Method for determining a change in the instantaneous demand of respective auxiliary consumers of a work machine by means of a neural network

Abstract

Method for determining a change in the instantaneous demand of respective auxiliary consumers (10) of a work machine by means of a neural network, the method comprising at least the following steps: - Continuously detecting (40) an output speed, an engine speed and an engine torque as input data; - inputting (42) the continuously acquired input data into the neural network; - determining (48) the change in torque demand at respective auxiliary consumers as output data by means of the neural network based on a relationship, learned by the neural network, between the continuously detected output speed, the continuously detected engine speed and the continuously detected engine torque and the change in torque demand at respective auxiliary consumers (10); - checking (44) whether the instantaneous requirement change determined by the neural network exceeds a significance threshold; and - Outputting (46) the instantaneous demand change of respective auxiliary consumers (10) determined by the neural network, provided that the test has shown that the significance threshold has been exceeded.

Description

Technical field

[0001] The present invention relates to a method for determining a change in the instantaneous demand of respective auxiliary consumers of a work machine using a neural network. The invention also relates to a method for training a neural network, a computer program product, and a work machine. State of the art

[0002] In the drive train of a work machine, it is usually known which torque is generated by a drive motor. In a work machine, however, this torque is not always completely made available to a drive output. Instead, a portion of the torque can be directed to the respective auxiliary consumers of the work machine. The effort required to record the number and type of active auxiliary consumers and their torque requirements can be very high. Therefore, it is usually not known how much torque is directed to the respective auxiliary consumers. Accordingly, it may be unknown, for example, which torque is made available to a drive transmission. For vehicle control, however, it is desirable to know the torque actually directed to an input of the drive transmission. For example, this torque is used to control the drive transmission, if available.This can significantly improve transmission control. For example, in certain vehicle conditions, it may be necessary to limit the maximum power delivered to the transmission. Otherwise, excessive transmission wear or even damage may occur.

[0003] The actual torque delivered to the respective auxiliary consumers is therefore estimated for active auxiliary consumers, for example. However, this estimate can be very inaccurate. A conservative design can therefore result in excessive engine torque reduction, resulting in a very low maximum available power. However, if the estimate is not conservative enough, excessive wear or even damage can still occur.

[0004] DE 101 15 045 A1 describes a control device for compensating a change in torque.

[0005] DE 10 2020 211 364 A1 describes a method and a device for operating a work device using artificial intelligence methods.

[0006] DE 10 2018 222 064 B4 describes an operating strategy for adjusting a tractive force taking into account the operating conditions of a working machine.

[0007] DE 10 2019 113 765 A1 describes a method for controlling a hydraulic system of a mobile work machine.

[0008] DE 10 2020 007 453 A1 describes a method for determining a vehicle mass average and its output in the vehicle.

[0009] DE 10 2019 008 212 A1 describes a method for using artificial neural networks to calculate engine operating points during operation of a motor vehicle with at least one drive motor. Description of the invention

[0010] A first aspect relates to a method for determining a change in the torque requirement of respective auxiliary consumers of a work machine using a neural network. A torque requirement for the drive motor can correspond to a drive power required to drive a traction drive and respective auxiliary consumers of the work machine. The change in the torque requirement can accordingly be, for example, a change in the drive torque required to continue operating a traction drive and all auxiliary consumers of the work machine as desired, for example due to the activation or deactivation of an auxiliary consumer.

[0011] The work machine can, for example, have a drive motor. A drive motor can, for example, be designed as a traction motor. The drive motor can, for example, be designed as an internal combustion engine or an electric machine. An engine torque can be a torque that is provided on a motor shaft. A work machine can, for example, be designed as an agricultural machine or a construction machine. An example of an agricultural machine is a tractor. An example of a construction machine is a wheel loader. An auxiliary consumer can, for example, be a fan, an air conditioning compressor, an alternator, a control hydraulic system, a power take-off shaft with devices connected to it, or even a working device on a power take-off shaft.

[0012] A neural network can be a program that can be used to universally approximate functions. In this process, weights can be assigned to each input data to generate output data. The structure of the network and the respective weights are determined, for example, during training of the neural network using a reward function and training data. The relationships between input data and output data are therefore not calculated analytically or specified manually.

[0013] The method comprises a step of continuously recording an output speed, an engine speed, and an engine torque as input data. The output speed can, for example, be a speed at an output shaft of the drive transmission or at a driven wheel of the work machine. The engine speed can, for example, be a speed of a motor shaft or a rotor of the drive motor. The engine speed can, for example, be a turbine speed. The engine speed can correspond to an input speed at the drive transmission. The engine torque can, for example, be recorded by a sensor or calculated by an engine control system for recording, for example using a table. Continuous recording can be achieved by individual recordings at discrete intervals or by a continuous measurement signal. Continuous recording enables the calculation of changes, for example of an engine torque gradient.The continuously recorded data can therefore also take gradient relationships into account as input data for the neural network.

[0014] The method includes a step of inputting the continuously acquired input data into the neural network. The neural network may have been previously trained with training data. Inputting can occur, for example, by transmitting the acquired data to the neural network. Gradients may have been previously calculated from the continuously acquired data, which are then fed to the neural network as input data. However, the continuously acquired data can also be fed to the neural network as acquired, without prior calculations.

[0015] The method comprises a step of determining the change in torque demand at respective auxiliary consumers as output data using the neural network based on a relationship, learned by the neural network, between the continuously recorded output speed, the continuously recorded engine speed, and the continuously recorded engine torque and the change in torque demand at respective auxiliary consumers. Thus, for example, the neural network outputs a change in torque demand at respective auxiliary consumers based on the input data. This allows corresponding actual input torques and their change to be determined at the transmission and made available for transmission control. The change in torque demand at respective auxiliary consumers can be determined by the trained neural network with little effort and high precision.

[0016] The method includes a step of checking whether the torque demand change determined by the neural network exceeds a significance threshold. If the check does not reveal that the significance threshold is exceeded, the torque demanded by the respective auxiliary consumers is determined to be unchanged. For this purpose, for example, output data from the neural network may not be output or may be changed to zero. The significance threshold can be used to filter out engine torque changes and torque demand changes due to external influences and respective work processes of the work machine, i.e., changes which are not attributable, for example, to a change in the state of auxiliary consumers. This can improve the accuracy of determining torque demand changes. The check can be used to filter output data from the neural network.

[0017] Alternatively or additionally, the input data can also be tested. For example, it can be checked whether a change in the recorded output speeds, recorded engine speeds, and recorded engine torques exceeds a significance threshold. For example, vibrations in the recorded data, such as the height, frequency, and shape of a vibration, can be considered as a significance threshold. This allows the input data for the neural network to be filtered. This can improve the accuracy of determining changes in torque requirements, as, for example, changes in input data due to external environmental influences can be filtered out.

[0018] The method includes a step of outputting the change in torque demand of respective auxiliary consumers determined by the neural network, provided the test has shown that the significance threshold has been exceeded. This enables improved control of the transmission and other parts of a drive train of the work machine. Complex sensor-based determination of changes in torque demand of respective auxiliary consumers can thus be dispensed with. The output data can, for example, only contain an overall change in torque demand. The determination can be particularly precise, and the neural network requires little computing power. However, the output data can also contain changes in torque demand assigned to specific auxiliary consumers. The output data can also contain information about which auxiliary consumers have changed their torque demand.Output data with more information allows for more further processing and more control signals for the working machine. When outputting, an unchanged instantaneous demand of the respective auxiliary consumers, i.e., a change in the instantaneous demand of the respective auxiliary consumers equal to zero, can be output if no exceedance of the significance threshold was detected during the step of checking whether the significance threshold has been exceeded.

[0019] The method may include a step of determining the changed engine torque demand on the drive motor as a function of the torque demand change of respective auxiliary consumers determined by the neural network. The method may include a step of determining an input torque of the transmission as a function of the torque demand change of respective auxiliary consumers determined by the neural network.

[0020] In a further embodiment of the method, the significance threshold has a minimum level of the change in torque requirement determined by the neural network. Below this minimum level, it can be assumed, for example, that no auxiliary consumer has been switched on or off. The minimum level can, for example, correspond to a torque requirement caused by switching on an auxiliary consumer with the lowest torque requirement. The minimum level can also be set such that below this level there is no significant influence on systems of the work machine. For example, the minimum level can be selected such that a control signal for the drive transmission is only changed when the minimum level is exceeded.

[0021] In a further embodiment of the method, the significance threshold has a minimum duration of the change in torque requirement determined by the neural network. For example, the minimum duration can be one second. This can avoid hysteresis in the control of the drive transmission. For example, a short-term activation or deactivation of auxiliary consumers can be ignored for the control of other systems, such as the drive transmission. Furthermore, the minimum duration can also be selected such that in the case of shorter changes in the input data and, alternatively or additionally, the output data, it can be assumed that these are not based on a change in the state of an auxiliary consumer. A change in the state of an auxiliary consumer can, for example, be activation, deactivation, or a changed control.

[0022] In a further embodiment of the method, it is provided that the method further comprises a step of determining a current instantaneous demand of respective auxiliary consumers by offsetting the output instantaneous demand change of respective auxiliary consumers against a previous instantaneous demand of respective auxiliary consumers. The previous instantaneous demand of respective auxiliary consumers can be a starting value, for example zero. The previous instantaneous demand of respective auxiliary consumers can be an instantaneous demand of respective auxiliary consumers stored in a previous iteration of the method. An output instantaneous demand change signal can be added to or subtracted from a memory block in this step. The addition or subtraction can occur depending on whether there has been a decrease or increase in instantaneous demands by respective auxiliary consumers.The idea behind this step is that the neural network only detects changes, for example, and therefore does not output an absolute instantaneous demand from the respective auxiliary consumers during continuous monitoring. The determined current instantaneous demand from the respective auxiliary consumers can easily be used to control the systems of the working machine.

[0023] In a further embodiment of the method, the neural network has a first subnetwork and a second subnetwork, the second subnetwork being connected in series behind the first subnetwork to the first subnetwork. Each subnetwork can correspond to a neural network. The first subnetwork receives, for example, the input data. The second subnetwork generates, for example, the output data. Connected in series here can mean that the first subnetwork generates intermediate data as output data, and the intermediate data is input to the second subnetwork as input data. The first subnetwork can carry out data extraction, for example by pattern recognition. The second subnetwork can carry out data interpretation, for example by evaluating the recognized pattern. The pattern can indicate a deviation in the input data. By providing two subnetworks, different subnetworks can be created for each task orFor each subtask of the analysis, a network structure should be chosen which is particularly suitable.

[0024] In a further embodiment of the method, the first subnetwork is designed as a convolutional neural network and the second subnetwork as a fully connected network. The convolutional neural network, also abbreviated to CNN in the field, can be a folding neural network. Such a network architecture can be particularly suitable for pattern recognition. The structure of a convolutional neural network can, for example, consist of one or more convolutional layers followed by a pooling layer. A fully connected network can be particularly suitable for data interpretation and also for the evaluation of recognized patterns. The method can therefore be particularly efficient.

[0025] A second aspect relates to a method for training a neural network designed to determine a change in the torque demand of respective auxiliary consumers of a work machine. Through training, the neural network can be trained to be able to determine changes in the torque demand of respective auxiliary consumers as a function of the continuously recorded output speed, the continuously recorded engine speed, and the continuously recorded engine torque. The neural network can be designed as a network that is used in the method according to the first aspect. Respective advantages and further features can be derived from the description of the first aspect, wherein embodiments of the first aspect also form embodiments of the second aspect, and vice versa.

[0026] The training method comprises a step of providing continuously recorded output speeds, continuously recorded engine speeds, and continuously recorded engine torques as input data of a training data set. This input data of the training data set can be generated, for example, in driving tests or through simulations. The training method comprises a step of providing the change in torque demand at respective auxiliary consumers as output data of the training data set. This output data of the training data set can also be generated, for example, in driving tests or through simulations. For this purpose, a work machine can, for example, be equipped with additional sensors to generate the training data during driving tests, which can also take place during real use.Training data can be real vehicle data, for example, where a transmission input torque is measured to calculate the change in torque demand at the respective auxiliary consumers. Alternatively, these changes in torque demand at the respective auxiliary consumers can also be measured directly. Training data can, for example, be determined specifically for each work machine or generally for all vehicles in a series using a work machine. The respective output data and, alternatively or additionally, output data from the training data set can be filtered according to whether they exceed a significance threshold. For example, only data that exceeds the significance threshold can be used for training.

[0027] The training method includes a step of training the neural network with the training data set in order to learn a relationship between the continuously measured output speed, the continuously measured engine speed, and the continuously measured engine torque and the change in torque demand at the respective auxiliary consumers. This generates, for example, respective weightings in the neural network and, alternatively or additionally, links the respective nodes of the neural network to one another.

[0028] A third aspect relates to a computer program product for determining a change in the instantaneous demand of respective auxiliary consumers of a work machine. The computer program product can have the neural network used in the method according to the first aspect. The computer program product can have been generated using a training method according to the second aspect. The computer program product can be stored on a non-volatile data storage device, such as a hard disk or a CD. Respective advantages and further features can be found in the description of the first and second aspects, respectively, wherein embodiments of the first and second aspects also form embodiments of the third aspect, and vice versa.

[0029] A fourth aspect relates to a work machine with a drive motor and a transmission, as well as at least one auxiliary consumer. The transmission can be designed, for example, to transmit the engine torque to an output of the work machine. The gear ratio can be changed by a circuit. The work machine can be designed to generate training data, which are used in the training method according to the second aspect. The work machine can be designed to carry out the method according to the first aspect. The work machine can have a data memory on which the computer program product according to the third aspect is stored. Respective advantages and further features can be derived from the description of the first, second, and third aspects, wherein embodiments of the first, second, and third aspects also form embodiments of the fourth aspect, and vice versa.

[0030] The work machine has a detection device configured to continuously detect an output speed, an engine speed, and an engine torque. The detection device may have respective sensors for this purpose and may alternatively or additionally be configured to access data in a vehicle bus of the work machine.

[0031] The work machine has a computing device which is designed to determine a change in the torque requirement at respective auxiliary consumers by means of the neural network based on a relationship, learned by the neural network, between the continuously detected output speed, the continuously detected engine speed and the continuously detected engine torque and the change in the torque requirement at respective auxiliary consumers.

[0032] The computing device may have a non-volatile memory on which the neural network is stored and can be retrieved alternatively or additionally.

[0033] The work machine has a testing device configured to check whether the change in the instantaneous demand determined by the neural network exceeds a significance threshold. Data relating to the significance threshold can also be stored in the non-volatile data memory. The testing device has, for example, a microprocessor. The test can be performed, for example, by a tabular comparison.

[0034] The work machine has an output device configured to output the change in the torque demand of the respective auxiliary consumers determined by the neural network, provided the test has shown that the significance threshold has been exceeded. The output device can, for example, have an electronic interface by means of which the data can be transmitted to the vehicle bus of the work machine and, alternatively or additionally, to a control unit, such as a transmission control for the drive transmission.

[0035] In a further embodiment of the work machine, the work machine is provided with a transmission control system configured to control the drive transmission as a function of an engine torque and the change in the torque demand of the respective auxiliary consumers on the drive motor, as determined by the neural network. The transmission control system can, for example, comprise a microprocessor. The transmission control system can, for example, use a characteristic map to control the drive transmission. Short description of the characters Fig. 1 schematically illustrates a method for determining a change in the instantaneous demand of respective auxiliary consumers of a work machine by means of a neural network. Fig. 2 schematically illustrates the working machine which is used to carry out the method according to Fig. 1 is trained. Detailed description of embodiments

[0036] Fig. 1 schematically illustrates a method for determining a change in the instantaneous demand of respective auxiliary consumers 10 of a work machine using a neural network. The work machine is schematically shown in Fig. 2 and has a drive transmission 12, by means of which drive power can be transmitted from the drive motor 14 to a driven axle 16. In addition, the work machine has respective auxiliary consumers 10, which can also be supplied with drive power by the drive motor 14.

[0037] In a step 40, an output speed of the drive transmission 12, an engine speed of the drive motor 14, which corresponds to an input speed at the drive transmission 12, and an engine torque generated by the drive motor 14 are continuously recorded as input data at discrete intervals by means of a recording device 18 of the work machine. In a step 42, the continuously recorded input data are input into a neural network on a computing device 20 of the work machine. In a step 48, the neural network determines the change in torque demand at respective auxiliary consumers 10 as output data based on a relationship, learned by the neural network, between the continuously recorded output speed, the continuously recorded engine speed, and the continuously recorded engine torque and the change in torque demand at respective auxiliary consumers.This makes it possible to estimate the torque flowing to the auxiliary consumers 10. With this information, the transmission input torque at the drive transmission 12 can now be calculated much more accurately, thereby improving shift comfort, transmission temperature calculation, and shift timing. This determination is possible because engaging or disengaging the auxiliary consumers 10 does not necessarily change the driving speed, but it can result in a change in the engine output speed.

[0038] In a step 44, a test device 22 of the working machine checks whether the change in instantaneous demand determined by the neural network exceeds a significance threshold. In the example shown, a change in instantaneous demand at the respective auxiliary consumers must be determined for longer than 1 second and also be greater than a specified minimum level. This prevents a vibration that only lasts for a short time from being recognized as a change in instantaneous demand at the respective auxiliary consumers. Likewise, noise with only a low amplitude can be filtered out.

[0039] In a step 46, the change in torque demand for each auxiliary consumer determined by the neural network is output via an output device 24 of the work machine, provided the test has shown that the significance threshold has been exceeded. The determined change in torque demand for each auxiliary consumer is loaded into a memory block and offset against the previously determined change in torque demand. In this way, a current torque demand by all auxiliary consumers is determined. This current torque demand of the auxiliary consumers is offset against an engine torque to determine a corrected input torque for the drive transmission 12. This corrected input torque is then used for vehicle control, such as the improved drive transmission control described above. Reference symbol 10 secondary consumers 12 drive gears 14 Drive motor 16 driven axles 18 Detection device 20 Calculating device 22 Test device 24 Dispensing device 40 Step: Entering the input data 42 Step: Entering the input data Step 44: Checking a change in current requirements 46 Step: Outputting a current requirement change 48 Step: Determine the instantaneous demand change

Claims

[1] Method for determining a change in the instantaneous demand of respective auxiliary consumers (10) of a work machine by means of a neural network, the method comprising at least the following steps: - Continuously detecting (40) an output speed, an engine speed and an engine torque as input data; - inputting (42) the continuously acquired input data into the neural network; - determining (48) the change in torque demand at respective auxiliary consumers as output data by means of the neural network based on a relationship, learned by the neural network, between the continuously detected output speed, the continuously detected engine speed and the continuously detected engine torque and the change in torque demand at respective auxiliary consumers (10); - checking (44) whether the instantaneous requirement change determined by the neural network exceeds a significance threshold; and - Outputting (46) the instantaneous demand change of respective auxiliary consumers (10) determined by the neural network, provided that the test has shown that the significance threshold has been exceeded. [2] Method according to claim 1, characterized by that the significance threshold has a minimum level of the instantaneous requirement change determined by the neural network. [3] Method according to claim 1 or 2, characterized by that the significance threshold has a minimum duration of the instantaneous demand change determined by the neural network. [4] Method according to one of the preceding claims, characterized bythat the method further comprises a step of determining a current instantaneous demand of respective secondary consumers (10) by offsetting the output instantaneous demand change of respective secondary consumers (10) against a previous instantaneous demand of respective secondary consumers (10). [5] Method according to one of the preceding claims, characterized by that the neural network has a first subnetwork and a second subnetwork, wherein the second subnetwork is connected in series to the first subnetwork behind the first subnetwork. [6] Method according to claim 5, characterized by that the first subnetwork is designed as a convolutional neural network and the second subnetwork as a fully connected network. [7] A work machine with a drive motor (14) and a transmission (12) and at least one auxiliary consumer (10), wherein the work machine further comprises a detection device (18) which is designed to continuously detect (40) an output speed, an engine speed and an engine torque, a computing device (20) which is designed to determine (48) a change in the torque requirement at respective auxiliary consumers (10) by means of the neural network based on a relationship, learned by the neural network, between the continuously detected output speed, the continuously detected engine speed and the continuously detected engine torque and the change in the torque requirement at respective auxiliary consumers (10), a testing device (22) which is designed to check (44) whether the change in the torque requirement determined by the neural network exceeds a significance threshold, and an output device,which is designed to output (46) the instantaneous demand change of respective auxiliary consumers (10) determined by the neural network, provided that the test has shown that the significance threshold has been exceeded., [8] Working machine according to claim 7, characterized by that the work machine has a transmission control which is designed to control the drive transmission (12) as a function of an engine torque and the change in the torque requirement of respective auxiliary consumers (10) to the drive motor (14) determined by the neural network.

Images