimated x is composed of the following states vx ( k ), âx ( k ), g.sin(â ( k ) ), for example:

[0130] x _ as■ n ( g ( ) ]r

[0131] The corresponding measurement vector y is defined by:

[0132] A .af J - dyn , " x r dyn , has x j MU ^

[0133] The state transition matrix F based on the above state evolution system taking into account the vector x, and the model noise covariance matrix Q for filter prediction are fixed as follows:

[0134] F = '1 0 .0 AT 0' and Q = 0 0 0' 0 1 0 0 1. 0 0 ^3.

[0135] with At the recursion step (the time in seconds between two successive estimates, which must take into account the possible deletion of measures during preprocessing 241) and ^2, and y3 of the regularization parameters defined in the range [0; 10].

[0136] The measurement modeling matrix H based on the above measurement system taking into account the vector y, and the measurement noise covariance matrix R for filter correction are fixed as follows:

[0137] '1 0 0' and '*1 0 0' H - 0 1 0 0 K2 0 .0 1 1. 0 h 0 *3.

[0138] with ^2, and ^3 regularization parameters defined in the range [0 ; 10]. Note here that the measurement modeling matrix H includes an additional row compared to the matrices used classically, due to the consideration of the acceleration (Vx(k) in the measurement system: ax(t)= wx( / ).rdyn(t).

[0139] The covariance matrix P (internal parameter of the filter) is initialized to [30xI where IeR3x3 is the identity matrix of dimension 3 and [30 is a regularization parameter defined in the range [0 ; 100]. The estimated vector x is initialized to (0,0,0).

[0140] Figure 2A illustrates the use of the angular acceleration of the wheels wx(k) generated by the polynomial smoother PKS 242 in the recursive estimation of the slope of the road â(k) by the slope estimator 243.

[0141] The recursive estimator of radius 244 receives, as input, the smoothed signal cox(k) as well as the estimates of the longitudinal velocity $x(k). This estimator estimates the dynamic rolling radius rdyn(k) of the tires (the estimate is denoted (k)) from the measurement of the angular velocity of the wheels and the longitudinal velocity estimated by the slope estimator 243. In particular, the recursive estimator of radius 244 is an estimator based on the following underlying model: vx(t)=cox(t).rdyn(t).

[0142] The use of estimator 244 allows for a noise-robust (sensor, smoothing) and therefore accurate estimation.

[0143] By reciprocally using an estimate of the other estimator, the recursive estimator of slope 243 and recursive estimator of radius 244 are coupled to perform simultaneous, higher quality estimates.

[0144] In one embodiment, the coupling is performed between two successive recursions. In this case, the longitudinal velocity estimate vx(k-1) is used by the recursive radius estimator during recursion k to estimate the dynamic radius. Conversely, the dynamic radius estimate rjynÇk- 1) is used by the recursive estimator with slope 243 during recursion k to estimate the slope a(k) and the actual longitudinal acceleration ax(k) and actual longitudinal velocity vx(k). With this coupling between two successive recursions, the inputs of the estimator 244 for the recursion k are vx(k-1) and ojk), while those of the estimator 243 are ( k -1 ), aXjlMU(k), cox(k) and cox(k ). This coupling is illustrated for example in [Fig.2B],

[0145] In another embodiment, estimators 243 and 244 are put in series: the first uses the (k-1) estimate of the second when estimating for the current recursion k, while the second (downstream in the series) uses this k estimate of the first during its recursion k. This serial coupling is illustrated for example in [Fig.2C].

[0146] In the first configuration, the recursive estimator of radius 244 is upstream of the recursive estimator of slope 243. Estimator 244 receives vx(k-1) from the previous recursion, calculates ïdyn(k) which is passed to estimator 243. The latter calculates â(k) and vx(k) as a function of (k).

[0147] In the second configuration, the recursive estimator with slope 243 is upstream of the recursive estimator with radius 244. Estimator 243 receives îdyn(k-1) from the previous recursion, calculates a(k) and vx(k). vx(k) is passed to estimator 244, which calculates rdyn(k) as a function of vx(k).

[0148] In yet another embodiment, coupling is performed in real time, meaning that the current estimation (at recursion k) is used by the other estimator during the same recursion. Typically, the recursive slope estimator and the recursive radius estimator perform iterative estimations during an estimation recursion, meaning, for example, that they run their respective filters several times (iterations) using the same input measurements, until convergence (for example, their estimates no longer change) or the expiration of a time (based, for example, on the time Δt between two successive measurements). The real-time coupling provides that the two estimators exchange their previous iterative estimations (obtained at iteration i-1) to perform the next iterative estimation (iteration i).

[0149] For example, the recursive slope estimator 243 iteratively estimates the slope of the road a and the longitudinal speed vx at recursion k, i.e. several estimates cç(k), vr.(k) are made at several iterations i during recursion k. The estimation of Oj(k) and \rxi(k) is based on the measures aXjlMU(k), cox(k) and a)x(k) as well as on the estimate rdynM(k) provided by the recursive radius estimator 244. ïdyn.o(k) =Tdyn(k-1) to allow the initialization of recursion k.

[0150] Symmetrically, the recursive estimator with radius 244 iteratively estimates the dynamic rolling radius of the tires rdyn at recursion k, that is, several estimates rayrû(k) are made at several iterations i during recursion k. The estimation of rdynj(k) is based on the measure cox(k) as well as on the estimate Vxyj ( k ) provided by the recursive estimator with slope 243.

[0151] The recursive estimator of radius 244 can implement a recursive least squares or RLS filter based on the underlying model above.

[0152] For example, the parameter vector to be estimated is defined by r ]r, the vector X — p dyn] regression by a = [mx] and the vector of measures by y = estimated by estimator 243. The forgetting factor is fixed to a value between 0 and 1; the covariance matrix P (internal parameter of the filter) is initialized to [3iXl where theR1 x 1 is the identity matrix of dimension 1 and [3i is a regularization parameter defined in the range [0; 100]; and the estimated vector x is initialized to 0.

[0153] The recursive mass estimator 245 receives as input the signal Meng(k) (optionally smoothed by a smoother 242), the slope estimates â(k), longitudinal acceleration estimates âx(Æ), and longitudinal velocity estimates vx(k) from estimator 243, and the dynamic radius estimates L / v„(k) from estimator 243 when calculated. This estimator estimates a mass mv(k) of the vehicle (the estimate is denoted ^(k)) from the estimated road slope and, where applicable, from an estimated dynamic rolling radius of the tires.

[0154] More precisely, the recursive mass estimator 245 can have as its underlying model the model (2) or “equivalent”, i.e., when, for example, the term mv(t).g.(cro+cri.vx(t)+cr4.vx(t)4) is substituted by another expression for the rolling resistance Fron(t). Also, the estimator 243 estimates, in addition to the mass mv(A:) of the vehicle, the rolling resistance coefficients Cjo, cr[, cr4, and the aerodynamic coefficient of the vehicle cd.

[0155] The recursive mass estimator 245, for example, implements a recursive least squares RLS filter based on the underlying model (2) or equivalent.

[0156] For example, the parameter vector to be estimated is defined by x=[mv, mv, m;.c,n. mv .cri, mv.cr4, ]T, the regression vector is defined by a=[ax, g.sin(a), g, g.vx, 2”ajr td g.vx4, vx2], and the output measure of the regression is defined by . The factor rdyn The forgetting parameter is set to a value between 0 and 1; the covariance matrix P is initialized to P2XI where the R6 x 6 is the 6-dimensional identity matrix and

[32] is a regularization parameter defined in the range [0; 100]; and the estimated vector x is initialized to the 6-dimensional vector 0.

[0157] Fig. 3 illustrates, using a flowchart, the steps of a supervisory process implementing the control device 220.

[0158] Before executing the algorithm in real time, the different estimators are initialized in step 300.

[0159] At step 305, the device obtains, via interface 230, measurements of longitudinal acceleration of the vehicle ax>1MU(t) and angular velocity of the vehicle wheels cox(t), as well as measurements of the motor torque Meng(t), acquired by sensors 112.

[0160] The following steps are performed at each recursion time.

[0161] In step 310, these measurements are preprocessed by units 241 and 242 to obtain synchronous and smoothed measurements: ax>1MU(k), cox(k) and Meng(k). At this time, measurements of the angular acceleration of the vehicle's wheels Mx(k) are generated by differentiating the angular velocity of the wheels.

[0162] In step 315, the road gradient (Uk), as well as the vehicle's longitudinal speed vx(k) and longitudinal acceleration ax(k), are determined by the gradient estimator 243 from ax > 1MU(k), cox(k) and u)x(k), and from the dynamic rolling radius estimated by the radius estimator 244, for example in the previous recursion k-1). Also, simultaneously with step 315, the radius estimator 244 estimates in step 320 the dynamic rolling radius of the tires from The angular velocity of the wheels cox(k) and the longitudinal velocity estimated by the slope estimator 243, for example, at the previous recursion vx(k-1). Steps 315 and 320 are followed by step 325, during which the recursive mass estimator 245 performs the estimates of m, ( / < ), cr0, cri, cr4, and cd, from a(k), ax(k), vx(k), Meng(k), and also the estimated dynamic rolling radius (k). Note that implementing this coupling without using the smoother from step 310, but with an angular velocity differentiator, for example, can be considered, while still maintaining an improved final estimate of the vehicle mass.

[0163] Step 325 can be followed by a step 330 where the estimated mass (m^k) is used by the control unit 250 to generate a control signal to the vehicle. As given above by way of example, such a signal can consist of activating a tire overload indicator, indicating tire wear, adjusting a vehicle electric range indicator, or routing the vehicle to charging stations.

[0164] The vehicle control system described above offers the advantage of an improved estimation of the actual road slope by taking into account the angular acceleration of the wheels, which is naturally estimated in the conventional smoothers used during measurement preprocessing. It also offers the additional advantage of a joint and high-quality estimation of the vehicle's longitudinal speed and dynamic rolling radius, making complex models observable. This results in a higher-quality estimation of the vehicle's mass using such now-observable complex models.

[0165] Fig. 3A illustrates, using a flowchart, steps of a coupled estimation process for the slope of a road, the longitudinal speed of a vehicle and a dynamic rolling radius of the vehicle.

[0166] Before executing the algorithm in real time, the different estimators are initialized in step 370.

[0167] At step 375, the device obtains, via interface 230, measurements of longitudinal acceleration of the vehicle ax>1MU(t) and angular velocity of the vehicle wheels œx(t), acquired by sensors 112.

[0168] At step 380, the road slope a(k) and the longitudinal speed of the vehicle vx(k) are determined by the slope estimator 243, from ax > 1MU(k), cox(k) and the dynamic radius, for example that of the previous recursion r <ivn(k- 1) estimée par l’estimateur de rayon 244.

[0169] At step 385, the radius estimator 244 estimates the dynamic rolling radius of the tires rjv / z( &) from the measurement of the angular velocity of the wheels cox(k) and the longitudinal velocity estimated by the slope estimator 243, for example at the previous recursion vv(k-1 ) .

[0170] As mentioned above, variants may provide for carrying out step 380 before step 385 (or vice versa), in which case the estimate carried out by the upstream estimator during recursion k is provided for estimate k by the downstream estimator.

[0171] In other variants, the two estimators exchange, during recursion k, iterative estimates that they perform.

[0172] Fig. 4 illustrates a computer hardware architecture of a control device 220 according to embodiments.

[0173] The device 400 includes a communication bus 401 to which the following are preferably connected: - one or more central processing units 402, such as one or more CPU processors and / or one or more microprocessors; - a 403 storage memory, of type ROM and / or hard disk and / or flash memory, for the storage of computer programs intended to implement all or part of the operations described above; - 404 RAM, or even video RAM (VRAM), for storing the executable code of computer programs as well as registers adapted to record variables and parameters necessary for their execution; - a 405 communication interface connected to a network (e.g. CAN bus or Wifi network) in order to communicate with external equipment, e.g. external 210 sensors or any other unit in the vehicle (especially when part of the estimation is carried out outside the vehicle); - One or more 406 I / O inputs / outputs allowing an operator to interact with computer programs, both during configuration and operation. Typically, the inputs / outputs may include a screen serving as a graphical interface with the operator and displaying the value of unknown parameters, and / or a speaker to output audio content and / or a keyboard or other pointing device allowing the operator to interact.

[0174] Preferably, the communication bus 401 ensures communication and interoperability between the various elements included in or connected to the device 400. The bus representation is not limiting and, in particular, the central processing unit can be used to communicate instructions to any element of the computer device 400 directly or by means of another element of the computer device.

[0175] The executable code stored in memory 403 can be received via the communication network, through interface 405, for storage there prior to execution. Alternatively, the executable code is not stored in non-volatile memory 403 but can be loaded into volatile memory 404 from a remote server via the communication network for direct execution.

[0176] The central processing unit 402 is preferably adapted to control and direct the execution of instructions or parts of software code of the computer program(s). Upon power-up, the program(s) stored in non-volatile memory 403 or on the remote server are transferred / loaded into the main memory 404, which then contains the executable code of the program(s), as well as registers for storing the variables and parameters necessary for the implementation of the invention.

[0177] Of course, the invention is by no means limited to the variant embodiments described above, the person skilled in the art being able in particular to isolate or freely combine the aforementioned characteristics, or to substitute equivalents for them.

Claims

Demands

1. Supervision device (220) of a vehicle operating on a road, the device comprising: a communication interface (405, 230) for obtaining, from one or more sensors (210), measurements of longitudinal acceleration of the vehicle (ax>1MU) and angular velocity of the vehicle's wheels (œx), one or more processors (402) implementing: - a recursive slope estimator (243) estimating a slope of the road (a) and a longitudinal speed of the vehicle (vx), from the measurements of longitudinal acceleration and angular velocity of the wheels and a dynamic rolling radius of the tires (rdyn), and - a recursive radius estimator (244) estimating the dynamic rolling radius of the tires (rdyn) from the measurement of angular velocity of the wheels (œx) and the longitudinal speed (vx) estimated by the recursive slope estimator.

2. Supervisory device (220) according to claim 1, wherein the recursive slope estimator (243) estimates the slope of the road and the longitudinal speed of the vehicle, further from an angular acceleration of the wheels (0½) obtained from the angular velocity measurements of the vehicle's wheels.

3. Supervisory device (220) according to claim 1 or 2, wherein the recursive slope estimator (243) comprises a linear Kalman filter estimating the road slope (a), a longitudinal acceleration of the vehicle (ax) and a longitudinal speed of the vehicle (vx).

4. Supervisory device (220) according to any one of claims 1 to 3, wherein the underlying model of the recursive slope estimator (243) includes the following state evolution system: vx(t) = vx(tl) + ax(t).At, ax(t) = ax(tl) g.sin(a(t)) = g.sin(a(tl)) and the following measurement system: vx(t) = cox(t).rdyn(t), ax(t) = ü\(O.rdyn(t), ax,iMu(t) = ax(t) + g.sin(a(t)), where ax is an estimated longitudinal acceleration of the vehicle, g the gravitational constant, At the time step between two estimation recursions, and an angular acceleration of the vehicle's wheels.

5. Supervisory device (220) according to any one of claims 1 to 4, wherein the recursive slope estimator (243) estimates the road slope at recursion k from the dynamic rolling radius of the tires estimated by the recursive radius estimator at recursion k-1, and / or the recursive radius estimator estimates the dynamic rolling radius of the tires at recursion k from the longitudinal speed estimated by the recursive slope estimator at recursion vi

6. K 1. Supervisory device (220) according to any one of claims 1 to 4, wherein the recursive slope estimator (243) and the recursive radius estimator (244) perform iterative estimations during an estimation recursion and exchange their previous iterative estimations to perform the next iterative estimation.

7. Supervisory device (220) according to any one of claims 1 to 6, wherein the underlying model of the recursive radius estimator (244) includes: vx(t)=œx(t).rdyn(t).

8. Supervisory device (220) according to claim 7, wherein the recursive radius estimator (244) comprises a recursive least squares filter whose parameter vector is r 1^, the regression vector is a — [ wx ] and the measures are =

9. Supervisory device (220) according to any one of claims 1 to 8, further comprising a recursive vehicle mass estimator (245) estimating a vehicle mass (mv) from the estimated road slope (a), the estimated longitudinal speed (vx) and the estimated dynamic rolling radius of the tires (rdyn).

10. System (200) comprising a vehicle monitoring device (230) according to any one of claims 1 to 9 and a vehicle control unit (250) for controlling an on-board function from the estimates of the monitoring device.

11. A method for monitoring a vehicle operating on a road, the method comprising the following steps: obtaining (300), via a communication interface (405, 230), measurements of the vehicle's longitudinal acceleration (aXjlMU) and speed angular velocity of the vehicle wheels (œx) acquired by one or more sensors (210), estimate (310), recursively, a road slope (a) and a longitudinal vehicle speed (vx), from the longitudinal acceleration and angular velocity measurements of the wheels and a dynamic rolling radius of the tires (rdyn), and estimate (315), recursively, the dynamic rolling radius of the tires (rdyn) from the angular velocity measurement of the wheels and the estimated longitudinal speed (vx).

12. Computer program product embedded in a non-transient recording medium comprising computer-readable program code for implementing the method according to claim 11.

Images