Method for actuating an electrohydraulic actuator

Abstract

The invention relates to a method for an optimized actuation of an electrohydraulic actuator and a corresponding training method for a neural network, to a neural network, and to a method for using the neural network and a corresponding device in order to carry out the method. The method for an optimized actuation of an electrohydraulic actuator has the steps of applying (S1) a target current curve to the electrohydraulic actuator, measuring (S2) the actual current curve set on the electrohydraulic actuator, repeatedly carrying out the applying (S1) and measuring (S2) steps for a plurality of parameterizations, determining (S3) correction values on the basis of the target current curve and the actual current curve, and actuating (S4) the electrohydraulic actuator on the basis of the determined correction values.

Description

[0001] Method for controlling an electrohydraulic actuator

[0002] Technical area

[0003] The present invention relates to a method for the optimized control of an electrohydraulic actuator and an associated training method for a neural network, a neural network and a method for using the neural network as well as an associated device for carrying out the methods.

[0004] State of the art

[0005] When using electro-hydraulic actuators, a current signal is fed into the electro-hydraulic actuator. Based on the current signal, mechanical energy is then converted into pressure energy of the hydraulic oil by a power element, and the hydraulic energy is converted into mechanical energy with the help of actuators. First, the target current value to be applied is determined based on a pressure specification using a stored characteristic map. The actual current flowing is then regulated by a current controller. The control units used in industrial technology are usually software current controllers because they are inexpensive. Due to insufficient accuracy, however, the target current value and the actual current value differ significantly, especially at low effective current values. This can lead to the actual pressure, which results in a clutch path, for example, being higher than the desired pressure.This can be problematic, for example, when using electro-hydraulic actuators in industrial engineering during inching maneuvers and during pressure equalization phases during power shifting.

[0006] From the prior art DE 10 2016202 999 A1, a method for controlling pressure regulators using a dither function is known, within which the control transfer functions are determined for each switching element path and system pressure or supply pressure path of a slide pressure regulator depending on defined system input variables serving as input signals. Description of the invention

[0007] In a first aspect, the present invention relates to a method for the optimized control of an electrohydraulic actuator. An electrohydraulic actuator generally has a standard current signal as its input. Based on the current signal, mechanical energy is then converted into pressure energy of the hydraulic oil by a power element, and the hydraulic energy is converted into mechanical energy using actuators.

[0008] In a first step, a target current curve is applied to the electrohydraulic actuator. This means that, for example, certain known target current values ​​are applied to the electrohydraulic actuator in a sequential sequence. These various target current values ​​together form the target current curve.

[0009] In a further step, the actual current curve occurring at the electrohydraulic actuator is measured. For this purpose, the current is recorded using a shunt installed on a control unit, for example. This also results in a multitude of actual current measured values, which together represent an actual current curve.

[0010] The aforementioned steps of applying a target current curve and measuring the resulting actual current curve are then performed for a variety of parameter settings of the electrohydraulic actuator. Such parameter settings can, for example, be different P, I, and D components of a PID controller. Other parameters required for control can also be mapped sequentially.

[0011] Then, based on the target current curves determined in this way and the actual current curves determined in this way, which result from the repeated execution of the steps, corresponding correction values ​​are determined. These correction values ​​allow the deviation between the respective target current and the respective actual current to be compensated for. For example, a calculation of the correction values ​​from the actual current and the target current is possible as a determination. The result of the determination can therefore be a large number of correction values ​​which, depending on the respective parameterization of the electro-hydraulic actuator as well as on the current target current and the current actual current, indicate the correction value with which a corresponding correction must be made in order to minimize the deviation between the target and actual values.

[0012] In a further step, the electrohydraulic actuator can then be controlled based on the determined correction values.

[0013] The above-mentioned method thus makes it possible to use a current controller with insufficient accuracy, especially at low

[0014] Current effective values, by means of the application of a correction factor, to minimize the deviation between the target current and the actual current and thus to control the electro-hydraulic actuator in such a way that there is only a small difference between the desired pressure and the actually output pressure.

[0015] In one aspect, it is also possible to measure the actual current curve and the target current curve repeatedly, for example, three times, for each parameterization in order to satisfy the measurement uncertainty. Furthermore, it is possible to discretize the actual current curve and the target current curve, as well as any other parameterization values, such that the influence of the parameter change is greater than the measurement uncertainty during the repeated measurements.

[0016] In one aspect, the effective value, or Root Mean Square Value, can be used to determine the current values.

[0017] In a further aspect, in addition to the aforementioned values, the inductance and resistance of the actuator can also be used to generalize the method. For this purpose, the steps of applying, measuring, and repeating the measurement are performed for a large number of actuators, each with a different inductance and resistance. The step of determining the correction values ​​is then performed depending on the actual current curve, the target current curve, the parameterization, and the respective inductance and resistance of the actuators, and the result is stored, for example, in a structured data set.

[0018] Accordingly, the determined correction values ​​depend not only on the actual current, target current, and parameterization, but also on the inductance and resistance of the actuator. This allows for generalization to different actuators with different inductances and resistances.

[0019] A structured data set can be, for example, a look-up table. The control can either read the data from the look-up table or interpolate it using the look-up table, with the values ​​in the look-up table serving as support points.

[0020] In order to allow a better prediction of the necessary correction value and also to allow a better prediction for actuators that were not directly measured, the use of a neural network is also possible in one aspect.

[0021] To do this, the target current curve, the actual current curve, the parameterization, and a desired correction value are first entered into a neural network as training data in the step of determining correction values. For example, the individual values ​​of the actual current curve and the target current curve can be entered into the neural network.

[0022] Since continuous-time data is not typically used here, the determined average current, i.e., the effective value, can serve as the input value. The desired correction value between the actual current and the target current serves as the label of the neural network. This correction value can be determined, for example, mathematically. Then, in the determination step, a target current, an actual current, and the parameterization are entered into the trained neural network. Here, too, the effective value can be used for the input, as described above. The neural network then predicts the correction value it receives as output. The electrohydraulic actuator can then be controlled using this correction value determined by the neural network.

[0023] The neural network can, for example, be an FCNN.

[0024] The use of a neural network enables a more reliable determination of the correction value and, at the same time, a generalization and thus a more general applicability to different actuators and different parameterizations.

[0025] According to the above explanations, in the step of determining, the respective resistance of the actuators and the respective inductance of actuators can also be part of the training data, whereby training also takes place for a large number of different actuators with different resistances and inductances.

[0026] In the determination step, the corresponding inductance and the corresponding resistance of the actuator are then entered into the neural network as input.

[0027] Thus, this approach allows further generalization for different electrohydraulic actuators.

[0028] Another aspect concerns a training method for training a neural network for the optimized control of an electrohydraulic actuator.

[0029] This method initially involves receiving data sets, each containing a target current curve applied to an electrohydraulic actuator, and an actual current curve that occurs on the electrohydraulic actuator after the target current curve has been applied. Furthermore, the data sets also include the corresponding parameterization of the electrohydraulic actuator, as well as, if applicable, an inductance and a resistance of the electrohydraulic actuator. A desired correction value, which can be determined mathematically from the actual current and the target current, is also included in the training data.

[0030] Regarding the definition and the data to be used for this purpose and how to obtain them, reference is made to the above explanations.

[0031] The neural network is then trained by inputting the data sets as training data into the neural network. It should be noted again that the RMS value can be used for the current instead of a continuous-time data set.

[0032] A further aspect then includes the neural network, which was trained using the above-mentioned method.

[0033] A further aspect then includes a method for the optimized control of an electrohydraulic actuator.

[0034] For this purpose, as already described above with reference to the aforementioned methods, a target current, an actual current, a parameterization, as well as an inductance and a resistance of an electrohydraulic actuator are first input into the previously discussed, trained neural network. This then outputs a correction value, which can then be used to control an electrohydraulic actuator.

[0035] A further aspect of the present invention comprises an apparatus for carrying out a method according to any one of the preceding claims.

[0036] A further aspect includes an electrohydraulic actuator with correction values ​​determined using the preceding method.

[0037] Brief description of the figures Figure 1 shows a flow diagram of a method according to one aspect.

[0038] Figure 2 shows an example of a neural network according to one aspect.

[0039] Detailed description of embodiments

[0040] Figure 1 shows a flow diagram of a method for the optimized control of an electrohydraulic actuator. This initially comprises step S1 of applying a target current curve to the electrohydraulic actuator. For example, a linearly increasing and subsequently decreasing target current is applied to the actuator. Then, in step S2, the actual current curve occurring at the electrohydraulic actuator is measured. In the present example, these steps are repeated several times for each parameterization in order to compensate for any corresponding measurement uncertainty. Then, as shown by the backward arrow between step S2 and step S1 in Figure 1, steps S1 and S2 are carried out for a large number of different parameterizations. This means that various parameters of the electrohydraulic actuator are set and the corresponding current curve is applied again.Then, as also indicated by the backward arrow shown here between step S2 and step S1, the measurements are repeated for a large number of actuators, each with different inductances and different resistances. The result after step S2 is a data set with numerous dependencies between the actual current curve, the target current curve, the parameterization, the resistance of the actuators, and the inductance of the actuators. Then, in step S3, a necessary correction value is determined, for example, calculated. In the example shown here, the data sets obtained above are entered into a neural network as training data to determine the correction values.This means that in step S3a, the target current curve, the actual current curve, the parameterization, the respective inductance and resistance of the actuators, and a desired correction value are entered into the neural network as training data. If a necessary correction value is then to be determined for a specific actuator in a specific situation, a data set consisting of a target current, an actual current, a parameterization, and an inductance and resistance of the actuator can be entered into the thus trained neural network in step S3b. The output of the neural network is then the correction value, which can be used to control the electrohydraulic actuator in step S4.

[0041] Figure 2 shows the corresponding inputs and outputs of a neural network. In Figure 2, the neural network is implemented as an FCNN. As described above, the parameterization, the target current, the actual current, as well as the resistance and inductance of the electrohydraulic actuator serve as input values, and the correction value serves as the output value. Accordingly, during training, the correction value is used as the label of the neural network.

[0042] Reference symbol

[0043] 51 Applying a target current curve

[0044] 52 Measuring the actual current curve

[0045] 53 Determining correction values

[0046] S3a Training of the neural network

[0047] S3b Determining correction values ​​using the neural network

[0048] 54 Control of the electrohydraulic actuator

Claims

Patent claims 1 . Method for the optimized control of an electrohydraulic actuator, comprising the steps: - applying (S1) a target current curve to the electro-hydraulic actuator; - Measuring (S2) the actual current curve at the electro-hydraulic actuator; - Repeatedly performing the step of applying (S1) and measuring (S2) for a plurality of parameterizations; - Determining (S3) correction values ​​based on the target current curve and the actual current curve; - Control (S4) of the electro-hydraulic actuator based on the determined correction values.

2. The method according to claim 1, wherein - the steps of applying (S1), measuring (S2) and repeating are carried out for a plurality of actuators each having a different inductance and resistance; and wherein - in the step of determining (S3) the correction values ​​are determined depending on the actual current curve, the target current curve, the parameterization, the respective inductance and resistance of the actuators and stored in a structured data set.

3. The method according to claim 1, wherein - in the step of determining correction values, the target current curve, the actual current curve, the parameterization and a desired correction value are entered into a neural network as training data (S3a), and - in the step of determining, a target current, an actual current and the parameterization are input into the trained neural network and a correction value is obtained as output from the neural network (S3b), where - the electro-hydraulic actuator (S4) is controlled based on the determined correction value.

4. The method according to claim 2, wherein - in the step of determining correction values, the target current curve, the actual current curve, the parameterization, the respective inductance and the respective resistance of the actuators and a desired correction value are entered into a neural network as training data (S3a), and - in the step of determining, a target current, an actual current, an inductance and a resistance of an actuator and the parameterization are input into the trained neural network and a correction value is obtained as output from the neural network (S3b), wherein the electrohydraulic actuator (S4) is controlled based on the determined correction value.

5. Training method for training a neural network for the optimized control of an electrohydraulic actuator, comprising the steps: - Receiving data records with: - a target current curve applied to an electro-hydraulic actuator; - an actual current curve which is set at the electro-hydraulic actuator after the application (S1) of the target current curve; - a parameterization; - an inductance and a resistance of the electrohydraulic actuator; and - desired correction values; - Training the neural network by inputting the data sets as training data into a neural network (S3a).

6. Neural network trained by the method of claim 5.

7. Method for the optimized control of an electrohydraulic actuator, comprising: - Input - a target current applied to an electro-hydraulic actuator; - a resulting actual current; - a parameterization; and - an inductance and a resistance of the electrohydraulic actuator into the neural network according to claim 5; - Obtaining a correction value as output; - Controlling the electro-hydraulic actuator (S4) based on the determined correction value.

8. Apparatus for carrying out a method according to any one of the preceding claims.

9. Electrohydraulic actuator with correction values ​​determined by the method according to claim 7.

Images