Method for recalibrating a control device of an electromechanical brake

Abstract

The invention relates to a method (300) for recalibrating a control device (101; 102) of an electromechanical brake (100; 200), wherein the electromechanical brake has an electromechanical actuator (104) that can be controlled by means of the control device. The method has the steps of: determining (302) a quality function (F(a)) which assigns a quality value to a control parameter set (a) of the control device, the quality value being a measure of the control performance of the control device; determining a control parameter set (a*) which maximizes the determined quality function; and defining (304) the determined control parameter set as a recalibrated control parameter set.

Description

[0001] Description

[0002] title

[0003] Method for recalibrating a control device of an electromechanical brake

[0004] The present disclosure relates to a method for recalibrating a control device of an electromechanical brake.

[0005] In conventional hydraulically actuated friction brakes in vehicles, braking torque is typically varied and controlled via a given hydraulic pressure. Changed boundary conditions, e.g., in electric vehicles, can also make the use of electromechanical (electromechanically actuated) brakes (EMB) attractive.

[0006] Even with electronic brake boosters (EBBs), it is necessary to regulate braking torque to guarantee safe vehicle operation. Braking torque control is usually achieved indirectly by adjusting the actuation force. While force control in conventional hydraulic brakes is based on the value measured by a pressure sensor and its deviation from the target pressure, in EMBs the actuation force at each wheel is either measured directly via a physical force sensor or indirectly reconstructed via a stored stiffness characteristic, estimation methods, or sensor fusion.

[0007] The control of electromechanical actuators in electromechanical brakes is based on control parameters (e.g., gain factors) that significantly influence the performance criteria of the brake control. The system behavior of an electromechanical actuator typically changes over its service life due to aging effects. Even after brake maintenance, such as the installation of a different brake pad, the behavior can change. The control system must continue to function reliably despite these changes, which are often difficult to predict.

[0008] The object of the invention is to provide a method by which the control parameters of a control device of an electromechanical brake can be reliably and safely recalibrated.

[0009] This problem is solved by a method for recalibrating a control device of an electromechanical brake, which has an electromechanical actuator controllable by means of the control device, wherein the method comprises:

[0010] • Determining a performance function that assigns a performance value to a set of control parameters of the control device, where the performance value is a measure of the control performance of the control device,

[0011] • Determining a set of control parameters that maximizes the determined performance function, and

[0012] • Defining the determined set of control parameters as a recalibrated set of control parameters, wherein determining the performance function comprises the following steps: a) selecting a new set of control parameters starting from a current set of control parameters, b) supplying a time-varying control signal with a waveform to the electromechanical actuator such that the electromechanical actuator is caused to generate a predetermined brake actuation pattern, c) detecting a response of the electromechanical actuator to the control signal, d) determining a performance value based on the detected response of the electromechanical actuator, and e) updating the performance function based on the determined performance value.

[0013] The specified signal shape can be designed in such a way that a relevant dynamic range of the actuator is covered. Examples of signal shapes include:

[0014] • a stepped function that corresponds to different braking force levels,

[0015] • a chirp signal, i.e. a time-varying signal (e.g. a sinusoidal signal) whose frequency varies over time (e.g. in a range between 1-10 Hz),

[0016] • Signal shapes that correspond to force build-up and force reduction ramps (i.e., with predetermined braking force gradients).

[0017] The signal shape can be changed or updated over the lifetime of the electromechanical brake, for example wirelessly (e.g. via so-called "over-the-air updates") or as part of maintenance.

[0018] The response of the electromechanical actuator can be detected, for example, by sensors that are standard on the brake, such as a position, force, and / or current sensor. Accordingly, the actuator's response can include a detected position and / or braking force and / or a requested electrical current (a requested electrical power).

[0019] The quality value can be a scalar. Quality can be defined such that an increasing quality value corresponds to increasing control performance.

[0020] Changing the set of control parameters, followed by evaluating the control behavior based on the actuator's recorded response and updating the performance function, provides an efficient way to determine a better set of control parameters compared to the current set, which can achieve improved control behavior.

[0021] The procedure can be performed iteratively, i.e., steps a) to e) can be performed repeatedly, using the last determined set of control parameters as the starting point in each new iteration step.

[0022] The performance function can be modeled as a Gaussian process. This approach offers the possibility of approximating a function that is unknown a priori.

[0023] Selecting a new set of control parameters from a current set can be performed using Bayesian optimization. Bayesian optimization is suitable for finding the maximum of an unknown function, especially an unknown function modeled using a Gaussian process (here, the performance function). Bayesian optimization and the Gaussian process can be carried out based on the methods / algorithms described in P.I. Frazier, "A Tutorial on Bayesian Optimization", arXiv:1807.02811v1, July 8, 2018, and their extensions from F. Berkenkamp et al., "Safe Controller Optimization for Quadrotors with Gaussian Process", arXiv:1509.01066v4, August 16, 2017. This allows for computationally and time-efficient recalibration.

[0024] The selection of the new set of control parameters can be carried out, for example, based on one or both of the following criteria:

[0025] • The new set of control parameters maximizes an expected performance function and

[0026] • The new set of control parameters lies on a boundary of a safe set of control parameters, which includes several sets of control parameters with which stable control can be carried out.

[0027] The quality factor can be determined, for example, based on a deviation from a reference trajectory (i.e., a reference behavior) of the electromechanical actuator and a requested electrical power. For instance, the quality factor can be determined based on the difference between a detected braking force and a reference braking force, as well as a requested electrical power. Alternatively, the quality factor can be determined based on the difference between a detected position of an electromechanical actuator (e.g., an electric motor) and a reference position, as well as a requested electrical power.

[0028] Determining the performance function can be performed taking one or more boundary conditions into account. This prevents damage to the brake and suppresses instabilities in the control system. For example, such a boundary condition could include limiting the requested setpoint current, as an excessively high setpoint current can permanently damage the electromechanical actuator. Alternatively or additionally, a boundary condition could stipulate that an integrator does not saturate, since saturation can impair control accuracy.

[0029] One or more boundary conditions can be taken into account using a boundary condition function modeled as a Gaussian process. The boundary condition function can also be updated based on a corresponding performance value. Both the performance function and the boundary condition function can be determined / updated iteratively using the algorithms mentioned above.

[0030] In another aspect of the present disclosure, a data processing facility (data processing unit) is provided which is configured to perform a previously described procedure.

[0031] Furthermore, according to the present disclosure, a computer program is provided containing instructions which, when executed by a processor, cause the processor to perform a previously described procedure. In another aspect of the disclosure, a computer-readable medium is provided which stores instructions which, when executed by a processor, cause the processor to perform a previously described procedure.

[0032] The present disclosure is explained below by reference to the accompanying drawings.

[0033] Figure 1 is a schematic representation of an exemplary electromechanical brake according to the present disclosure.

[0034] Figure 2 is a schematic representation of another exemplary electromechanical brake according to the present disclosure.

[0035] Figure 3 shows a flowchart of an exemplary method for recalibrating a control device of an electromechanical brake according to the present disclosure.

[0036] Figure 4 is a graphic representation of an exemplary brake actuation pattern.

[0037] Figure 5 shows graphical representations illustrating the determination of a quality function and a boundary condition function.

[0038] Figure 1 is a schematic representation of an electromechanical brake 100, which includes a control unit 102 and an electromechanical actuator 104. The control unit 102 is configured to control the electromechanical actuator 104. The electromechanical actuator 104 has an electromechanical controller 104-1, which can be configured, for example, as an electric motor, optionally as a direct current motor (DC motor), e.g., as a brushless DC motor (BLDC motor). As shown in Figure 1, the control unit 102 receives input variables including a target braking force F. soll , an actual actuating force (or an estimate thereof, i.e., an estimated actuating force) F, as well as an I-position change rate Ö ist supplied to the electromechanical actuator 104-1.

[0039] The control device 102 can be configured to derive the target braking force F from the target braking force. solla reference trajectory (reference braking force) F using a reference model 102-1 ref to calculate. The reference model 102-1 can generally be designed such that the reference trajectory it generates meets the requirements of the braking performance (braking capacity), but only within the limits of the physical capabilities of the actuator 104. This ensures, for example, that only those actuator position changes are requested that can actually be realized by the actuator 104-1.

[0040] The control device 102 has a P controller 102-2 and a downstream PI controller 102-3, with which a cascade control can be implemented.

[0041] The P-controller 102-2 is designed to calculate a deviation (differential force) AF between the reference braking force F. ref and the actuating force F a reference position change rate 0 refto determine. The control parameter of the P-controller 102-2 is shown in Figure 1 with reference numeral K. p F designated.

[0042] From the reference position change rate Ö ref The actual position change rate Ö is then calculated. ist subtracted, resulting in a differential position change rate AÖ, which is fed to the PI controller 102-3. The control parameters of the PI controller 102-3 are shown in Figure 1 by K p ö and K i ö The PI controller 102-3 provides a setpoint torque r. soll out, which is supplied to the electromechanical actuator 104.

[0043] In actuator 104, the target torque r is set. soll via a link 104-2 into a target current I soll converted, which serves as the basis for controlling the electromechanical actuator 104-1. The electromechanical actuator 104-1 sets the target current I sollinto a spreading path x, by which a brake pad is to be moved by the actuator 104-1. The spreading path therefore corresponds to an actuation force.

[0044] The control system outlined in Figure 1 is based on a measured (estimated) actuation force. This control scheme is subsequently referred to as sensor-based control.

[0045] Alternatively, the actuator 104 can be controlled based on a previously known and / or operationally learned / adapted stiffness characteristic (force-displacement characteristic). Such a control scheme is subsequently referred to as stiffness-based control.

[0046] An electromechanical brake 200 with a control device configured to control an electromechanical actuator 104 on the basis of stiffness-based control is described below with reference to Figure 2.

[0047] The electromechanical brake 200 shown in Figure 2 comprises a control unit 202 and an electromechanical actuator 104. The control unit 202 is configured to control the electromechanical actuator 104 by means of stiffness-based control. The electromechanical actuator 104 can be configured identically to the electromechanical actuator shown in Figure 1. To avoid repetition, reference is made to the preceding description of the actuator 104 in connection with Figure 1.

[0048] As shown in Figure 2, the control device 202 receives, among other things, a target braking force F as input variables. soll , an actual position 0 ist as well as a rate of change in actual position Ö ist supplied to the electromechanical actuator 104-1.

[0049] The control device 202 can be configured to derive the target braking force F from the target braking force. sollbased on a stiffness characteristic (force-displacement characteristic) 202-1, a target actuator position 0 soll to calculate, from which a reference position 0 can be determined using a reference model 202-2 ref The reference model 202-2 can generally be designed such that the reference trajectory it generates (e.g., actuator position trajectory) meets the requirements of the braking performance (braking capacity), but only within the limits of the physical capabilities of the actuator 104. This ensures, for example, that only actuator position changes are requested that can actually be implemented by the actuator 104-1.

[0050] The control device 202 has a P controller 202-3 and a downstream PI controller 202-4, with which a cascade control can be implemented.

[0051] The P-controller 202-3 is designed to compensate for a deviation A0 between the reference controller position 0 ref and the I position 0 ist a setpoint position change rate of 0 soll to determine. The control parameter of the P-controller 202-3 is shown in Figure 2 with reference numeral K. P;0 designated.

[0052] From the target position change rate 0 soll The I-position change rate is then set to 0. ist This is subtracted, resulting in a differential controller position change rate A0, which is fed to the PI controller 202-4. The control parameters of the PI controller 202-4 are shown in Figure 1 by K p ö and Kj ö The PI controller 202-4 provides a setpoint torque r. soll from, which is supplied to the electromechanical actuator 104, which, as described above, is driven by the target torque T SO11 a spreading path x is calculated.

[0053] As described above, the control parameters K are fed into the control unit 102. p F , K ö and K i ö one and into the 202 control parameters the control parameters K p 0 , K p ö and K i ö , which significantly influence the quality criteria of the brake control.

[0054] The behavior of an actuator in an electromechanical brake typically changes over its lifetime due to aging effects. Maintenance work on the brake, such as installing a different brake pad, can also alter its behavior. The control system must continue to function reliably despite these changes, which are often difficult to predict.

[0055] These unpredictable changes in system behavior can be countered by recalibrating the control device 102, 202, thereby readjusting the control parameters.

[0056] The following describes a procedure for recalibrating the control devices 102 and 202 described above. Recalibration means finding a new set of control parameters starting from a current set of control parameters. The control parameters to be recalibrated in the control device 102 are: K p F , K p ö and K i ö The control parameters to be recalibrated in the control unit 202 are: K p 0 , K p ö and K i ö The control parameters of the respective control devices 102, 202 are hereinafter referred to as control parameter set a.

[0057] Figure 3 shows a flowchart of an exemplary procedure 300 for recalibrating control parameters of an electromechanical brake.

[0058] As shown in the flowchart in Figure 3, the process 300 has:

[0059] • Determine 302 a performance function F(a) which assigns a performance value G to a set of control parameters a of the control device 102, 202 a assigns, whereby the quality value G a a measure of the regulatory capacity of the regulatory institution 102, 202 is,

[0060] • Determine 304 a set of control parameters a* that maximizes the determined performance function F(a), and

[0061] • Defining the determined set of control parameters a* as a recalibrated set of control parameters, wherein determining the performance function F(a) comprises the following steps: a) selecting a new set of control parameters starting from a current set of control parameters, b) supplying a time-varying control signal with a waveform to the electromechanical actuator 104 such that the electromechanical actuator 104 is caused to generate a predetermined brake actuation pattern, c) detecting a response of the electromechanical actuator 104 to the control signal, d) determining a performance value G a from the recorded response of the electromechanical actuator 104 and e) updating the performance function F(a) based on the determined performance value G a .

[0062] The specified signal shape can be designed such that a relevant dynamic range of the actuator 104 is covered. Examples of signal shapes and / or brake actuation patterns include:

[0063] • a stepped function that corresponds to different braking force levels,

[0064] • a chirp signal, i.e. a time-varying signal (e.g. a sinusoidal signal) whose frequency varies over time (e.g. in a range between 1-10 Hz),

[0065] • Signal shapes that correspond to force build-up and reduction ramps (i.e., with predetermined braking force gradients).

[0066] Figure 4 shows an exemplary operating pattern F s displayed over time. The activity pattern F s It has a time-varying amplitude and frequency.

[0067] The signal shape can be changed or updated over the lifespan of the electromechanical brake, for example wirelessly (e.g., via so-called "over-the-air updates") or during maintenance. If the electromechanical brake 100, 200 is installed in a vehicle, the actuator 104 can be supplied with the specified signal shape in predefined situations where the vehicle's driving behavior is not affected, for example, during refueling or charging, during maintenance in a workshop, etc.

[0068] The performance function F(a) can be a function that assigns a performance value G to a set of control parameters a. a , i.e., assigns a scalar. The quality can be defined such that increasing quality values ​​correspond to increasing control performance.

[0069] Changing the set of control parameters a, followed by evaluating the control behavior based on the recorded response of actuator 104 and updating the performance function F(a), provides an efficient way to determine a better set of control parameters compared to the current set, with which improved control behavior can be achieved.

[0070] The procedure can be performed iteratively, i.e., steps a) to e) can be repeated, with the most recently determined set of control parameters a) being used in each new iteration step. (n) It can be used as a starting point for further optimization. The iteration can be terminated after a predetermined number of iteration steps.

[0071] The performance function F(a) can be modeled as a Gaussian process. This allows an a priori unknown function to be approximated.

[0072] Selecting a new set of control parameters a (n+1) starting from a current set of control parameters a (n) This can be performed using Bayesian optimization. Bayesian optimization is suitable for finding the maximum of an unknown function modeled using a Gaussian process (here, the performance function F(a)). The Bayesian optimization and the Gaussian process can be carried out based on the methods / algorithms described in P.I. Frazier, "A Tutorial on Bayesian Optimization", arXiv:1807.02811v1, July 8, 2018, and their extension from F. Berkenkamp et al., "Safe Controller Optimization for Quadrotors with Gaussian Process", arXiv:1509.01066v4, August 16, 2017. This allows the recalibration to be performed in a computationally efficient manner.

[0073] The response of the electromechanical actuator 104 can be detected, for example, by sensors already provided as standard on the brake, such as a position, force, and / or current sensor. Accordingly, the response of the actuator 104 can include a detected position and / or braking force and / or a requested electrical current (a requested electrical power). Thus, no additional sensors are required. Furthermore, the method can be used with different types of electromechanical brakes, such as drum brakes, disc brakes, etc.

[0074] The response from actuator 104 can be used not only to calculate a performance rating, but also to determine the overall performance of the actuator, for example, to ascertain whether its overall performance is above a certain threshold. If this threshold is not met, a warning can be issued.

[0075] The time series data recorded during the execution of the actuation pattern, which correspond to the response of actuator 104, are used to evaluate the quality criterion, i.e., to determine a quality value G. a to calculate which corresponds to the current set of control parameters a (n) assigns a quantifiable quality.

[0076] In the case of sensor-based control (Figure 1), the time series data includes the course of the reference force F®, the measured (or estimated) actuation force F (k)as well as the requested actuator current (setpoint actuator current) 1^ for the sampling times k = 1, 2, ... , K. In stiffness-based control (Figure 2), the reference actuator position 0® and the measured actuator position (actual actuator position) 0® are recorded analogously instead of the reference force and the measured actuation force. In this case, the reference actuator position change rate 0® and / or the actual actuator position change rate 0® can also be recorded for performance evaluation.

[0077] For a given set of control parameters a, the quality, i.e., a quality value G, can be determined. a , for sensor-based control, calculated using the following expression:

[0078] In this expression (F® - F (k) ) represents a deviation from the reference fk') trajectory. 02!^ represents the electrical power requirement, e.g., the required electrical power. The weighting factors and c2 are positive constants that are predetermined.

[0079] For a given set of control parameters a, the quality, i.e., a quality value G, can be determined. a , for stiffness-based control, can be calculated analogously using the following expression:

[0080] In this expression, 0® - 0® ) represents a deviation from the reference fk') trajectory. 02!^ represents the electrical power requirement, e.g., the required electrical power. The weighting factors and c2 are positive constants that are predetermined.

[0081] Other brake-specific control parameters (e.g., the time required to build up locking force or the maximum overshoot) can, in principle, also be considered when calculating the performance factor. Determining the performance factor F(a) can be carried out taking one or more boundary conditions into account. This can prevent damage to the brake and suppress instabilities in the control system. For example, such a boundary condition could be a limit on a requested setpoint current I. soll This is because an excessively high target current can permanently damage the electromechanical actuator 104, i.e., I soll should be smaller than a maximum current I max This condition is subsequently referred to as condition r, > 0 and can be expressed mathematically as follows:

[0082] Fi = I max - max ltsollct)! > °' where t is the time and T is the total signal duration.

[0083] Alternatively or additionally, a further boundary condition can stipulate that an integrator does not saturate, as saturation can impair control accuracy. This condition is subsequently referred to as the condition r² > 0 and can be expressed mathematically as follows: where t is the time and T is the total signal duration.

[0084] The one or more boundary conditions can be taken into account by means of a boundary condition function S(a), which is determined / modeled using a Gaussian process and based on the methods / algorithms described in PI Frazier, “A Tutorial on Bayesian Optimization”, arXiv:1807.02811v1, 8 July 2018 and their extension from F. Berkenkamp et al. “Safe Controller Optimization for Quadrotors with Gaussian Process”, arXiv:1509.01066v4, 16 August 2017) based on the calculated performance value G. a will be updated.

[0085] These algorithms can thus approximate an unknown performance function F*(a) using an approximation function F(a). Furthermore, an unknown boundary condition function S*(a) can also be approximated by an approximation function S(a). The fundamental assumption here is that these functions are sufficiently smooth. This means that small changes in the control parameters a result in only small changes in the function value (see Figure 5). With this assumption and the modeling of the respective functions as a Gaussian process, the search for an optimal set of control parameters a* can be implemented using the algorithms described above.

[0086] Figure 5 shows the unknown performance function F*(a) and the approximate performance function F(a) plotted as functions of the control parameter set a in the upper graph. The area highlighted in the upper graph represents the variance of the approximate performance function F(a).

[0087] Figure 5 shows the unknown boundary condition function S*(a) and the approximate boundary condition function S(a) plotted as functions of the control parameter set a in the lower graph. The area highlighted in the lower graph represents the variance of the approximate boundary condition function S(a).

[0088] In Figure 5, a (0) A current set of control parameters to be recalibrated, a*, is an optimized set of control parameters obtained by a recalibration procedure according to the present disclosure. The recalibrated set of control parameters a* satisfies the exemplary boundary condition S(a) > 0.

[0089] An exemplary implementation of the recalibration method according to the present disclosure is described below. It is assumed that the electromechanical brake is mounted on a vehicle and functions as a vehicle brake.

[0090] The starting point is a current set of control parameters a (0) as well as a safe set of control parameters P o in an environment of a (0) in a state where the vehicle is in a predetermined condition, e.g., a state where braking is not required, such as during charging / refueling or in a workshop. A safe set of control parameters can be a set of control parameters that includes at least one set of control parameters with which stable control can be achieved.

[0091] Based on this, the following steps can be performed multiple times:

[0092] • Extend the safe control parameter set based on the envelope (variance) of S(a) and store the resulting control parameter set as P n ,

[0093] • Select new control parameter set a (n) e P nbased on the following criteria: a. new set of control parameters maximizes expected performance (based on the envelope (variance) of F(a)) or b. new set of control parameters increases the safe set of control parameters (i.e. a (n) (lies on the edge of the safe set of control parameters)

[0094] • Perform actuation patterns with new control parameters a (n) select and evaluate the quality criterion (quality value) as well as boundary conditions.

[0095] • Use the evaluation result to update the performance function F(a) and the boundary condition function S(a), i.e., the Gaussian processes F(a) and S(a).

[0096] Subsequently, a new, optimized set of control parameters is selected according to: a* = arg max F(a) aEP.

[0097] For electromechanical brakes used as vehicle brakes, as mentioned above, stationary phases, such as refueling or charging, are particularly suitable for recalibration. The frequency of recalibration can be adjusted based on the brake's usage profile. For example, critical driving situations, such as intense thermal stress on the brake / actuator, can be detected. These situations can alter the control parameters due to factors like heavy wear, outgassing of the brake pads, etc., and recalibration can be performed during the next stationary phase. Another situation where recalibration is particularly beneficial is after extended periods of vehicle inactivity (e.g., several weeks). During such periods, the actuator (e.g., due to insufficiently distributed / degraded lubricant in the transmission) or the brake pads can change. In these situations, recalibration should also be performed promptly.

[0098] Furthermore, if the actuator is known to have a strong dependence on environmental parameters, it is possible to perform and save targeted recalibrations under specific environmental conditions, so that the control parameters can be used when these conditions occur. For example, the friction in certain types of gears (e.g., worm gears) can be highly temperature-dependent, which affects the control. It would then be possible to perform targeted recalibrations at specific ambient temperatures (e.g., very cold temperatures, such as below freezing) and to store the control parameters for such situations.

[0099] The method according to the present disclosure can be carried out by one or more computers with one or more data processing units. The term "data processing unit" can be understood as any device configured to process data or signals. The data or signals can, for example, be processed according to at least one (i.e., one or more than one) special function performed by the data processing unit. A data processing unit can comprise or be configured as an analog circuit, a digital circuit, a logic circuit, a microprocessor, a microcontroller, a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), an integrated circuit, a programmable gate array (FPGA), or any combination thereof.Any other way of implementing the respective functions described in more detail here can also be understood as a data processing unit or logic circuit arrangement. One or more of the process steps described in detail here can be executed (e.g., implemented) by a data processing unit through one or more special functions performed by the data processing unit.

[0100] According to various explanations, the process is therefore implemented in a computer-based manner.

Claims

Claims 1. Procedure (300) for recalibrating a control device (102; 202) an electromechanical brake (100; 200) which has an electromechanical actuator (104) which can be controlled by means of the control device (102; 202), wherein the method (300) has: • Determine (302) a performance function (F(a)) which assigns a performance value to a set of control parameters (a) of the control device (102; 202), where the performance value is a measure of the control performance of the control device (102; 202), • Determine (302) a set of control parameters (a*) that maximizes the determined performance function (F(a)), and • Defining (304) the determined set of control parameters (a*) as a recalibrated set of control parameters, wherein determining (302) the performance function (F(a)) comprises the following steps: a) selecting a new set of control parameters starting from a current set of control parameters, b) supplying a time-varying control signal with a waveform to the electromechanical actuator (104) such that the electromechanical actuator (104) is caused to generate a predetermined brake actuation pattern, c) detecting a response of the electromechanical actuator (104) to the control signal, d) determining a performance value based on the detected response of the electromechanical actuator (104), and e) updating the performance function (F(a)) based on the determined performance value.

2. Method (300) according to claim 1, wherein the selection of the new set of control parameters is carried out from the current set of control parameters by means of Bayesian optimization.

3. Method (300) according to claim 1 or 2, wherein steps a) to e) are repeated.

4. Method (300) according to any one of claims 1 to 3, wherein the selection of the new set of control parameters is carried out on the basis of one or both of the following criteria: • the new set of control parameters maximizes the expected performance function (F(a)) and • The new set of control parameters lies on a boundary of a safe set of control parameters, which includes several sets of control parameters with which stable control can be carried out.

5. Method (300) according to one of claims 1 to 4, wherein the quality value is determined on the basis of a deviation from a reference trajectory of the electromechanical actuator (104) and a requested electrical power.

6. Method (300) according to any one of claims 1 to 5, wherein the determination of the quality function (F(a)) is carried out taking into account one or more boundary conditions.

7. Method (300) according to claim 6, wherein the one or more boundary conditions are taken into account by means of a boundary condition function (S(a)) modeled as a Gaussian process.

8. Method (300) according to claim 7, wherein the determination of the boundary condition function (S(a)) is carried out on the basis of the determined quality value.

9. Data processing unit configured to execute a method (300) according to any one of claims 1 to 8.

10. Computer program with instructions which, when executed by a processor, cause the processor to execute a method (300) according to any one of claims 1 to 8.

11. Computer-readable medium which stores instructions which, when executed by a processor, cause the processor to execute a method (300) according to any one of claims 1 to 8.

Images