METHOD FOR MANUFACTURED A SYSTEM FOR ESTIMATING THE CLAMPING FORCE EXERCISED BY AN ELECTRIC BRAKE AND A SYSTEM FOR ESTIMATING THE CLAMPING FORCE

Abstract

A method for manufacturing a system for estimating the clamping force exerted by an electric brake, comprising: a) Providing a set of machine learning means (ML1, ML2,…MLn) for determining an estimated clamping force, b) Providing the values ​​of the selected brake characteristics and clamping force values ​​measured by means of a sensor to the learning means, c) Generating a clamping force value (Fest1, Fest2,…Festn) by the learning means (ML1, ML2,…MLn), d) Combining said values ​​to generate an estimated clamping force (by means of a combination function), e) Comparing the estimated (Fest) and measured clamping forces, f) Adapting the learning means (ML1, ML2,…MLn) and repeating steps b) to e) until the estimated (Fest) and measured clamping forces are within a given error threshold interval, g) Integrating the means learning (ML1, ML2,…MLn) in the storage area and integration of the combination function. [Fig.3],

Description

Title of the invention: METHOD FOR MANUFACTURED A SYSTEM FOR ESTIMATING THE CLAMPING FORCE EXERCISED BY AN ELECTRIC BRAKE AND A SYSTEM FOR ESTIMATING THE CLAMPING FORCE TECHNICAL FIELD AND PREVIOUS ART

[0001] The present invention relates to a method for manufacturing a system for estimating a clamping force exerted by an electric brake and an estimation system obtained by such a method.

[0002] A motor vehicle is equipped with a brake at each wheel. This can be a disc brake or a drum brake.

[0003] The brake can be a hydraulic brake or an electromechanical brake designated EMB (“Electromechanical Brake” in Anglo-Saxon terminology).

[0004] Parking brakes are increasingly electrically activated, and it is interesting to be able to produce fully electrically actuation service brakes. For example, a screw-nut system operated by an electric motor causes the brake pads to be applied against the disc in the case of a disc brake, and the brake linings against the drum in the case of a drum brake.

[0005] We wish to be able to know the clamping force exerted by each brake in order to be able to check if the expected braking level is reached and / or to control the service brake in the case of an electromechanical brake.

[0006] Furthermore, we wish to reduce or limit the number of sensors, in particular we do not wish to implement a force sensor to know the clamping force.

[0007] Currently, the position of the rotor in the brake is used to deduce the clamping force using predetermined curves that relate the rotor position to the clamping force. However, it appears that these curves do not reflect the actual behavior of the brake. Indeed, knowing the rotor position alone is not always sufficient to estimate the clamping force with sufficient accuracy in certain configurations, particularly for an electromechanical service brake, because a large number of parameters are involved. Description of the invention

[0008] It is an object of the present invention to provide a method for manufacturing a precise and robust system for estimating the clamping force exerted by an electric brake requiring a reduced number of sensors.

[0009] It is also an object of the present invention to provide a system for estimating clamping force.

[0010] The stated goal above is achieved by a method of manufacturing a brake clamping force estimation system implementing the training of several machine learning means using selected features of the brake and its environment, each of the machine learning means providing an estimated clamping force value, which are then combined to provide an estimated clamping force.

[0011] At least part of the brake characteristics used for estimating the clamping force is advantageously estimated by a combination of estimates obtained each by machine learning means.

[0012] The estimation of the clamping force obtained through the invention is less sensitive to estimation errors than prior art devices, since the effect of possible estimation errors is reduced due to the number of machine learning means implemented.

[0013] Preferably, among the characteristics used to estimate the clamping force, at least the temperature of the electric motor rotor and the wear of the brake friction elements are used.

[0014] In a very advantageous way, each characteristic estimated at time t-1 obtained from estimated values ​​of this characteristic is used to estimate the characteristic at time t.

[0015] The estimation system thus designed includes in its storage area either the machine learning means, or the calculation functions established by these machine learning means.

[0016] In a preferred example, the features used in the estimation of the clamping force were selected from a very large number of features using machine learning means.

[0017] In other words, the manufacturing process implements a plurality of machine learning means whose results are combined to provide the estimated value of the clamping force.

[0018] The present invention relates to a method for manufacturing a system for estimating the clamping force exerted by an electric brake of a motor vehicle, said estimation system comprising a storage area including executable machine instructions, a memory and a processor for generating an estimated clamping force, said method comprising the steps: a) Provision of a first group of several machine learning methods for determining an estimated clamping force, b) Providing the selected brake characteristic values ​​and clamping force values ​​measured by means of a sensor to each of the machine learning means of the first group, c) Generation by each of the machine learning methods of the first group of a clamping force value, d) Combining said clamping force values ​​to generate an estimated clamping force using a first combination function, e) Comparison of said estimated clamping force to the measured clamping force value, f) Adaptation of each of the machine learning methods and repetition of steps b) to e) until the estimated clamping force and the measured clamping force are within a given error threshold interval, g) Integration of the first group of machine learning means into the storage area or extraction of the computation functions of each of the machine learning means of the first group and integration of said functions into the storage area and integration of the first combination function.

[0019] In an advantageous example, step b) comprises the step of generating at least one estimated braking characteristic, including the substeps of: a') Provision of a second group of several machine learning means for determining said estimated characteristic, b') Providing the values ​​of the selected brake characteristics and values ​​of said characteristic measured by means of a sensor to each of the machine learning means of the second group, c) Generation by each of the machine learning methods of the first group of a value of said characteristic, d') Combining said values ​​of said feature to generate an estimated value of feature by means of a second combination function, e') Comparing said estimated value of feature to the value of the measured feature, f') Adapting each of the machine learning methods of the second group and repeating steps b') to e') until the estimated value of feature and the measured feature are within a given error threshold interval, g') Integration of the second group of machine learning means into the storage area or extraction of the computation functions of each of the machine learning means of the second group and integration of said functions into the storage area, and integration of the second combination function.

[0020] In a preferred example, at least one other second group of several machine learning means for determining another estimated feature is provided, and in which steps b') to g') are carried out.

[0021] For example, the estimated characteristic is the temperature and the other estimated characteristic is the wear of the friction element(s).

[0022] The manufacturing process according to the invention may include, prior to step a), a step of selecting the selected characteristics obtained at least in part by means of machine learning.

[0023] Step d) and / or step d') may implement an arithmetic mean, an average, a weighted sum and / or machine learning means.

[0024] The present invention also relates to a system for estimating the clamping force of an electric brake of a vehicle manufactured according to the invention, comprising a storage area in which at least the first group of machine learning means or the calculation functions of each of the machine learning means are written.

[0025] In an example embodiment, the second group or groups of machine learning means or the computation functions of each of the machine learning means of the second group or groups are written in the storage area.

[0026] The estimation system is advantageously configured to take into account the estimated value of the clamping force at time t-1 when calculating the estimated value of the clamping force at time t.

[0027] The present invention also relates to a braking system for a motor vehicle comprising at least one brake disposed at the level of a wheel, said brake being at least partly electrically actuated, a control unit for said brake and a system for estimating the clamping force exerted by said brake according to the invention and means for providing the selected characteristics to said estimation system.

[0028] The brake is for example a parking brake and / or an electric service brake. BRIEF DESCRIPTION OF THE FIGURES

[0029] The following description will be better understood with the aid of the attached drawings, in which: - [Fig. 1] is a schematic representation of a vehicle comprising a braking system capable of implementing the invention, - [Fig.2] is a schematic representation of an example of a manufacturing process for a clamping force estimation system according to the invention, - [Fig.3] is a schematic representation of an example of a clamping force estimation system according to the invention applied by an electric brake, - [Fig.4] is a schematic representation of an example of a learning method for the estimation system according to the invention, - [Fig. 5] is a schematic representation of an example of a braking control by a braking system comprising an estimation system according to the invention, - [Fig. 6] is a schematic representation of another example of a system for estimating the clamping force applied by an electric brake according to the invention. DETAILED DESCRIPTION OF EMBODIMENTS

[0030] In [Fig.1], we can see a vehicle V, represented schematically, comprising a braking system including brakes FR equipping each wheel.

[0031] In this example, service braking can be provided by hydraulic brakes or electric brakes.

[0032] The braking system also includes a parking braking device comprising at least a first parking brake FP1 at the right rear wheel and a second parking brake FP2 at the left rear wheel.

[0033] The FP1 and FP2 parking brakes are electric parking brakes.

[0034] Generally the parking brake is integrated into the service brake.

[0035] Each electric parking brake comprises an actuator equipped with an electric motor and means for converting the rotational motion of the electric motor into a translational motion applying the brake pads against the brake disc or the brake shoes against the drum. The electric brake can be equipped with any type of electric motor, for example, a DC motor, such as a brushless DC motor.

[0036] The braking system may or may not include an ABS and / or ESP control unit.

[0037] The braking system includes an electronic control unit (ECU) (Electronic control unit in Anglo-Saxon terminology) for controlling the parking brakes FP1 and FP2. The parking brakes are controlled, for example, by pressing a button B located in the passenger compartment.

[0038] The invention will be described in the context of an application to one of the electric brakes, but it will be understood that the invention applies to every electric brake, including electromechanical brakes implemented for service braking.

[0039] Figure 2 shows an example of a manufacturing process for a system S for estimating the clamping force applied by an electric brake obtained by the manufacturing process according to the invention.

[0040] The estimation system can be developed for an existing brake that is to be adapted to a new vehicle (new platform). To do this, the instrumented brake is mounted on the new vehicle, data is acquired, and machine learning means are trained on the basis of this data.

[0041] The estimation system can also be developed for a newly developed brake that is to be fitted to a vehicle. The new instrumented brake is Mounted on the vehicle, data is acquired and machine learning systems are trained based on this data. The estimation system is integrated into the computer.

[0042] According to a first embodiment, the estimation system ([Fig.3]) comprises a processor P which is advantageously, for example, a graphics card, a memory area M, and a storage area Z in which executable machine instructions I and a plurality of machine learning means are loaded.

[0043] The manufacturing process for the estimation system comprises: - A step 100 of providing a first group of several machine learning methods ML1, ML2, ML3, ... MLn for determining an estimated clamping force, - a step 200 of supplying selected brake characteristics Cl, C2... Ck and clamping force values ​​measured by means of a sensor to each of the machine learning means ML1, ML2, ML3, .. .MLn of the first group, - a 300th generation step by each of the machine learning means of the first group of a clamping force value Festl, Fest2...Festn, - a step 400 combining said clamping force values ​​Festl, Fest2, ... Festn to generate an estimated clamping force, - a step 500 comparing said estimated clamping force to the measured clamping force value, - a step 600 of adapting each of the machine learning methods, - a step 700 repeating steps 200 to 600 until the estimated clamping force and the measured clamping force are within a given error threshold range, - a step 800 of Integration of the first group of machine learning means in the storage area.

[0044] We will now go over the different elements implemented by the manufacturing process.

[0045] Figure 3 shows an example of a clamping force estimation system obtained by the manufacturing process according to the invention.

[0046] The estimation system thus fabricated comprises several machine learning means ML1, ML2, .. .MLn; each configured to provide an estimated value of the clamping force Festl, Fest2, .. .Festn and means 4 to combine the estimated clamping forces Festl, Fest2, .. .Festn and provide a final clamping force value Fest.

[0047] In this application, "machine learning means" means artificial intelligence means, for example, artificial neural networks implementing one or more algorithms and establishing a function for calculating an output value from one or more input values. This function is obtained by learning using "training data" or "dataset" in Anglo-Saxon terminology. This data is acquired at least in part by means of one or more real sensors on a brake fitted to a vehicle in motion or possibly on a test bench.

[0048] The neural network, presented here as an example, comprises several units, called neurons, connected to each other by connections, called synapses, and between which signals are transmitted. Each neuron receives an input signal which is modified by weights and biases, and generates an output signal which is transmitted to the other neurons. The neural network can be hardware and / or software. Furthermore, the neural network can comprise several layers.

[0049] By way of example, the machine learning means are selected from XGboost Regressor, Extra Tree regressor + Bagging Regressor, Gradient Boosting Regressor, Decision Tree Regressor, Support Vector Regressor (SVR), MLP Regressor (Neural Network), Random Forest Regressor.

[0050] The means for combining 4 of the values ​​Festl, Fest2,..., Festn include, for example, means for performing an arithmetic mean, a weighted mean, an average... In the case of a weighted mean, the weight of each value can be determined by machine learning means.

[0051] Implementing several machine learning methods to determine the clamping force reduces the effect of an error in the estimation performed by one of the machine learning methods. Furthermore, it allows for the consideration of a large number of features.

[0052] Each of the machine learning means receives as input signals data from the brake, the vehicle, its environment and estimates a clamping force Fest.

[0053] Very advantageously, the estimation system includes means 9 The estimated clamping force Fest at time t-1 is fed back into the machine learning tools ML1, ML2,... MLn to estimate the clamping force at time t. For example, the clamping force is estimated every 50 ms. Using the value estimated at t-1 to estimate the value at t allows for the detection of discrepancies in the estimation and can potentially generate an alert on these machine learning tools.

[0054] By injecting the estimate made at time t-1 into the machine learning means to estimate the clamping force at time t, the learning means are informed of the latest estimated values, which improves the performance of the learning methods. Very advantageously, such information can be used to prevent the learning methods from providing an estimated value at time t that is inconsistent with previously estimated values.

[0055] It will be understood that the means 9 can reinject one or more values ​​estimated at t-1, t-2...tn.

[0056] The input data consists of characteristics Cl, C2... Ck of the brake and its environment, such as the vehicle. Some of the characteristics are measured by a real sensor and designated CR, while others are estimated by mathematical models or machine learning methods. The estimated characteristics use, in particular, values ​​measured by sensors and information about the brake, the vehicle, and / or the environment. The CR characteristics are provided to the system, which processes them and estimates other characteristics.

[0057] The group of characteristics Cl, C2... Ck among which the characteristics used include, for example and without limitation: vehicle acceleration, vehicle deceleration, brake motor rotor position or piston position, brake pad temperature, brake disc temperature, brake pad wear which can be measured or estimated, braking force applied to the brake pedal, vehicle speed, speed during brake activation and deactivation, braking duration, known vehicle mass, time interval between two braking operations, number of braking operations, road condition, road gradient, brake pad thickness, vehicle type, number of pads per brake, brake pad type, coefficient of friction, brake pad surface area, wear coefficient,tire pressure, weather conditions such as humidity, ambient temperature...

[0058] By way of example only, the vehicle includes, at each brake, a wheel rotation sensor, a sensor for the supply voltage of the brake's electric motor, and a sensor for the current consumed by the electric motor. The data then provided to the machine learning means are the wheel speed, the electrical voltage applied to the motor, and the current consumed by the motor.

[0059] The characteristics that are estimated are, for example, the stroke of the electric brake piston.

[0060] It will be understood that these lists are given only as examples, and that they may vary depending on the brake, the vehicle...

[0061] In a preferred example, the manufacturing process provides for one or more groups of machine learning means to estimate some of the data input data is used to estimate the clamping force. Preferably, and as will be described below, all or part of the input data is itself estimated by implementing several machine learning methods, the output values ​​of which are combined to provide a final estimated value.

[0062] Particularly advantageously, at least the rotor temperature of the electric brake motor and the wear of the friction elements are estimated by implementing several machine learning methods. The development of these machine learning methods is achieved by repeating the steps of [Fig. 2],

[0063] In [Fig. 3], the estimation system comprises 6 estimation means for the temperature obtained by the process according to the invention. The estimation means 6 comprise several machine learning means TML1, TML2...TMLm, each estimating the temperature T and providing values ​​Testl, Test2...Testm.

[0064] The different machine learning methods TML1, TML2.. .TMLm can be constructed with different parameter sets and trained with different data sets.

[0065] The means 6 also include means 8 for combining the estimated temperature values ​​Test, Test2...Testm and delivering a final estimated temperature value Test. Similar to the combining means 4, the means 8 include, for example, means for calculating an arithmetic mean, a weighted mean... Alternatively, the combining means include machine learning means.

[0066] Very advantageously, the estimation system also includes means 13 for re-injecting the estimated temperature Test at time t into the machine learning means TML1, TML2,... .TMLm to estimate the temperature at time t+1.

[0067] The system includes means 10 for estimating the wear obtained by the process according to the invention. The means 10 include several machine learning means WML1, WML2...WMLp, each estimating the wear and providing Westl, West2...Westp values.

[0068] The means 10 also include means 12 for combining the estimated wear values ​​West1, West2, ..., Westp and delivering a final estimated wear value West. Similar to the combining means 4 and 8, the means 12 include, for example, means for calculating an arithmetic mean, a weighted mean, etc. Alternatively, the combining means include machine learning means.

[0069] Very advantageously, the estimation system also includes means 14 for reinjecting the estimated wear West at time t into the machine learning means WML1, WML2,... WMLp to estimate the wear at time t+1.

[0070] The estimated values ​​Test and West are provided to each of the machine learning means ML1, ML2.. .MLn implemented to estimate the clamping force Fest.

[0071] The number of machine learning means implemented to estimate clamping force, temperature and wear may be the same or different.

[0072] In the example shown, the same characteristics are provided as input to the machine learning means TML1, TML2.. .TMLm and WML1, WML2... WMLp, as well as to the machine learning means ML1, ML2...MLn, the latter receiving in addition the estimated temperature Test and the estimated wear West.

[0073] In another example, the features provided for estimating clamping force, temperature, and wear are different and advantageously selected as described below. The set of features selected for estimating temperature is designated CT, the set of features selected for estimating wear is designated CW, and the set of features selected for estimating clamping force is designated CS. The initial sets of features from which a selection is made may differ among the various machine learning methods implemented for estimating either temperature, clamping force, or wear.

[0074] According to the manufacturing process, each of the machine learning means ML1, ML2.. .MLn, TML1, TML2.. .TMLm and WML1, WML2.. .WMLp is trained prior to its integration into the memory area of ​​the estimation system.

[0075] In a particularly advantageous example, prior to step 200 of supplying brake characteristics, a step of selecting said characteristics from among a large number of characteristics takes place using at least in part machine learning means.

[0076] The machine learning means process all the characteristics, checks the influence of each of them on the clamping force and their interdependence.

[0077] A learning phase of the automatic learning means for estimating the clamping force then takes place.

[0078] This learning phase is carried out on the basis of the data acquired for the selected characteristics, this data being acquired by means of the clamping force sensor located on the brake and other sensors and mathematical models, and / or machine learning means.

[0079] This data acquisition is referred to as the calibration step; the brake is then instrumented, notably with a force sensor and possibly other sensors, including a temperature sensor. This calibration step takes place either on a A test vehicle that moves on a circuit, or on a test bench. In this case, since the wear of the friction elements is taken into account, the duration of the calibration step is long enough to obtain substantial wear of the friction elements, or even total wear.

[0080] The measured values, which the machine learning methods will approximate during their training until the estimated values ​​are sufficiently close to the measured values, are referred to as "ground truth" in Anglo-Saxon terminology. GTF designates the measured values ​​of the clamping force, GTT designates the measured values ​​of the temperature, and GTW designates the measured values ​​of the wear.

[0081] We then have a set of measured clamping force values ​​associated with the values ​​of the selected characteristics CFA.

[0082] The machine learning means for temperature estimation and those for wear estimation are also trained during a learning phase. For these learning processes, a set of measured temperature values ​​associated with the values ​​of a number of measured and estimated characteristics (CTA), and a set of measured wear values ​​associated with the values ​​of a number of measured and estimated characteristics (CWA) are also available.

[0083] This learning phase is shown schematically in [Fig.4].

[0084] Each of the machine learning means is trained separately.

[0085] Learning takes place on the basis of selected features which may differ for each of the groups of machine learning means ML1, ML2.. .MLn, TML1, TML2.. .TMLm and WML1, WML2.. .WMLp.

[0086] The group of characteristics selected to estimate the temperature is designated CT, the group of characteristics selected to estimate the wear is designated CW and the group of characteristics selected to estimate the clamping force is designated CS.

[0087] By way of example, the CT group includes the pressure exerted on the rotor, the surface area of ​​the friction elements, the thickness of the friction elements, the wear rate...

[0088] In one example, the learning of each machine in the same group is based on a selection of features different from that used for the learning of the other machine learning methods. In another example, the same selection is used by all the machine learning methods in the same group.

[0089] For training purposes, the ground truth and selected features are provided to each of the machine learning methods ML1, ML2...MLn, TML1, TML2...TMLm and WML1, WML2...WMLp to enable them to train. To this end, each machine learning method is subjected to an error threshold interval between the estimated values ​​and the ground truth.

[0090] For example, for the machine learning means ML1, ML2.. .MLn intended to estimate the clamping force, a deviation of 1% is imposed.

[0091] For the machine learning means TML1, TML2.. .TMLm intended to estimate the temperature of the electric motor rotor, a maximum deviation of 20°C is fixed.

[0092] For the machine learning means WML1, WML2.. .WMLp intended to estimate the wear of the friction elements, a maximum deviation of 500 pm is fixed.

[0093] The measured and estimated characteristics and the threshold intervals used for learning are referred to as "training data" or "dataset" in Anglo-Saxon terminology.

[0094] During training, each of the machine learning means runs an algorithm on the training dataset and optimizes the algorithm and adapts its configuration, for example its rules and data structures, so that the estimated value is within the error threshold interval.

[0095] A backpropagation loop (RF, RT, RW) is applied to each of the machine learning means, in which the estimated value is compared to the measured value (GTF, GTT, GTW). When the estimated value is outside the expected threshold range, the machine learning means adapt to bring the estimated value closer to the ground truth value. In the case where the machine learning means include a neural network, for example, the weights and biases are modified, and the connections and / or layers of the neural network are modified.

[0096] The input signals are again fed into the modified machine learning means, a new value is estimated and compared to the ground truth value. When the estimated value and the expected value are within the limits of the expected error threshold, the machine learning means are considered optimal.

[0097] Machine learning means can be trained to take into account the aging of the brake and / or the vehicle.

[0098] In an example embodiment, all the means of machine learning thus trained are integrated into an estimation system.

[0099] In another embodiment, a selection of the most effective machine learning methods is made. The selected machine learning methods are then integrated into the estimation system.

[0100] In one example, the selection of machine learning methods is carried out manually. In another example, the selection of machine learning methods is carried out using machine learning methods.

[0101] In one example, the same clamping force estimation system is implemented to estimate the clamping force of the right rear brake and the left rear brake. In another example, a learning process is performed for the right rear brake and a learning process is performed for the left rear brake.

[0102] An example of brake control is shown schematically in [Fig.5].

[0103] The user activates the electric parking brake. The control unit sends a command to activate the electric motors of each parking brake so that they each apply a given clamping force, for example, to a brake disc in the case of disc brakes. The given clamping forces depend, in particular, on the incline on which the vehicle is located.

[0104] In order to verify whether the clamping forces actually applied are those commanded, the clamping force applied by each of the parking brakes is estimated by the estimation system according to the invention.

[0105] At time t, the rotor temperature Test(t) is estimated from selected characteristics CT, and the wear of the friction elements West(t) is estimated from characteristics CW. Test(t) is estimated by means 6 and West(t) is estimated by means 10.

[0106] T(est), W(est) and other features are provided to the machine learning means ML1, ML2,... .MLn to estimate Fest.

[0107] This value Fest(t) is compared to an expected value Fth. An instruction I is sent to the brake FR based on the result of the comparison. If Fest and Fth are within the tolerance range, the braking is enabled and the brake does not apply additional clamping force; otherwise, the motor(s) are reactivated to modify the clamping force.

[0108] In another example of an embodiment represented schematically in [Fig.6], the estimation system behaves in its storage area Z, not the learning means but the calculation functions of each of the machine learning means and the function of combining the results provided by each of these functions.

[0109] Indeed, at the end of the learning, we obtain for each of the automatic learning means for estimating the clamping force ML1, ML2, ... MLn, a function fl, f2.. .fn.

[0110] Similarly, in the case where temperature and wear are estimated in the same way, we obtain for each of the machine learning means for temperature estimation TML1, TML2, .. .TMLm gl, g2.. .gm; we obtain for each of the machine learning means for wear estimation WML1, WML2, ...WMLp hl, h2...hp.

[0111] This makes it possible to implement an estimation system using a standard processor.

[0112] The operation of this estimation system is as follows.

[0113] The CS characteristics are provided to the processor, which calculates Test and West from the functions gl, g2, ..., gm and hl, h2, hp, respectively. The values ​​calculated by the functions gl, g2, ..., gm are combined by the selected combination function gc, and the values ​​calculated by the functions hl, h2, ..., hp are combined by the selected combination function hc. These calculated values, along with the other selected characteristics, are provided to the functions fl, f2, ..., fn, which calculate the values ​​festl, fest2, ..., festn. These values ​​are combined by the selected combination function fc, providing the value Fest.

[0114] Advantageously, Fest(tl), Test(tl) and West(tl) are used by the calculation functions of Fest, Test and West respectively.

[0115] These estimated values ​​at t-1 will be part of the input characteristics / parameters for training machine learning models. Thus, Fest(tl), Test(tl) and West(tl) will be part of CF, CT and CW respectively.

[0116] The clamping force estimation system also applies to estimating the clamping force of an electric service brake. Furthermore, the invention applies to both disc brakes and drum brakes. REFERENCES

[0117] 4: means of combining Fest1, Fest2,... Festn 6: Methods of estimating temperature 8: Means of combining Test1, Test2...Testm 9: means to reinject Fest(tl) 10: Methods for estimating wear and tear 12: means of combining Westl, West2...Westp 13: means to reinject Test(tl) 14: means to reinject West(tl) 100, 200, 300, 400, 500, 600, 700, 200-600: Manufacturing process steps B: Button Cl, C2, C2..., CR: brake characteristics CS: Selected characteristic group for estimating clamping force; CT: Selected characteristic group for estimating temperature; CW: Selected characteristic group for estimating wear; CFA: Set of measured clamping force values ​​associated with the selected characteristic values CTA: set of measured temperature values ​​associated with the values ​​of the selected characteristics CWA: set of measured wear values ​​associated with the values ​​of the selected characteristics fl, f2.. .fn, g2,.. .gm, h2, h2,.. .hp, f2,.. .fn: calculation functions ML1, ML2, ML3,...MLn: machine learning methods for estimating clamping force TML1, TML2, TML3,...TMLm: machine learning methods for estimating temperature WML1, WML2, WML3,...MLp: machine learning methods for estimating wear Festl, Fest2,...Festn: estimated clamping forces Testl, Test2...Testm: estimated temperatures Westl, West2...Westp: estimated wear Fth: expected clamping force value FR: brake FP1: First parking brake P2: Second GTF ground truth parking brake clamping force GTF: Ground Truth Clamping Force GTT: temperature ground truth TW: Ground Thruth Wear I: Executable machine instructions M: memory area P: processor; S: clamping force estimation system RF, RT, RW: backpropagation loop Test: Estimated final temperature is: Estimated final wear West: final estimated wear value V: vehicle Z: storage area

Claims

Demands

1. A method for manufacturing a system for estimating the clamping force exerted by an electric brake of a motor vehicle, said estimation system comprising a storage area (Z) including executable machine instructions, a memory, and a processor (P) for generating an estimated clamping force, said method comprising the steps: a) Providing a first group of several machine learning means for determining an estimated clamping force, b) Providing selected brake characteristic values ​​and clamping force values ​​measured by means of a sensor to each of the machine learning means of the first group, c) Generating a clamping force value by each of the machine learning means of the first group, d) Combining said clamping force values ​​to generate an estimated clamping force by means of a first combination function,e) Comparison of said estimated clamping force with the measured clamping force value, f) Adaptation of each of the machine learning means and repetition of steps b) to e) until the estimated clamping force and the measured clamping force are within a given error threshold interval, g) Integration of the first group of machine learning means into the storage area or extraction of the computational functions of each of the machine learning means of the first group and integration of said functions into the storage area and integration of the first combination function.

2. A manufacturing method according to claim 1, wherein step b) comprises the step of generating at least one estimated braking characteristic comprising the substeps of: a') Providing a second group of several machine learning means for determining said estimated characteristic, b') Providing the values ​​of the selected brake characteristics and values ​​of said characteristic measured by means of a sensor to each of the machine learning means of the second group, c') Generation by each of the machine learning means of the first group of a value of said characteristic, d') Combination of said values ​​of said characteristic to generate an estimated value of the characteristic by means of a second combination function, e') Comparison of said estimated value of the characteristic to the value of the measured characteristic, f) Adaptation of each of the machine learning means of the second group and repetition of steps b') to e') until the estimated value of the characteristic and the measured characteristic are within a given error threshold interval,(g') Integration of the second group of machine learning means into the storage area, or extraction of the computational functions from each of the machine learning means of the second group and integration of said functions into the storage area, and integration of the second combination function.

3. A manufacturing method according to claim 2, wherein at least one other second group of several machine learning means for determining another estimated characteristic is provided and wherein steps b') to g') are carried out.

4. A manufacturing method according to claim 3, wherein the estimated characteristic is temperature and the other estimated characteristic is the wear of the friction element(s).

5. A manufacturing method according to any one of claims 1 to 4, comprising prior to step a), a step of selecting the selected features obtained at least in part by machine learning means.

6. A manufacturing method according to any one of claims 1 to 5, wherein step d) and / or step d') implement an arithmetic mean, an average, a weighted sum and / or machine learning means.

7. A system for estimating the clamping force of an electric brake of a vehicle manufactured according to any one of the claims previous ones, comprising a storage area in which at least the first group of machine learning means or the computational functions of each of the machine learning means are written.

8. Estimation system according to claim 7, wherein the second group or groups of machine learning means or the computation functions of each of the machine learning means of the second group or groups are written in the storage area.

9. Estimation system according to claim 7 or 8, configured to take into account the estimated value of the clamping force at time t-1 when calculating the estimated value of the clamping force at time

10. 1. Braking system of a motor vehicle comprising at least one brake disposed at a wheel, said brake being at least partly electrically actuated, a control unit of said brake and a system for estimating the clamping force exerted by said brake according to claim 7, 8 or 9 and means for providing the selected characteristics to said estimation system.

11. Braking system according to the preceding claim, wherein the brake is a parking brake and / or an electric service brake.

Images