Method for determining a control law to control at least one actuator

Abstract

Method for Determining a Control Law to Control at Least One Actuator. The invention relates to a computer-implemented method for determining a control law (Lc) to control at least one actuator (20, 30) for a system, said method comprising the following steps: Provision (100) of a physical model (M10) of the system; Provision (200) of a driven model (M20, M30) of the dynamics of at least one actuator (20, 30); Determination (300) of a hybrid model (MH10) of the system by concatenating the physical model (M10) of the system and the model (M20, M30) of the dynamics of at least one actuator (20, 30); Determination (400) of the control law (Lc) from the hybrid model (MH10). (Figure 1)

Description

Title of the invention: Method for determining a control law to control at least one actuator

[0001] The present invention relates to the field of mechanics, and in particular to the automotive sector, and more specifically concerns a method for determining a control law to control at least one actuator for a system, for example, a vehicle. The invention aims to take into account the actual dynamics of each actuator in order to ensure, on the one hand, efficient control of the actuators and therefore of the vehicle in which they are implemented, and on the other hand, to guarantee the energy efficiency of the control system as a whole.

[0002] In what follows, the invention will be described in its automotive application. However, as will become clear from reading the description of the invention, the invention can also be applied in other technical fields where it is necessary to electronically control an actuator, as is the case, for example, in systems such as airplanes, trains, electric bicycles, motorcycles, lifting systems, and household appliances. Furthermore, the application of the invention is not limited to transportation systems. It can also be applied in the field of manufacturing, since there too, different actuators with different dynamics must be electronically controlled to achieve a target or reference trajectory.

[0003] An actuator is a component that acts on a system in such a way as to modify its state or behavior. In the automotive field, an actuator can be defined as a motor used by driver assistance software to assist the driver, the system then being the vehicle. Vehicles currently under development, and even more so future vehicles, are and will be increasingly equipped with actuators to control their mobility: front and rear steering actuators (4RD for 4-wheel steering, or 4WS in English for 4-wheel steering), braking actuators, various thermal and / or electric propulsion actuators, and in-wheel motor actuators (known by the abbreviation IWM, from the English term In-Wheel Motors) are cited as non-limiting examples.

[0004] Each actuator exhibits its own unique dynamics, resulting from the technical solution implemented to perform its intended function. An actuator's dynamics correspond to the time it takes to execute a command and its ability to reach the given setpoint. For example, for a wheel, the setpoint is the steering angle, and the dynamics consist of the time required to reach this target steering angle and the Accuracy relative to the setpoint. In most of the various control laws used in automobiles, the actuators are considered to be ideal for the sake of simplicity. Therefore, when designing a control system, it is assumed that this control will be applied instantaneously, without amplitude or phase distortion.

[0005] This is not entirely accurate, but the loss of efficiency created by this simplifying assumption has not been considered significant until now. However, in the current context of the energy crisis, and even more so in the future, vehicle control based on multiple actuators will need to be energy-efficient. This is all the more possible given the existence of control assignment techniques capable of guiding the choice of actuators used towards the most efficient actuators in a vehicle (taking into account the efficiency of movement and the energy required), particularly when the vehicle is equipped with more actuators than there are degrees of freedom to control (over-actuated system).

[0006] In existing control solutions, the control laws are designed without taking into account the actual dynamics of the actuators. These laws correct the vehicle's behavior to follow a given setpoint or trajectory by addressing, through closed-loop control, the inaccuracies introduced by the actuator dynamics.

[0007] It is therefore understood that this correction is made at the expense of the overall electrical energy required to carry out the mission. This correction, although effective, is all the more significant (and therefore energy-intensive) because the behavioral models used to design the correctors are developed without taking into account the dynamics of the actuators.

[0008] The invention aims to overcome all or part of the problems mentioned above by proposing a control law design for actuators which takes into account the actual dynamics of each actuator contributing to the control of the vehicle, from the control law design phase.

[0009] The invention thus optimally solves the vehicle control problem from an energy perspective by introducing dynamic models of all the actuators onboard the vehicle into the control models. This results in personalized, and therefore perfectly efficient, control.

[0010] To this end, the invention relates to a computer-implemented method for determining a control law to control at least one actuator for a system, said method comprising the following steps: - Provision of a physical model of the system; - Provision of a trained model of the dynamics of at least one actuator; - Determination of a hybrid model of the system by concatenating the physical model of the system and the model of the dynamics of at least one actuator; - Determination of the control law from the hybrid model.

[0011] Thanks to these characteristics, the resulting control law is based both on the physical model of the system and the trained model(s) of the actuator dynamics. Taking into account the actual dynamics of the actuators avoids the need to correct for inaccuracies that would be introduced by assuming perfect actuator operation. This results in a dual effect: on the one hand, improved vehicle stability and performance are achieved, and on the other hand, energy efficiency in vehicle operation is increased since there is no longer a need to correct for erroneous actuator contributions.

[0012] According to an optional feature of the invention, the trained model of the dynamics of at least one actuator is obtained by means of machine learning, which takes as input a set of input data from the actuator and a set of output data from the actuator associated with the input data. The machine learning makes it possible to introduce real input and output data into an initially chosen model and to adapt its parameters to best match the output data from the model with the real output data from the actuator.

[0013] According to an optional feature of the invention, the step of determining a model of the actuator dynamics comprises: - a first sub-step of choosing a model defined by initial parameters; - a second sub-step of feeding the model with a first input data from the actuator to obtain a first output data from the model; - a third sub-step of determining a difference between the first output data of the model and the first output data of the actuator associated with the first input data of the actuator; - a fourth sub-step of adjusting the initial parameters and carrying out the second and third sub-steps until the gap is less than a first predefined value.

[0014] By proceeding in this way, a model of the actuator dynamics is determined by iteration on the basis of a set of real data.

[0015] According to an optional feature of the invention, the step of determining a model of the actuator dynamics further comprises a fifth substep for validating the actuator dynamics model, defined by performing the second substep with a second input data point, distinct from the first input data point, to obtain a second output data point for the model, and performing the third substep with the second output data point. If the deviation is greater than a second predefined value, a sixth alert substep is performed. If the deviation is less than this second predefined value, then the actuator dynamics model is validated.

[0016] This fifth substep allows for the consideration of a second set of input / output data to consolidate the actuator dynamics model. In the event of a significant discrepancy between the output data from the actuator dynamics model and the actual output data from the actuator, the parameters of the actuator dynamics model are adjusted. This substep therefore serves to validate the model and ensure the robustness of the method of the invention. The warning substep, which may be optional, alerts a user to the fact that the actuator dynamics model may have deficiencies.

[0017] In one embodiment of the invention, the actuator dynamics model is a linear time-invariant model.

[0018] In another embodiment of the invention, the actuator dynamics model is a linear model with varying parameters.

[0019] In variants of the invention, the control law is of the linear-quadratic-Gaussian type or of the predictive control type.

[0020] The invention also covers a vehicle comprising at least one actuator and computing means configured to implement the method described above, in which the system is the vehicle.

[0021] The invention also relates to a computer program comprising instructions for the execution of such a method, when the program is executed by a processor.

[0022] Finally, the invention also relates to a processor-readable recording medium on which is recorded a program containing instructions for the execution of said method, when the program is executed by a processor.

[0023] Other features and advantages of the invention will become apparent from the following description on the one hand, and from several illustrative and non-limiting examples of embodiments given with reference to the accompanying schematic drawings on the other hand, in which:

[0024] [Fig. 1] schematically represents the main steps of the method for determining a control law according to the invention,

[0025] [Fig.2] schematically represents the sub-steps of the step of determining a actuator dynamics model according to the invention,

[0026] [Fig.3] schematically represents the sub-steps of the determination step of a model of the actuator dynamics according to the invention with an optional sub-step for validating the actuator dynamics model,

[0027] [Fig.4] represents a control law that controls the yaw rate of a vehicle and the associated yaw moment without and with application of the method of the invention,

[0028] [Fig. 5] represents a control law that controls the yaw rate of a vehicle and the rear steering of the associated 4WS without and with application of the method of the invention.

[0029] The features, variants, and different embodiments of the invention, as described or as they will be presented in the detailed description that follows, can be combined in various ways, provided that they are not incompatible or mutually exclusive. In particular, variants of the invention may be conceived comprising only a selection of features described hereafter in isolation from the other described features, if this selection of features is sufficient to confer a technical advantage and / or to differentiate the invention from the prior art.

[0030] For the sake of clarity, the same elements are designated by the same references in the different figures.

[0031] As already mentioned, the invention is described in its automotive application to determine a control law to control an actuator for a vehicle, but is by no means limited to it and can be applied to an actuator for any type of system.

[0032] Figure 1 schematically represents the main steps of the method for determining a control law according to the invention. The method for determining a control law Le is implemented by computer. The control law Le allows for the control of at least one actuator 20, 30 for a system, in particular a vehicle 10. The method for determining the control law according to the invention includes a step 100 of providing a physical model M10 of the vehicle 10. The physical model M10 is a physical model of the vehicle, based on the laws of physics, such as those used in current solutions for controlling a vehicle.

[0033] In what follows, the invention is illustrated by taking into account the actuators of the vehicle's lateral control, that is to say, the wheel and differential braking actuators. Of course, the invention applies equally to any type of actuator, by adapting the models used.

[0034] By way of example, for longitudinal control, the actuators involved are the engine or braking actuators. For vertical control, the actuators involved are the active or semi-active suspension actuators.

[0035] Lateral or yaw control of the vehicle usually employs a linear dynamic model, commonly called the bicycle model of the vehicle. The bicycle model is the most widely used for trajectory generation and the development of control laws for vehicle control.

[0036] This model results from Newton's first two laws, which allow us to obtain the vehicle dynamics equations, referenced below (6) and (9). The table below presents the physical parameters and signals characterizing this physical model. Symbol Description Distance from center of gravity to front axle ^■2 Distance from center of gravity to rear axle Front axle drift stiffness Di Rear axle drift stiffness fi Road friction coefficient vx Longitudinal speed m Vehicle mass Z? Vehicle drift angle Front axle drift angle 52 Rear axle drift angle 7' Yaw rate <5 / Front wheel steering angle 8r Rear wheel steering angle (4WS) Mz Yaw moment Ff Transverse force on front axle Fr Transverse force on rear axle

[0037] According to Newton's first law: the sum of the external forces is equal to the mass multiplied by the acceleration (equation (1)) and we obtain: / y + fj. = 4 g>) (2) Dj^i + D252 = mVxQ3 + (3) 5i=3f K) 32=8r-^ + ^ (5) kf £ = - +1-1++ to sf +¾¾ w

[0038] According to Newton's second law, the sum of the moments of the external forces is equivalent to the derivative of the angular momentum at the center of gravity of the vehicle (equation (7)). We then obtain: YMsxt = Hca (7} OjLjÆj — D2L2S2 + Mz = ¢8)

[0039] Starting from equations (6) and (9), we obtain the mathematical description of the bicycle model of the vehicle in the form of a referenced state representation (10): (i = Ax + 5" lv = Cx ■ (.¾ 4- .¾) rn « 1¾. ■ 4 (10) Æ Li 4 has i 4-

[0040] with:

[0041] where:

[0042] x = ( ÿ ) is the state variable,

[0043] _ / ôf \ is the control vector, with 5f the front wheel steering, the "- HAS \ / DAE (short for Electric Power Steering), or the rear wheel steering, the 4WS or 4 Wheels Steering (for 4 steering wheels), and Mz the overall yaw moment to be allocated to the braking actuators or to the IWM (In Wheel Motor) or to any other actuator capable of performing this yaw moment.

[0044] v _ ( P ] is the output of the model.

[0045] The determination method according to the invention includes a step 200 of providing a driven model M20, M30 of the dynamics of at least one actuator 20, 30. If the vehicle implements a single actuator 20, the method includes providing the driven model M20. If the vehicle implements a single actuator 30, the method includes providing the driven model M30. If the vehicle implements both actuators 20 and 30, the method includes providing the driven models M20 and M30. Of course, a similar procedure can be used with a plurality of actuators, for example 10, 50, or 100.

[0046] For each actuator 20 (respectively 30), the trained model of the actuator dynamics is obtained by means of machine learning receiving as input a set of input data E20 (respectively E30) of the actuator 20 (respectively 30) and a set of output data S20 (respectively S30) of the actuator 20 (respectively 30) associated with the input data E20 (respectively E30).

[0047] For each actuator 20, 30, input data E20, E30 and output data S20, S30 are collected. Based on this data, a machine learning technique is used to produce a model of the dynamics of the actuators involved in controlling the vehicle.

[0048] The determination of the trained model (step 205) will be detailed below.

[0049] Next, the method of the invention includes a step 300 of determining a hybrid model MH10 of vehicle 10 by concatenation of the physical model M10 of vehicle 10 and the model M20, M30 of the dynamics of at least one actuator 20, 30.

[0050] The hybrid model is a concatenation of the previously provided physical model of the vehicle and the actuator dynamics model previously obtained by Machine Learning.

[0051] Finally, the method of the invention includes a step 400 of determining the control law Le from the hybrid model MH 10. Instead of using the physical model of the vehicle, as is done in the prior art, the method of the invention makes it possible to design the control law based on the hybrid model, comprising the physical model of the vehicle and the model of the dynamics of the actuators.

[0052] The determination of the control law Le is done in a traditional manner. A control law is easily determined based on knowledge of all the states of the system. The model, being a function that knows the vehicle's modeling, calculates an output adapted to an actuator based on the inputs. For example, for a longitudinal speed actuator, the model links a speed command to a torque to be applied by the motor.

[0053] The control law is for example of linear-quadratic-gaussian type or of predictive control type.

[0054] It is thus understood that, thanks to the invention, the known vehicle control model based on the laws of physics (called the vehicle's physical model) remains unchanged. It is supplemented or augmented, for each actuator used by the vehicle, by a model of the dynamics of said actuator. This yields the hybrid model.

[0055] The method of the invention contributes to improved vehicle stability, thereby increasing the safety of vehicle occupants. Furthermore, the method of the invention reduces the energy required to control the vehicle.

[0056] According to the invention, the method for determining the control law Le is implemented on the vehicle. The step of determining a model of the actuator dynamics, however, is carried out offline, prior to determining the control law and before proceeding to use the vehicle.

[0057] Fig. 2 schematically represents the sub-steps of the step of determining a model of the dynamics of the actuator according to the invention.

[0058] As described previously, the trained model of the dynamics of at least one actuator is obtained by means of machine learning, taking as input a set of input data E20, E30 from actuator 20, 30 and a set of output data S20, S30 from actuator 20, 30 associated with the input data E20, E30 (step 205). Step 205 is a step for determining the dynamic model, which is implemented offline, prior to determining the control law Le. This model is obtained using machine learning techniques based on the input and output data of each actuator. The model is obtained solely through data, hence the name Data Driven Model, as opposed to a physical model obtained through the laws of physics.

[0059] In the context of the invention, the Data Driven Model approach is identical to the Supervised Machine Learning approach or to System Identification. It consists of estimating the parameters of the actuator dynamics model from the input and output signals.

[0060] Step 205 of determining a model M20, M30 of the dynamics of the actuator 20, 30 includes a first substep 210 of choosing a model Mi defined by initial parameters Pi. In order to limit the complexity of the calculations, a reduced-order model may be used, but the invention applies identically to higher-order models.

[0061] For actuators, a reduced-order LTI (Linear Time Invariant) model is suitable, but other types of models can be used, for example the LPV (Linear Model with Variing Parameters).

[0062] The inputs should preferably sweep across the entire amplitude of the actuator, in order to obtain a model that is reliable over the entire operating range of the actuator.

[0063] Step 205 of determining a model M20, M30 of the dynamics of the actuator 20, 30 includes a second substep 220 of feeding the model Mi with a first input data El of the actuator to obtain a first output data SMI of the model Mi.

[0064] This second substep 220 is carried out following the activation of each actuator in predefined configurations (with variable amplitudes and frequencies) to obtain corresponding input and output data sequences. This data is processed according to modeling assumptions: frequency filtering, bias elimination, weighting of data sequences, etc. For example, for a wheel steering actuator, the input is the steering setpoint, and the output is the actuator's response (i.e., the actual steering angle applied). After obtaining the input / output data pairs, the model Mi is successively fed an input data El to produce an output data for the SMI model.

[0065] Step 205 of determining a model M20, M30 of the actuator dynamics 20, 30 includes a third substep 230 of determining a gap 40 between the first SMI output data of the model and the first SI output data of the actuator associated with the first El input data of the actuator. In other words, the SMI output data of the model is compared with the actual SI output data of the actuator to determine a gap between the output signal measured on the actuator and the output signal of the model subjected to the same input signal. If the gap is greater than a predefined value, it is then necessary to adjust the characteristics of the model in order to minimize the gap.

[0066] Step 205 of determining a model M20, M30 of the dynamics of the actuator 20, 30 includes a fourth substep 240 of adjusting the initial parameters Pi and carrying out the second and third substeps 220, 230 until the gap 40 is less than a first predefined value 50. For convex optimization, the identified parameters of the model are those that minimize the L2 norm (least squares norm) of the gap between the model and reality. This means that the parameters Pi of the model are modified and then the substeps of feeding the model with the input data and determining the gap between The output data from the model and the actual data are repeated until the difference obtained is less than the predefined value of 50.

[0067] The learning phase, i.e., step 205 of determining a model of the actuator dynamics, is done "offline", prior to the use of the vehicle. It is done once for the entire range of vehicles.

[0068] Figure 3 schematically represents the substeps of the actuator dynamics model determination step according to the invention, with an optional actuator dynamics model validation substep. In addition to the steps presented previously, the actuator dynamics model determination step 205 further includes a fifth actuator dynamics model validation substep 250 defined by carrying out the second substep 220 with a second input data E2, distinct from the first input data E1, to obtain a second output data SM2 of the model, and carrying out the third substep 230 with the second output data SM2, S2, and if the deviation 40 is greater than a second predefined value 51, a sixth alert substep 260.

[0069] This substep allows the obtained model to be identified and validated by verifying its predictive nature, that is, its ability to account for the actuator output, for input sequences different from those used for identification. This step is optional and allows the model's quality to be ensured.

[0070] The step of determining the hybrid model will now be explained. As an example, a model of the actuator dynamics determined in step 205 may have an LTI-type structure. Such a model is first-order and allows approximating the actuator dynamics. Such a model for the 4WS (four-wheel steering) application may have the following structure: 1 * $

[0071] with Kp 4^5, the actuator gain; its response time constant, and 4 the Laplacian. As an example, the actuator gain can take the value 1 and the time constant can take the value 0.05.

[0072] Previously, the method for identifying the dynamics of the actuators was illustrated with the example of the 4WS actuator. The hybrid model used for the design of the control law is a concatenation of the physical model of the vehicle and the model of the dynamics of the different actuators of the vehicle.

[0073] To construct this hybrid model, a physical model of the vehicle and a dynamic model of each actuator implemented in the vehicle are therefore required. The model The vehicle physics was presented previously (see referenced equation (10)). Regarding the actuator dynamics model, based on the example of the 4WS actuator, the corresponding state representation is:

[0074]

[0075]

[0076]

[0077]

[0078]

[0079]

[0080]

[0081]

[0082] With 4a “ î wrzn77777 Q. Ww„4ws is the actuator excitation input (i.e., the setpoint), and the actuator output ^C_4W5=-^«_4W5. The Data Driven approach applied to braking actuators that perform the overall yaw moment gives the following state representation: with : 4 -__. p ■"a Srake T / ■ °a Brake' Kp.grake 'tn. TV fl rp_5rake - ua_Brake is the excitation input of the brake actuator, and the output of the actuator ya_Brake=Xa_Brake. The Data Driven approach applied to DAE yields the following state representation: with : 4a Erased"- "i 77777 OAE” AED - ua_DÆ is the excitation input of the DAE actuator, and the output of the actuator ya^DAE^0-0^. The hybrid model is obtained by concatenating v, ^4^5=^^4^5, yajBrake=xa_Brake and ya DÆ=Xa-DAE. Denoting by the index relevant to the hybrid model, the new state variable is:

[0083]

[0084]

[0085]

[0086]

[0087]

[0088] We obtain the state representation of the hybrid model: i -¾ + <■ hh with : —1 + 0 wut X - 0 0 0 1 ■5—Z? .i'1 s ; L ; 0 0 0 H 0 ir r, 0 1.? O; Bft = 0 0 — DAE 0 0 0 0 0 s 1 0 y 0 0 and Ch = Fl 10 0 1 0 0 for Vi ÜJ The control law design is based on the hybrid model. Although the DAE actuator, 4WS, and braking are used in the hybrid model, the method for obtaining this hybrid model will apply analogously with other actuators. Figure 4 represents a control law that controls the yaw rate of a vehicle (upper graph) and the associated yaw moment without and with application of the method of the invention (lower graph). Figure 5 represents a control law that controls the yaw rate of a vehicle (upper graph) and the rear steering of the associated 4WS without and with application of the method of the invention (lower graph). Figures 4 and 5 illustrate three configurations: - Yaw rate obtained without an IP control law (No Controller, noted NC) - The yaw rate obtained with a control law designed on the physical model ip (Controller based on Vehicle mode!, noted CB - MV) - The yaw rate obtained with a control law designed on the hybrid model ip (Controller based on Hybrid Vehicle model, noted CB-MHV)

[0089] The application illustrated in figures 4 and 5 corresponds to an overtaking maneuver, with a first lane change to overtake, then a second lane change to return to the original line.

[0090] The "reference" control law represents the control law giving rise to the ideal trajectory.

[0091] The NC (“no controller”) control law is quite far from the reference trajectory, because driver errors are not compensated by trajectory aids.

[0092] The CB-MV (“controller based on vehicle model”) control law, based on the physical model alone, offers an improvement over the driver-only control, but exhibits peaks relative to the reference. These peaks are overshoots of the setpoint, linked, for example, to amplitude errors in the actuators. For example, if the desired setpoint is a 3.5° steering angle for the front wheels, and the actuator only applies 3°, the setpoint compensates by increasing the requested steering angle, resulting in increased energy consumption.

[0093] The CB-MHV control law designed on the hybrid model offers better performance than the control law designed on the physical model in terms of stability and control of the setpoint or reference. Indeed, the CB-MHV control law designed on the hybrid model eliminates the peaks introduced by the failure to account for the dynamics of the actuators in the physical model.

[0094] In addition, the control required for this performance is reduced by approximately 20% in the CB-MHV control law using the hybrid model developed within the framework of this invention, which gives it the character of energy efficiency.

[0095] The lower graph of Figures 4 and 5 illustrates this reduction with a low yaw moment and a low steering angle of the 4WS (5) for the control law designed on the basis of the hybrid model of the vehicle.

[0096] In figures 4 and 5, the use of yaw moment and rear steering 4WS are done independently of each other.

[0097] For a performance which proves to be better with the control law using the hybrid model (upper graph of [Fig.4]), the control energy (the yaw moment) is reduced by about 20% for the control law using the hybrid model (lower graph of [Fig.4]).

[0098] For a performance which proves to be better with the control law using the hybrid model (upper graph of [Fig.5]), the control energy (the rear steering setpoint 4WS) is also reduced by about 20% for the control law using the hybrid model (lower graph of [Fig.5]).

[0099] Thanks to the method of the invention, a more stable vehicle behavior is obtained, guaranteeing increased passenger safety. This stability is of particularly crucial importance in emergency obstacle avoidance situations where the vehicle (i.e., the vehicle and all the actuators involved in the avoidance maneuver) must be perfectly controlled.

[0100] The method of the invention also ensures, due to the use of the hybrid model for the determination of the control law, a lower energy consumption of the control, resulting in an increase in battery autonomy.

[0101] The invention also relates to a vehicle comprising at least one actuator and computing means configured to implement the method of determining a control law Le to control at least one actuator 20, 30 for the vehicle.

[0102] The invention also covers a computer program comprising instructions for executing the method according to the invention as described above, when the program is executed by a processor.

[0103] The invention also relates to a processor-readable recording medium on which is recorded a program comprising instructions for executing the method of determining a control law Le to control at least one actuator for a system, when the program is executed by a processor.

[0104] Although not limited to such applications, the embodiments of the invention are particularly advantageous for implementation in a vehicle, and particularly an autonomous vehicle.

[0105] A person skilled in the art understands that the system or subsystems according to embodiments of the invention can be implemented in various ways in the form of hardware, software, or a combination of hardware and software, in particular in the form of program code that can be distributed as a program product in various forms. In particular, the program code can be distributed using computer-readable media, which may include computer-readable storage media and communication media. The processes or methods described herein can, in particular, be implemented in the form of computer program instructions that can be executed by one or more processors in a computer processing device. These computer program instructions can also be stored in computer-readable media.

[0106] Of course, the invention is not limited to the examples just described, and many modifications can be made to these examples without departing from the scope of the invention. In particular, the characteristics of different embodiments variations of the invention can be combined to realize the invention, provided that these variations are not incompatible with each other.

Claims

Demands

1. A computer-implemented method for determining a control law (Le) to control at least one actuator (20, 30) for a system (10), said method comprising the following steps: - Providing (100) a physical model (M10) of the system (10); - Providing (200) a driven model (M20, M30) of the dynamics of at least one actuator (20, 30); - Determining (300) a hybrid model (MH10) of the system (10) by concatenating the physical model (M10) of the system (10) and the model (M20, M30) of the dynamics of at least one actuator (20, 30); - Determining (400) the control law (Le) from the hybrid model (MH10).

2. Method according to claim 1, wherein the trained model of the dynamics of at least one actuator is obtained by means of machine learning receiving as input a set of input data (E20, E30) of the actuator (20, 30) and a set of output data (S20, S30) of the actuator (20, 30) associated with the input data (E20, E30) (step 205).

3. A method according to claim 2, wherein the step (205) of determining a model (M20, M30) of the actuator dynamics (20, 30) comprises: - a first substep (210) of selecting a model (Mi) defined by initial parameters (Pi); - a second substep (220) of feeding the model (Mi) with a first input data (El) of the actuator to obtain a first output data (SMI) of the model (Mi); - a third substep (230) of determining a gap (40) between the first output data (SMI) of the model and the first output data (SI) of the actuator associated with the first input data (El) of the actuator; - a fourth substep (240) of adjusting the initial parameters (Pi) and carrying out the second and of the third sub-steps (220, 230) until the gap (40) is less than a first predefined value (50).

4. Method according to claim 3, wherein the step (205) of determining a model of the actuator dynamics further comprises a fifth substep (250) of validating the model of the actuator dynamics defined by carrying out the second substep (220) with a second input data (E2), distinct from the first input data (El), to obtain a second output data (SM2) of the model and carrying out the third substep (230) with the second output data (SM2, S2), and if the deviation (40) is greater than a second predefined value (51), a sixth substep (260) of alerting.

5. Method according to any one of claims 2 to 4, wherein the actuator dynamics model is a linear time-invariant model.

6. Method according to any one of claims 2 to 4, wherein the actuator dynamics model is a linear model with varying parameters.

7. Method according to any one of claims 1 to 6, wherein the control law is of linear-quadratic-Gaussian type or of predictive control type.

8. Vehicle comprising at least one actuator and computing means configured to implement the method according to claim 1 wherein the system (10) is the vehicle.

9. A computer program comprising instructions for carrying out the method according to any one of claims 1 to 7, when the program is executed by a processor.

10. Processor-readable recording medium on which is recorded a program containing instructions for executing the method according to any one of claims 1 to 7, when the program is executed by a processor.

Images