Efficient prediction method and device for friction torque of hydraulic motor driven by physical data fusion

Abstract

The present invention discloses a method and device for efficiently predicting the friction torque of a hydraulic motor driven by physical data fusion, which are applicable to the description of the friction behavior of the hydraulic motor and the calculation of the dynamic friction torque. Its essence is to deeply integrate the decomposition of the improved LuGre friction model with a neural network to obtain an improved LuGre-guided neural network (MLuGre-GNN). The friction torque prediction method driven by physical data fusion takes into account the advantages of both the physical friction model (i.e., the improved LuGre friction model) and the data-driven model (i.e., the neural network model), and has good physical interpretability and the strong local feature reconstruction ability of the neural network. In addition, the method can be approximately extended to all scenarios applicable to the LuGre model, providing an accurate and convenient new idea for friction prediction in mechanical systems.

Description

Technical Field

[0001] The present invention belongs to the field of non-linear dynamics prediction and analysis, and particularly relates to a method and device for efficiently predicting the frictional torque driven by physical data fusion applicable to hydraulic motors. Background Art

[0002] Hydraulic motors are widely used in major equipment such as shield machines and dredgers due to their advantages of high power density and high torque density. The strong non-linearity of friction directly affects the speed stability and control accuracy of hydraulic motors, and thus affects the operation safety of the equipment. Accurate friction prediction is the basis for studying the friction characteristics of hydraulic motors and is also the key to maintaining the safe and stable operation of major equipment. However, under low-speed and heavy-load conditions, the problems of numerous internal friction pairs and complex interfacial lubrication behaviors in hydraulic motors pose great challenges to establishing an accurate friction model.

[0003] The original LuGre friction model is only applicable to dry friction and non-radial force scenarios and is not applicable to the heavy-load lubrication conditions of hydraulic motors. The influence of radial pressure and lubricating oil makes the frictional torque more complex. At the same time, the LuGre friction model is a non-linear equation set with many parameters, and its parameter identification is extremely difficult. Especially for dynamic parameters, traditional methods require adjusting the controller parameters to make the system generate multiple sets of limit cycle oscillation curves in the pre-sliding stage. The experimental conditions are harsh and the experimental process is cumbersome, which is difficult to popularize in engineering practice.

[0004] In summary, the present invention proposes a method for predicting the frictional torque of a hydraulic motor driven by physical data fusion, which combines the advantages of physical prediction and data-driven prediction and can better calculate the lumped frictional torque of the hydraulic motor and describe the friction behavior. Summary of the Invention

[0005] In view of the challenges of accurately predicting the frictional torque of current hydraulic motors, the present invention proposes a method and device for efficiently predicting the frictional torque driven by physical data fusion, providing an accurate and convenient new idea for friction prediction in mechanical systems.

[0006] The object of the present invention is achieved by the following technical solutions: A method for efficiently predicting the frictional torque of a hydraulic motor driven by physical data fusion, which is composed of a physically-guided part and a data-driven part;

[0007] The process of the physical guidance part is as follows: First, based on the operating principle of the hydraulic motor, the influence of the working pressure is introduced on the basis of the original LuGre friction model to construct an improved LuGre friction model; then, the improved LuGre friction model is split into a steady-state part and a dynamic part, the steady-state part is used to construct the steady-state friction torque model of the hydraulic motor, and the dynamic part is replaced by a neural network model; finally, the steady-state characteristics obtained from the above-mentioned steady-state friction torque model of the hydraulic motor are used as physical friction characteristics and input into the neural network model to guide the training of the data-driven model;

[0008] The process of the data-driven part is as follows: First, according to the physical meaning of the improved LuGre friction model, a microscopic dynamic network and a dynamic friction network are constructed based on a multi-layer perceptron respectively; then, the physical friction characteristics are input into the microscopic dynamic network to obtain the microscopic dynamic characteristics at the corresponding time series, and then input into the dynamic friction network, and the predicted dynamic friction torque of the hydraulic motor is obtained by combining the speed damping characteristics; finally, the overall data-driven model is iteratively trained and parameter optimized to establish a dynamic friction torque model of the hydraulic motor, and torque prediction is realized based on the dynamic friction torque model of the hydraulic motor.

[0009] Furthermore, the specific process of the physical guidance part is as follows: First, an improved LuGre model considering the working pressure is proposed for the hydraulic motor, and the improved LuGre model is:

[0010]

[0011] where z represents the average bristle displacement; represents the angular velocity; represents the pressure difference between the high and low pressures of the hydraulic motor; represents the Stribeck model, represents the bristle stiffness; represents the bristle damping coefficient; represents the speed damping coefficient; represents the dynamic friction torque.

[0012] Furthermore, under steady-state conditions of the improved LuGre model, that is, when the friction torque is expressed as:

[0013]

[0014] where represents the Coulomb friction coefficient; represents the static friction coefficient; represents the no-load Coulomb friction torque; represents the no-load static friction torque; is the Stribeck angular velocity; for the parameter identification of the steady-state part, the least squares method is directly applied, and the optimal solution of the steady-state friction torque parameters can be obtained by minimizing the objective function; among them, is the i-th collected friction torque; is the i-th predicted steady-state friction torque; finally, the steady-state friction torque is decomposed into the Stribeck term and the velocity damping term , and they are respectively input into two subsequent constructed neural networks as prior physical knowledge to guide the training of the data-driven model.

[0015] Furthermore, the microscopic dynamic network is used for predicting the dynamic change of the bristles, describing the dynamic response of the bristle deformation. It inputs and integrates the angular velocity of the hydraulic motor, the working pressure state quantity of the hydraulic motor, and the static friction characteristic quantity in time series, and outputs the microscopic dynamic characteristic quantity corresponding to the time series; in order to control the output range and enhance the feature learning ability, the sigmoid function is applied as the activation function between layers.

[0016] Furthermore, the dynamic friction network is used to receive the output of the microscopic dynamic network and integrate the velocity damping characteristics. The output layer needs to combine the velocity damping characteristics corresponding to the input state quantity on the basis of the microscopic dynamic characteristic quantity; the output of the dynamic friction network is the predicted dynamic friction torque of the hydraulic motor.

[0017] Furthermore, the microscopic dynamic network consists of an input layer, two hidden layers and an output layer. The number of neurons in the two hidden layers is 4N p and 2N p respectively, and the dimension of the output layer is N p ; the dynamic friction network consists of an input layer, a hidden layer and an output layer. The number of neurons in the hidden layer is 2N p respectively, and the dimension of the output layer is N p .

[0018] Furthermore, the gradient descent method is applied to the overall framework MLuGre-GNN composed of the microscopic dynamic network and the dynamic friction network to find such that , where

[0019] In a second aspect, the present invention further provides a device for efficiently predicting the frictional torque of a hydraulic motor driven by physical data fusion, including a memory and one or more processors. Executable code is stored in the memory. When the processor executes the executable code, the method for efficiently predicting the frictional torque of a hydraulic motor driven by physical data fusion as described above is implemented.

[0020] In a third aspect, the present invention further provides a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, the method for efficiently predicting the frictional torque of a hydraulic motor driven by physical data fusion as described above is implemented.

[0021] In a fourth aspect, the present invention further provides a computer program product, including a computer program. When the computer program is executed by a processor, the method for efficiently predicting the frictional torque of a hydraulic motor driven by physical data fusion as described above is implemented.

[0022] Advantages of the present invention:

[0023] The method for efficiently predicting the frictional torque driven by physical data fusion provided by the present invention is used for predicting the dynamic frictional torque of a hydraulic motor, solving the problems that the original LuGre model does not consider the normal force and has difficulties in parameter identification, as well as the disadvantages of poor generalization and lack of physical meaning in data-driven prediction. Combining the advantages of physical prediction and data-driven prediction, it can better calculate the lumped dynamic frictional torque of the hydraulic motor and describe the frictional behavior, having unique advantages in terms of prediction convenience and accuracy. In addition, this method can be approximately corrected and extended to all scenarios applicable to the LuGre model to replace the traditional LuGre prediction method. Description of the drawings

[0024] Figure 1 It is a flowchart of the method for efficiently predicting the frictional torque of a hydraulic motor driven by physical data fusion proposed.

[0025] Figure 2 It is a flowchart of the data-driven part of the proposed method.

[0026] Figure 3 It is a structural diagram of the device for efficiently predicting the frictional torque of a hydraulic motor driven by physical data fusion according to the present invention. Detailed implementation manners

[0027] The present invention will be elaborated in detail below in conjunction with the drawings and specific examples:

[0028] As Figure 1 shown, the flowchart illustrates the process from signal acquisition of a certain type of hydraulic motor to dynamic frictional torque prediction:

[0029] The method proposed by the present invention mainly consists of a physical guidance part and a data-driven part. The process of the physical guidance part is as follows: First, the original LuGre friction model is improved to solve the problem of ignoring the normal pressure. Then, the improved LuGre friction model is split into a steady-state part and a dynamic part. The steady-state part is used to construct a physical model of the steady-state friction torque of the hydraulic motor, and the dynamic part is replaced by a data-driven model. Finally, the steady-state friction torque model of the hydraulic motor is input into the data-driven model as physical friction feature guidance for training the data-driven model. The process of the data-driven part is as follows: First, a microscopic dynamic network and a dynamic friction network are respectively constructed based on a multi-layer perceptron, which respectively correspond to the corresponding formulas in the improved LuGre friction model in terms of physical meaning. Then, physical friction features are input. Finally, the overall data-driven model is iteratively trained and parameter-optimized to establish a dynamic friction torque model of the hydraulic motor.

[0030] Signal acquisition is performed on the hydraulic motor, and state variables such as angular velocity and working pressure are used for physical prediction of the steady-state part of the improved LuGre model, and the dynamic friction torque is used for data-driven prediction.

[0031] The improved LuGre model is as follows:

[0032]

[0033]

[0034]

[0035]

[0036]

[0037] where z represents the average bristle displacement; represents the angular velocity; represents the pressure difference between the high and low pressures of the hydraulic motor; represents the Coulomb friction coefficient; represents the static friction coefficient; represents the no-load Coulomb friction torque; represents the no-load static friction torque; is the Stribeck angular velocity; represents the bristle stiffness; represents the bristle damping coefficient; represents the velocity damping coefficient; represents the dynamic friction torque.

[0038] The improved LuGre model under steady-state conditions (i.e., ) the friction torque is expressed as:

[0039]

[0040] For the parameter identification process of physical prediction, the least squares method is directly applied, and the optimal solution of the steady-state friction torque parameters is obtained by minimizing the objective function. The objective function to be optimized is expressed as:

[0041]

[0042] where is the i-th collected friction torque; is the i-th predicted steady-state friction torque; The steady-state friction torque is decomposed into the Stribeck term and the velocity damping term , which are respectively input into two subsequent constructed neural networks as prior physical knowledge to guide the training of the data-driven model.

[0043] The neural network parameter optimization process for data-driven prediction corresponds to Figure 1 the backpropagation part in , that is, the gradient descent method is applied to find such that , and the neural network training parameters and are optimized by minimizing the error between the output

[0044]

[0045] where N o is the model output dimension, . Define as the output of the j-th layer of the i-th network (i = 1, 2; j = 1, 2, 3). The gradient in the backpropagation of the microscopic dynamic network is expressed as:

[0046]

[0047] where ; represents the derivative of the sigmoid function. Since the output of the microscopic dynamic network combines the velocity damping characteristics and is directly input into the microscopic dynamic network, the gradient of the output layer of the microscopic dynamic network is provided reversely by the gradient of the input layer of the microscopic dynamic network in backpropagation, that is:

[0048]

[0049] The gradient in the backpropagation of the microscopic dynamic network is expressed as:

[0050]

[0051] For the MLuGre-GNN, update the training parameters after one round of gradient backpropagation :

[0052]

[0053] where t represents the number of training iterations, represents the gradient of the loss with respect to the current network parameters. After iterative training and loss reduction, the network prediction parameters of the MLuGre-GNN gradually approach the optimal, and the dynamic friction torque prediction of the hydraulic motor is completed.

[0054] As Figure 2 shown, the flowchart illustrates the feature transfer process between the data-driven parts of the proposed method, i.e., the micro-dynamic network and the dynamic friction network:

[0055] The present invention constructs a micro-dynamic network (Micro-dynamic net) and a dynamic friction network (Dynamic-friction net) based on a multi-layer perceptron respectively. The micro-dynamic network is used for predicting the dynamic bristle changes, describing the dynamic response of bristle deformation, inputting and integrating the state variables such as rotational speed and pressure in time series and the static friction feature variables, and outputting the micro-dynamic feature variables corresponding to the time series. Its input variables are:

[0056]

[0057] where N p represents the time step of the feature, which directly affects the number of features input for a single state variable; w is the angular velocity of the hydraulic motor actually collected; represents the working pressure of the hydraulic motor actually collected at the t-th moment, and g(w) is the Stribeck feature. The micro-dynamic network is composed of a 4-layer feedforward multi-layer perceptron, consisting of an input layer, two hidden layers and an output layer. Its output is expressed as:

[0058]

[0059] where, represents the functional structure of the dynamic friction network; represents the predicted output; represents the trainable parameters of the dynamic friction network, , represents the weight of the j-th layer of the micro-dynamic network; represents the bias of the j-th layer of the micro-dynamic network. The number of neurons in the two hidden layers of the micro-dynamic network is 4N p and 2N p , and the dimension size of the output layer is N p. To control the output range and enhance the feature learning ability, the sigmoid function is applied as the activation function between layers and is defined as: , then the output of the microscopic dynamic network is specifically:

[0060]

[0061] The dynamic friction network is also designed based on a multi-layer perceptron and is used to receive the output of the microscopic dynamic network and integrate the speed damping feature. It consists of an input layer, a hidden layer, and an output layer. The number of neurons in the hidden layer is 2N p , and the dimension of the output layer is N p . The output layer needs to combine the speed damping feature corresponding to the input state quantity on the basis of . Therefore, the input of the dynamic friction network is expressed as: :

[0062]

[0063] The output of the dynamic friction network is the predicted dynamic friction torque of the hydraulic motor and is specifically expressed as:

[0064]

[0065] Among them, represents the functional structure of the dynamic friction network; represents the predicted output of the dynamic friction network; represents the trainable parameters of the dynamic friction network, . represents the weight of the k-th layer of the dynamic friction network; represents the bias of the k-th layer of the dynamic friction network.

[0066] In summary, the embodiment of the present invention provides an efficient prediction method for the friction torque of a hydraulic motor driven by physical data fusion.

[0067] Corresponding to the foregoing embodiment of an efficient prediction method for the friction torque of a hydraulic motor driven by physical data fusion, the present invention also provides an embodiment of an efficient prediction device for the friction torque of a hydraulic motor driven by physical data fusion.

[0068] Referring to Figure 3 , an efficient prediction device for the friction torque of a hydraulic motor driven by physical data fusion provided by the embodiment of the present invention includes a memory and one or more processors. Executable code is stored in the memory. When the processor executes the executable code, it is used to implement the efficient prediction method for the friction torque of a hydraulic motor driven by physical data fusion in the foregoing embodiment.

[0069] An embodiment of the efficient friction torque prediction device for physical data fusion driving applicable to a hydraulic motor provided by the present invention can be applied to any device with data processing capabilities. Such a device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented by software, or by hardware, or by a combination of software and hardware. Taking software implementation as an example, as a logically defined device, it is formed by the processor of any device with data processing capabilities reading the corresponding computer program instructions in the non-volatile memory into the memory for operation. At the hardware level, as Figure 3 shown, it is a hardware structure diagram of any device with data processing capabilities where the efficient friction torque prediction device for physical data fusion driving applicable to a hydraulic motor provided by the present invention is located. In addition to Figure 3 the processor, memory, network interface, and non-volatile memory shown, for any device with data processing capabilities where the device in the embodiment is located, usually according to the actual functions of the device with data processing capabilities, other hardware may also be included, which will not be elaborated here.

[0070] The implementation processes of the functions and roles of each unit in the above device are specifically detailed in the implementation processes of the corresponding steps in the above method, which will not be elaborated here.

[0071] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present invention solution. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0072] The embodiment of the present invention also provides a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, it implements an efficient friction torque prediction method for physical data fusion driving applicable to a hydraulic motor in the above embodiment.

[0073] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium may also be an external storage device of any device with data processing capabilities, such as a plug-in hard disk, a Smart Media Card (SMC), an SD card, a Flash Card, etc. equipped on the device. Further, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and may also be used to temporarily store data that has been output or is to be output.

[0074] The present invention also provides a computer program product, including a computer program, which when executed by a processor, implements the physical data fusion-driven friction torque efficient prediction method applicable to a hydraulic motor described above.

[0075] The above are only the embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structural or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. An efficient prediction method for the friction torque of a hydraulic motor driven by physical data fusion, characterized in that This method consists of a physically-guided part and a data-driven part; The process of the physically-guided part is as follows: First, based on the operating principle of the hydraulic motor, the influence of the working pressure is introduced on the basis of the original LuGre friction model to construct an improved LuGre friction model; Then, the improved LuGre friction model is split into a steady-state part and a dynamic part. The steady-state part is used to construct the steady-state friction torque model of the hydraulic motor, and the dynamic part is replaced by a neural network model; Finally, the steady-state characteristics obtained from the above-mentioned steady-state friction torque model of the hydraulic motor are used as physical friction characteristics and input into the neural network model to guide the training of the data-driven model; The process of the data-driven part is as follows: First, according to the physical meaning of the improved LuGre friction model, a microscopic dynamic network and a dynamic friction network are respectively constructed based on a multi-layer perceptron; Then, the physical friction characteristics are input into the microscopic dynamic network to obtain the microscopic dynamic characteristics at the corresponding time series, and then input into the dynamic friction network, and combined with the speed damping characteristics to obtain the predicted dynamic friction torque of the hydraulic motor; Finally, the overall data-driven model is iteratively trained and parameter-optimized to establish a dynamic friction torque model of the hydraulic motor, and torque prediction is realized based on the dynamic friction torque model of the hydraulic motor.

2. The efficient prediction method for the friction torque of a hydraulic motor driven by physical data fusion according to claim 1, wherein The specific process of the physically-guided part is as follows: First, an improved LuGre model considering the working pressure is proposed for the hydraulic motor. The improved LuGre model is: ； where z represents the average bristle displacement; represents the angular velocity; represents the pressure difference between the high and low pressures of the hydraulic motor; represents the Stribeck model, represents the bristle stiffness; represents the bristle damping coefficient; represents the velocity damping coefficient; represents the dynamic friction torque.

3. A method for efficiently predicting the friction torque of a hydraulic motor driven by physical data fusion according to claim 2, characterized in that The improved LuGre model under steady-state conditions, i.e., when the friction torque is expressed as: ； Among them, represents the Coulomb friction coefficient; represents the static friction coefficient; represents the no-load Coulomb friction torque; represents the no-load static friction torque; is the Stribeck angular velocity; for the parameter identification of the steady-state part, the least squares method is directly applied, and the optimal solution of the steady-state friction torque parameters can be obtained by minimizing the objective function; among them, is the i-th collected friction torque; is the i-th predicted steady-state friction torque; finally, the steady-state friction torque is decomposed into the Stribeck term and the velocity damping term , which are respectively input into two neural networks constructed subsequently as prior physical knowledge to guide the training of the data-driven model.

4. A method for efficiently predicting the friction torque of a hydraulic motor driven by physical data fusion according to claim 1, characterized in that, The microscopic dynamic network is used for predicting the dynamic change of the bristles, describing the dynamic response of the bristle deformation. It inputs and integrates the angular velocity of the hydraulic motor, the working pressure state quantity of the hydraulic motor and the static friction characteristic quantity in time series, and outputs the microscopic dynamic characteristic quantity at the corresponding time series; In order to control the output range and enhance the feature learning ability, the sigmoid function is used as the activation function between layers.

5. The high-efficiency prediction method for the friction torque of a hydraulic motor driven by physical data fusion according to claim 1, wherein The dynamic friction network is used to receive the output of the microscopic dynamic network and integrate the speed damping characteristics. The output layer needs to combine the speed damping characteristics corresponding to the input state quantity on the basis of the microscopic dynamic characteristic quantity; The output of the dynamic friction network is the predicted dynamic friction torque of the hydraulic motor.

6. The high-efficiency prediction method for the friction torque of a hydraulic motor driven by physical data fusion according to claim 1, wherein, The microscopic dynamic network consists of an input layer, two hidden layers, and an output layer. The number of neurons in the two hidden layers is 4N p and 2N p , and the dimension of the output layer is N p ; The dynamic friction network consists of an input layer, a hidden layer, and an output layer. The number of neurons in the hidden layer is 2N p , and the dimension of the output layer is N p .

7. A method for efficiently predicting the friction torque of a hydraulic motor driven by physical data fusion according to claim 2, characterized in that Apply the gradient descent method to the overall framework MLuGre-GNN composed of the microscopic dynamic network and the dynamic friction network to find such that , denotes the predicted output of the dynamic friction network, and optimize the training parameters of the microscopic dynamic network and the dynamic friction network by minimizing the error between the output of MLuGre-GNN and the true measured friction torque.

8. An efficient prediction device for the friction torque of a hydraulic motor driven by physical data fusion, comprising a memory and one or more processors, wherein executable code is stored in the memory, characterized in that, When the processor executes the executable code, it implements a method for efficiently predicting the friction torque of a hydraulic motor driven by physical data fusion as described in any one of claims 1-7.

9. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements a method for efficiently predicting the friction torque of a hydraulic motor driven by physical data fusion as described in any one of claims 1-7.

10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements a method for efficiently predicting the friction torque of a hydraulic motor driven by physical data fusion as described in any one of claims 1-7.

Images