Method for estimating an excess oil volume flow in a working machine, method for training a machine learning model, computer program, device and working machine

Abstract

The present invention relates to a method for estimating an excess oil volume flow 119 in a work machine 100, a method for training a machine learning model 110, a computer program, a device 109, and a work machine 100. The method for estimating an excess oil volume flow 119 in a work machine 100 comprises the steps of obtaining S1 an oil pressure value 112, obtaining S2 an oil temperature value 114, and inputting S3 the oil pressure value 112 and the oil temperature value 114 as input signals into a trained machine learning model 110, wherein the machine learning model 110 was trained based on training data, each of which has an oil pressure value 112 and an oil temperature value 114 as input data and a value of the excess oil volume flow 119 of the work machine 100 as output data.The method further comprises determining an estimate of the excess oil volume flow 119 as an output of the trained machine learning model 110 based on the output data of the trained machine learning model 110.

Description

Technical field

[0001] The present invention relates to a method for estimating an excess oil volume flow in a work machine, a method for training a machine learning model, a computer program, a device and a work machine. State of the art

[0002] It is known from the prior art to use an oversized fixed-displacement pump to supply hydraulic consumers in work machines. The design point for dimensioning the fixed-displacement pump is a worst-case scenario at low speeds.

[0003] DE 10 2021 211 967 A1 discloses a hydraulic machine with an adjustable displacement volume for the pressure-controlled pressure medium supply to at least one hydrostatic consumer.

[0004] DE 10 2015 225 649 A1 discloses a method for controlling and / or regulating a hydraulic system for a transmission. Hydraulic oil is pumped by a pump, and a hydraulic network model is used to control and / or regulate the pump.

[0005] DE 10 2013 2017 708 A1 discloses a method for operating a drive system comprising a hydrostatic drive unit with at least one hydraulic pump and a hydraulic motor connected thereto via two working outputs. Description of the invention

[0006] A first aspect relates to a method for estimating an excess oil volume flow in a working machine.

[0007] The work machine can be a machine with a common oil supply. A common oil supply can mean that the work machine can include hydraulic systems, transmissions, and engines, and these use a common oil tank. A work machine can be a construction machine, such as a hydraulic excavator or a dump truck, an agricultural machine, such as a tractor or combine harvester, or a commercial vehicle, such as a bus.

[0008] The excess oil volume flow can be an oil volume flow that is pumped by a charging pump per unit of time in excess of the required amount.

[0009] The excess oil flow may be an excess of the oil flow at a lubrication pressure valve.

[0010] Estimating can include determining or calculating.

[0011] The method may be a computer-implemented method. The steps of the method may be executed by a computer using a corresponding computer program product, wherein the computer may be a control unit or part of a control unit.

[0012] The method includes obtaining an oil pressure value and obtaining an oil temperature value.

[0013] Obtaining a value can involve obtaining a raw sensor signal. Obtaining can involve obtaining a processed sensor value. Obtaining can also involve reading the value, for example, from a database. Obtaining the value can be done via a CAN, for example, a CAN bus.

[0014] The oil pressure value may include a lubricant pressure value. The oil pressure value may correlate with the oil volume flow and thus with an excess oil volume flow.

[0015] The oil temperature value can include a value of the lubricant temperature. The oil temperature value can influence the viscosity of the oil and thus correlate with the excess oil volume flow.

[0016] The method comprises inputting the oil pressure value and the oil temperature value as input signals into a trained machine learning model, wherein the machine learning model was trained based on training data each having an oil pressure value and an oil temperature value as input data and an excess oil volume flow of the working machine as output data.

[0017] The training data can be generated, for example, as part of a test setup. For this purpose, a volumetric flow meter can be integrated at a location in the working machine. The volumetric flow meter can be arranged in a gearbox of the working machine. The volumetric flow meter can be arranged in a line of the oil pressure valve. The oil pressure valve can be a passive pressure relief valve. Furthermore, an oversized fixed-displacement pump can be used to acquire the training data of the machine learning model. The volumetric flow meter can measure an excess oil volume flow, which can form the output data of the training data. The excess oil volume flow can be measured in the working machine, for example, in the gearbox. The input signals can be the measured oil pressure value and the measured oil temperature value.This allows the machine learning model to be trained based on the measured values. The training data can also be provided via a simulation.

[0018] The training data can include measurements across a variety of or all relevant states of the working machine. The training data can thus represent a representative mapping of the relevant states of the working machine and load spectra. Mapping a variety of or all relevant states of the working machine can, for example, enable a more accurate estimation of the excess oil volume flow.

[0019] The training data can be prepared before being entered.

[0020] Furthermore, the input signals can also include data that exhibits a time delay compared to the other input signals. This delayed data can, for example, represent an output signal of a PT1 element. For example, the PT1 element can represent a flow control. This allows a delay behavior to be represented even if the control unit capable of executing the method has low computing power. Thus, a time behavior can be represented.

[0021] The method includes determining an estimate of the excess oil volume flow as an output of the trained machine learning model based on the output data of the trained machine learning model.

[0022] The excess oil flow estimate may be the output of the trained machine learning model. The excess oil flow estimate may include post-processing the output of the trained machine learning model. The post-processing may, for example, include utilizing the determined excess oil flow estimate as a control input.

[0023] The proposed method can enable a conceptual and cost-effective implementation of estimating the excess oil volume flow in the working machine.

[0024] In one embodiment, the method may include controlling a pump of the work machine based on the determined estimate of the excess oil volume flow.

[0025] The pump of the driven machine can be a load pump. The pump of the driven machine can be part of a load-sensing system. The process can thus be integrated into a control strategy.

[0026] Control can be a dynamic process involving continuous measurement and adjustment. This can, for example, involve comparing an actual value with a target value. Control can also be the setting of a specific variable that is adjusted and occurs independently of continuous measurements or adjustments.

[0027] This allows for optimal pump control. The pump can adjust the oil flow rate so that excess oil flow is minimized or at least reduced. The energy supply to the pump can also be adjusted accordingly, so that reactive power can be minimized or at least reduced. This allows the pump's power consumption to be optimized.

[0028] In a further embodiment, the method may include mapping a virtual sensor based on the determined estimate of the excess oil volume flow.

[0029] A virtual sensor can provide calculated output values. The virtual sensor can be communicatively connected to other sensors or a unit for processing sensor data.

[0030] Mapping the virtual sensor may include outputting the determined estimate of the excess oil volume flow as an output of the virtual sensor. Thus, the determined estimate of the excess oil volume flow may be transmitted to another sensor or a unit for processing sensor data. Mapping the virtual sensor may include outputting the determined estimate of the excess oil volume flow on a user interface.

[0031] This allows the determined estimate of the excess oil volume flow to be mapped as a sensor value. This simplifies further processing or use.

[0032] In a further embodiment, the method may further comprise a step of receiving a transmission variable and, in the inputting step, inputting the received transmission variable as an input signal into the trained machine learning model, wherein the training data also includes the transmission variable as input data. In a further embodiment, the transmission variable may be a transmission input speed.

[0033] The gear size may include one or more values. The gear size may be time-varying. The gear size may be specific to the transmission or the work machine. The gear size may include a transmission input speed. The gear size may include a transmission output speed. The gear size may include one or more of a speed within the transmission, a torque, a number of teeth, a module, a gear material, an axle arrangement, a gear type, and a lubricant type.

[0034] This allows one or more additional parameters to be considered when estimating the excess oil flow. Transmission parameters, in particular, can be relevant for estimating the excess oil flow. The transmission input speed, in particular, can be relevant due to its influence on the heating of the gears and bearings. This enables a more accurate estimation of the excess oil flow using the trained machine learning model.

[0035] In another embodiment, the machine learning model can be a neural network. However, it is also possible for the machine learning model to be a support vector regression (SVR) model, a Gaussian process regression (GPR) model, or a convolutional neural network (CNN).

[0036] The neural network can be a regression model. The neural network can be a fully connected neural network (FCNN). This keeps the required computing power low. Thus, the method can be executed on common control units of work machines.

[0037] Using a neural network can provide a continuous value mapping. The continuous value mapping can be based on non-discrete classes. The neural network can also have a classification with discrete classes as output.

[0038] In another embodiment, the machine learning model may be a decision tree model.

[0039] The decision tree model can have a stepwise output. The stepwise output can be defined using thresholds. A threshold of the decision tree model can be in increments of 10 liters per minute for the determined estimate of the excess oil flow rate. The total oil volume in the working machine can be approximately 450 liters.

[0040] A second aspect relates to a method for training a machine learning model, for use in a method according to an embodiment of the first aspect. Training the machine learning model comprises the steps of inputting training data into the machine learning model, wherein the training data each comprise an oil pressure value and an oil temperature value as input data and a value of an excess oil volume flow of the work machine as output data.

[0041] This training data can, for example, result from an experimental setup described above. The aforementioned additional input of a transmission variable, such as the transmission input speed, can also be possible in the training process. The method for training the machine learning model can include a supervised learning method.

[0042] A third aspect relates to a computer program comprising instructions which, when executed by a computer, cause the computer to carry out the steps of the method for estimating an excess oil volume flow in a working machine.

[0043] A fourth aspect relates to a device configured to perform steps of a method according to an embodiment of the first aspect or the second aspect.

[0044] A fifth aspect relates to a work machine with a device for carrying out a method according to an embodiment of the first aspect, wherein the work machine comprises a pump, an oil tank and a hydraulic consumer, wherein the work machine and a device are configured to have a bidirectional connection to one another.

[0045] The hydraulic consumer can be one or more of a gearbox, a motor, a braking system, a cylinder, a pump, a steering system, a valve or similar.

[0046] Advantages and features described with respect to the method also apply to the device, the work machine, and the computer program, and vice versa. These are therefore described only once. Short description of the characters Fig. 1 shows schematically a work machine with a device according to an embodiment. Fig. Figure 2 schematically shows a trained machine learning model in a usage state. Fig. 3 shows a flowchart of a method for estimating an excess oil volume flow in a working machine. Detailed description of embodiments

[0047] Fig. 1 schematically shows a work machine 100 with a device 109 according to an embodiment.

[0048] The work machine 100 comprises an oil pressure sensor 102, an oil temperature sensor 104, an oil pressure valve 105, a pump 106, an oil tank 108, and a hydraulic consumer 107 in the form of a transmission 107. The work machine 100 further comprises a device 109 with a trained machine learning model 110. The device 109 is designed as a control unit. The work machine 100 and the device 109 are configured to have a bidirectional connection to one another. The device 109 is communicatively connected to the oil pressure sensor 102, the oil temperature sensor 104, the oil pressure valve 105, the transmission 107, and the pump 106. Furthermore, the pump 106, the oil tank 108, the oil pressure valve 105, and the transmission 107 are in fluid communication. During operation, the pump 106 delivers a volume flow of oil from the oil tank 108 to supply the gearbox 107 with lubricant.

[0049] The machine learning model 110 is a Fully Connected Neural Network (FCNN).

[0050] The oil pressure sensor 102 detects an oil pressure value 112, and the oil temperature sensor 104 detects an oil temperature value 114. These values are obtained by the device 109 and input to the trained machine learning model 110. Based on the output data of the trained machine learning model 110, an estimate of the excess oil volume flow 119 is determined.

[0051] The determined estimated value of the excess oil volume flow 119 is then used to control the pump 106 of the work machine 100 in order to adjust the oil volume flow delivered by the pump 106 accordingly. Furthermore, the determined estimated value of the excess oil volume flow 119 is used to map a virtual sensor (not shown).

[0052] In Fig. 2 shows a device 109 in the form of a control unit with a trained machine learning model 110.

[0053] The trained machine learning model 110 receives an oil pressure value 112, an oil temperature value 114, delayed data 116, and a transmission variable 117 in the form of a transmission input speed 118. The delayed data 116 are output values of a PT1 element. The transmission input speed 118 is measured using a speed sensor at the input of a transmission. The machine learning model 110 was trained with corresponding input data. The training data includes a measured value of the excess oil volume flow 119 of the work machine 100 for training the output of the machine learning model 110. Accordingly, the trained machine learning model determines an estimated value of the excess oil volume flow 119 as output.

[0054] Fig.3 shows a flowchart of a method for estimating an excess oil volume flow 119 in a work machine 100.

[0055] The method comprises a step S1 of obtaining an oil pressure value 112. A further step S2 comprises obtaining an oil temperature value 114. These values are input to the trained machine learning model as input signals in a step S3. In a step S4, an estimated value of the excess oil volume flow 119 is determined based on the output data of the trained machine learning model.

[0056] In an additional step S5, a pump 106 of the work machine 100 is controlled based on the determined estimated value of the excess oil volume flow 119. In a further step S6, a virtual sensor is mapped based on the determined estimated value of the excess oil volume flow 119. Reference symbol 100 work machines 102 Oil pressure sensor 104 Oil temperature sensor 105 Oil pressure valve 106 Pump 107 hydraulic consumer, gearbox 108 Oil tank 109 Device 110 Machine Learning Model 112 oil pressure value 114 Oil temperature value 116 delayed data 117 Gearbox size 118 Gearbox input speed 119 Value of excess oil volume flow S1 Obtaining an oil pressure value S2 Obtaining an oil temperature value S3 Input into a trained machine learning model S4 Determine an estimate of the excess oil volume flow S5 Control of a pump of the working machine S6 Mapping a virtual sensor

Claims

[1] Method for estimating an excess oil volume flow (119) in a work machine (100), comprising the steps of: - Obtaining (S1) an oil pressure value (112); - Obtaining (S2) an oil temperature value (114); - inputting (S3) the oil pressure value (112) and the oil temperature value (114) as input signals into a trained machine learning model (110), wherein the machine learning model (110) was trained based on training data which each have an oil pressure value (112) and an oil temperature value (114) as input data and a value of the excess oil volume flow (119) of the work machine (100) as output data; - Determining (S4) an estimated value of the excess oil volume flow (119) as an output of the trained machine learning model (110) based on the output data of the trained machine learning model (110). [2] Method according to claim 1, characterized bythat the method comprises a step of regulating (S5) a pump (106) of the working machine (100) based on the determined estimated value of the excess oil volume flow (119). [3] Method according to one of the preceding claims, characterized by that the method comprises a step of mapping (S5) a virtual sensor based on the determined estimated value of the excess oil volume flow (119). [4] Method according to one of the preceding claims, characterized by that the method further comprises a step of receiving a gear variable (117) and in the step of inputting the received gear variable (117) is input as an input signal into the trained machine learning model (110), wherein the training data also comprise the gear variable (117) as input data. [5] Method according to claim 3, characterized by that the gear size (117) is a gear input speed (118). [6] Method according to one of the preceding claims, characterized by that the machine learning model (110) is a neural network, or a support vector regression (SVR) model, or a Gaussian process regression (GPR) model, or a convolutional neural network (CNN). [7] Method according to one of claims 1 to 5, characterized by that the machine learning model (110) is a decision tree model. [8] A method for training a machine learning model (110) for use in a method according to any one of the preceding claims, wherein the training comprises the steps of: Inputting training data into the machine learning model, wherein the training data each have an oil pressure value (112) and an oil temperature value (114) as input data and a value of an excess oil volume flow (119) of the working machine (100) as output data. [9] A computer program comprising instructions which, when executed by a computer, cause the computer to carry out the steps of the method according to any one of claims 1 to 7. [10] Apparatus (109) adapted to carry out steps of the method according to any one of claims 1 to 8. [11] Work machine (100) with a device (109) for carrying out a method according to 1 to 7, wherein the work machine (100) comprises a pump (106), an oil tank (108) and a hydraulic consumer (107), wherein the work machine (100) and a device (109) are designed to have a bidirectional connection to one another.

Images