Plunger pump oil film field calculation method and system based on DeepONet

Abstract

The invention relates to the field of fluid pressure and the field of electric digital data processing, in particular to a DeepONet-based plunger pump oil film field calculation method and system, and the method comprises the steps: obtaining a training data set; an oil film thickness field-oil film pressure field calculation model based on DeepONet is constructed; training the oil film thickness field-oil film pressure field calculation model by adopting the training data set; and calculating a pressure value at a randomly generated space coordinate based on the trained oil film thickness field-oil film pressure field calculation model. According to the embodiment of the invention, by constructing the oil film thickness field-pressure field DeepONet, the calculation resolution and calculation precision of the oil film field are improved while the calculation speed is considered.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of fluid pressure and the field of digital data processing, in particular to a DeepONet-based plunger pump oil film field calculation method and system. BACKGROUND

[0002] The key friction pairs (porting pairs, plunger pairs and sliding shoe pairs) of the axial plunger pump work in high-pressure fluid medium, and the oil film characteristics of these friction pairs have a decisive influence on the energy efficiency and service life of the plunger pump. The oil film formed between these friction pair assemblies simultaneously undertakes the functions of bearing, sealing and lubrication, and the coupling of the oil film thickness field and the pressure field directly determines the bearing performance, leakage characteristics and lubrication state of the friction pair. Therefore, it is crucial to realize high-precision and high-resolution oil film pressure field calculation for in-depth analysis of the oil film characteristics of the friction pair, guidance of the optimization design of the plunger pump and reliability evaluation.

[0003] The traditional oil film pressure field calculation method is based on the Reynolds equation to establish a numerical model, and uses numerical calculation methods such as difference method or finite element method. However, the oil film field of the axial plunger pump involves strong coupling of fluid-solid-thermal multi-physical fields, making the traditional oil film field calculation method face the dual challenges of calculation speed and calculation accuracy. This method needs to discretize the known oil film thickness field into grids, and solve the grid node pressure value through a complex iterative process, resulting in low solving efficiency. Although the existing improved method (such as the fluid film pressure calculation method based on the Garlerkin idea, CN109815548A) simplifies the process by optimizing the basis function and reducing the order of calculation, its accuracy and resolution are still limited by the grid density. To obtain a higher-precision pressure field, the number of grids must be increased, which will lead to an exponential increase in the calculation scale.

[0004] In recent years, machine learning technology has provided a new way for oil film pressure field calculation. Some research (Development of a Machine Learning Model for Elastohydrodynamic Pressure Prediction in Journal Bearings) has constructed and trained a machine learning model to approximate the function mapping from the oil film thickness value to the oil film pressure value, significantly improving the calculation speed, but still has the following limitations: (1) The oil film field resolution is related to the output layer neurons of the machine learning model, and a large number of output layer neurons are needed to realize the calculation of high-resolution oil film field; (2) The mapping relationship is learned through function value to function value fitting, and the generalization ability is limited, when the input oil film thickness value exceeds the range of the training data set, the corresponding oil film field calculation accuracy will decrease significantly; (3) The pure data-driven black box model is difficult to meet the constraints of the Reynolds equation, limiting the improvement of the oil film field calculation accuracy.

[0005] Overall, the existing method still has obvious deficiencies in computing efficiency, accuracy and generalization ability, and a new solution is needed to break through these technical bottlenecks. SUMMARY

[0006] One of the purposes of the present application is to provide a DeepONet-based plunger pump oil film field calculation method and system to solve the technical problems of low oil film pressure field calculation efficiency, low accuracy and poor generalization ability in the prior art.

[0007] In order to achieve the above-mentioned purpose, the present application provides a DeepONet-based plunger pump oil film field calculation method, comprising: obtaining a training data set; constructing a DeepONet-based oil film thickness field-oil film pressure field calculation model; training the oil film thickness field-oil film pressure field calculation model using the training data set, wherein the loss function of the oil film thickness field-oil film pressure field calculation model comprises formulas (1) to (4): , (1) , (2) , (3) , (4) wherein, is the loss function, is the Reynolds equation constraint term, is the data constraint term, is the boundary condition constraint term, is the total number of training data set samples, is the number of pressure field coordinate points in each sample, is the spatial coordinate in the polar coordinate system, is the oil film thickness, is the kinematic viscosity of hydraulic oil, is the predicted pressure value of the oil film thickness field-oil film pressure field calculation model at the coordinate of the sample, is the tangential velocity of the shoe bottom surface relative to the swash plate, is the radial flow velocity caused by the change of the oil film thickness, is the reference solution of the sample at the coordinate , is the boundary pressure value, is the pressure solution of the sample at the boundary position ; The pressure value at the randomly generated spatial coordinate is calculated based on the trained oil film thickness field-oil film pressure field calculation model.

[0008] Optionally, the acquiring the training data set comprises: establishing oil film thickness field space polar coordinates; constructing an oil film thickness field model; constructing a plunger pump slipper pair Reynolds equation and solving a pressure field reference solution; discretely sampling the oil film thickness field to obtain oil film thickness field sampling point data; discretely sampling the oil film pressure field to obtain oil film pressure field sampling point data; acquiring a training data set based on the oil film thickness field sampling point data and the oil film pressure field sampling point data.

[0009] Optionally, the establishing oil film thickness field space polar coordinates comprises: taking the projection of the slipper bottom surface center on the swash plate as the origin establishing axis and axis on the normal of the contact surface of the oil film and the swash plate axis.

[0010] Optionally, the constructing an oil film thickness field model comprises: establishing an oil film thickness model according to formula (5), , (5) wherein, is the oil film thickness field, is the basic oil film thickness, is the central oil film thickness, is the slipper overturning angle, is the slipper overturning azimuth angle, is the radial coordinate, is the angle coordinate, is the slipper oil chamber radius, is the slipper bottom surface outer circle radius.

[0011] Optionally, the constructing a plunger pump slipper pair Reynolds equation and solving a pressure field reference solution comprises: establishing a plunger pump slipper pair Reynolds equation according to formula (6), , (6) wherein, is the oil film pressure field; inputting a given oil film thickness field into the Reynolds equation to obtain corresponding pressure field data as the oil film thickness field-oil film pressure field reference solution.

[0012] Optionally, the oil film thickness field is discretely sampled to obtain oil film thickness field sampling point data, comprising: The oil film thickness field is discretely sampled according to formula (7), and corresponding , , (7) wherein, is a fixed sampling point, is the number of sampling points, is a radial coordinate, is an angular coordinate, is the radius of the oil chamber of the sliding shoe, is the outer circle radius of the sliding shoe bottom surface.

[0013] Optionally, the oil film pressure field is discretely sampled to obtain oil film pressure field sampling point data, comprising: The oil film pressure field is discretely sampled according to formula (8), and corresponding , , (8) wherein, is a random sampling point, is a uniform probability distribution, is a uniform probability distribution, is the radius of the oil chamber of the sliding shoe, is the outer circle radius of the sliding shoe bottom surface.

[0014] Optionally, the oil film thickness field-oil film pressure field calculation model based on DeepONet comprises: A branch network and a trunk network are constructed, wherein the branch network adopts a series structure of an attention mechanism and a multilayer perception machine, the attention mechanism is used to capture global features, and the multilayer perception machine is used to process local features, the branch network is used to input oil film thickness values and output a global feature vector; the trunk network is a multilayer perception machine structure, used to input spatial coordinates and operating conditions and calculate a position feature vector; The pressure value is calculated according to the global feature vector and the position feature vector.

[0015] Optionally, the oil film thickness field-oil film pressure field calculation model is trained using the training data set, comprising: The training data set is input into the oil film thickness field-oil film pressure field calculation model, and a predicted pressure value is output; The loss function value is calculated; It is judged whether the loss function value is greater than a threshold value; In the case where it is judged that the loss function value is greater than the threshold value, the parameters of the oil film thickness field-oil film pressure field calculation model are updated, and the pressure field is recalculated; In a case where it is determined that the loss function value is less than or equal to a threshold value, the training is completed.

[0016] In another aspect, the present application provides a DeepONet-based plunger pump oil film field calculation system, comprising a processor configured to perform the method of any one of the above.

[0017] Advantages of the present application: The present application avoids the process of repeatedly solving partial differential equations by constructing an oil film thickness field-pressure field DeepONet. After the DeepONet model is trained, only one forward propagation is needed to quickly calculate the pressure field. Compared with the prior art, the present application avoids complex meshing and iterative calculation, and the pressure field calculation is faster.

[0018] The present application directly learns the global functional mapping relationship from the oil film thickness field to the oil film pressure field by constructing an oil film thickness field-pressure field DeepONet. The branch network is a series structure of attention mechanism and MLP. The oil film thickness field is discretized and global features are extracted. The backbone network encodes the spatial coordinates and working conditions, and outputs the position features. The global features and position features are summed to calculate the pressure value at any position in the pressure field domain. The generalization ability is strong, and the solving accuracy is still high on data not involved in training. At the same time, the loss function in the present application includes a Reynolds equation constraint term, a data constraint term and a boundary condition constraint term, which fully considers the consistency of the oil film field physical information and the boundary conditions, and further improves the solving accuracy.

[0019] The oil film thickness field-pressure field DeepONet constructed in the present application decouples the oil film thickness value and the spatial coordinates through the branch network and the backbone network. Only the spatial coordinates input into the backbone network need to be changed to calculate the oil film pressure value at any position. Therefore, the resolution of the oil film pressure field is only related to the number of spatial coordinates input into the backbone network. By increasing the number of spatial coordinates, high-resolution oil film pressure field can be quickly calculated without retraining the network.

[0020] Other features and advantages of the embodiments of the present application will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0021] The accompanying drawings are included to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used together with the following specific embodiments to explain the embodiments of the present application, but do not constitute a limitation on the embodiments of the present application. In the drawings: Figure 1 Flowchart of the DeepONet-based plunger pump oil film field calculation method according to an embodiment of the present application; Figure 2A flowchart of a method for obtaining a training data set according to one embodiment of the present invention; Figure 3 A schematic diagram of establishing a spatial polar coordinate system according to one embodiment of the present invention; Figure 4 A flowchart of a method for constructing an oil film thickness field-oil film pressure field calculation model based on DeepONet according to one embodiment of the present invention; Figure 5 Flowchart of a method for training an oil film thickness field-oil film pressure field calculation model according to one embodiment of the present invention; Figure 6 A framework diagram of an oil film thickness field-oil film pressure field calculation model based on DeepONet according to one embodiment of the present invention; Figure 7 This is an example 1 of the oil film field calculated by the DeepONet-based oil film thickness field-oil film pressure field calculation model according to one embodiment of the present invention; Figure 8 This is a second example of the oil film field calculated by the DeepONet-based oil film thickness field-oil film pressure field calculation model according to one embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following describes the specific implementation of the embodiment of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific implementation described herein is only used to illustrate and explain the embodiment of the present invention and is not used to limit the embodiment of the present invention.

[0023] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application are in compliance with the relevant provisions of laws and regulations. In the embodiments of this application, certain software, components, models, and other existing solutions in the industry may be mentioned. These should be considered as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of this application, but it does not mean that the applicant has or will necessarily use such solutions.

[0024] like Figure 1 FIG. 1 is a flow chart of a method for calculating the oil film field of a plunger pump based on DeepONet according to an embodiment of the present invention. Figure 1 The calculation method may include the following steps: In step S10, a training data set is obtained; In step S11, a DeepONet-based oil film thickness field-oil film pressure field calculation model is constructed; In step S12, the oil film thickness field-oil film pressure field calculation model is trained using a training data set, wherein the loss function of the oil film thickness field-oil film pressure field calculation model includes formulas (1) to (4): , (1) , (2) , (3) , (4) in, is the loss function, is the constraint term of the Reynolds equation, is the data constraint item, is the boundary condition constraint, is the total number of samples in the training dataset, is the number of pressure field coordinate points in each sample, is the spatial coordinate in the polar coordinate system, is the oil film thickness, is the kinematic viscosity of the hydraulic oil, The calculation model of oil film thickness field-oil film pressure field is Samples at coordinates The predicted pressure value at is the tangential velocity of the bottom surface of the sliding shoe relative to the swash plate, is the radial flow velocity caused by the change of oil film thickness, For the Samples at coordinates The reference solution at is the boundary pressure value, For the Samples at the boundary The pressure solution at ; In step S13 , the pressure value at the randomly generated spatial coordinate is calculated based on the trained oil film thickness field-oil film pressure field calculation model.

[0025] In this Figure 1 In the illustrated DeepONet-based plunger pump oil film field calculation method, step S10 is used to obtain a training data set. In this embodiment, the specific method for obtaining the training data set in step S10 can be various forms known to those skilled in the art. In one example of the present invention, step S10 may include: Figure 2 The steps shown in Figure 2 In the step S10, the following steps may be performed: In step S20, the oil film thickness field spatial polar coordinates are established; In step S21, an oil film thickness field model is constructed; In step S22, a Reynolds equation of the plunger pump slipper pair is constructed, and a pressure field reference solution is solved; In step S23, the oil film thickness field is discretely sampled to obtain oil film thickness field sampling point data; In step S24, the oil film pressure field is discretely sampled to obtain oil film pressure field sampling point data; In step S25, based on the oil film thickness field sampling point data and the oil film pressure field sampling point data, a training data set is obtained.

[0026] In the method as shown in the embodiment, Figure 2 Step S20 is used to establish an oil film thickness field space polar coordinate. Since the oil film of the plunger pump friction pair is annular, it is not convenient for subsequent calculation to establish a space rectangular coordinate system, therefore, in the embodiment, a space polar coordinate system can be established. Specifically, as shown in the embodiment, Figure 3 The projection of the slipper bottom surface center on the swash plate is taken as the origin , an axis and axis are established on the contact surface of the oil film and the swash plate in the normal direction of the contact surface, the oil film thickness field is represented by , and the oil film pressure field is represented by .

[0027] The plunger pump structure parameters and the plunger pump operating conditions can be given according to actual requirements. Specifically, in the example, the plunger pump structure parameters can be given as follows: the slipper oil chamber radius , the slipper bottom surface outer circle radius , the plunger distribution circle radius , the hydraulic oil performance parameters including the hydraulic oil kinematic viscosity are given. The plunger pump operating conditions are given as follows: the main shaft rotating speed and the load pressure are given as the solving conditions, and the pump shell cavity pressure is given as the boundary condition. In the example, the main shaft rotating speed 1000 rpm, the load pressure 9 MPa, the pump shell cavity pressure 0.1 MPa, and the main shaft rotating speed 5000 rpm, the load pressure 21 MPa, and the pump shell cavity pressure 0.1 MPa can be set as the two groups of operating conditions.

[0028] Step S21 is used to construct an oil film thickness field model. In the embodiment, the specific method of constructing the oil film thickness field model for this step S21 can be various forms known by those skilled in the art, and in one example of the present application, the step S21 can include: The oil film thickness model is established according to formula (5), , (5) wherein,​ for the oil film thickness field, for the base oil film thickness, for the center oil film thickness, for the shoe overturning angle, for the shoe overturning azimuth angle, for the radial coordinate, for the angular coordinate, for the shoe oil chamber radius, for the shoe bottom surface outer circle radius. In this example, the center point oil film thickness is set to {5, 6, 7, 8} , the shoe overturning angle is set to {0.001, 0.0015, 0.002}°, and the shoe overturning azimuth angle is set to {240, 245, 250}°.

[0029] Step S22 is configured to obtain a corresponding pressure field as a reference solution according to the Reynolds equation of the plunger pump friction pair. In this embodiment, the specific method for solving the pressure field reference solution of this step S22 can be in various forms known to those skilled in the art. In one example of the present application, this step S22 can include: establishing the Reynolds equation of the plunger pump shoe pair according to formula (6), , (6) wherein, is the oil film pressure field; inputting a given oil film thickness field into the Reynolds equation to obtain corresponding pressure field data as the oil film thickness field-oil film pressure field reference solution.

[0030] Specifically, in this embodiment, a specific oil film thickness field can be input into the Reynolds equation, and MATLAB or Pumplinx or other tools are used to obtain a corresponding pressure field as the oil film thickness field-pressure field reference solution, and the number of reference solutions is determined by the full permutation combination of the following parameters: center point oil film thickness, shoe overturning angle, shoe overturning azimuth angle, and operating condition. In this example, there are 72 groups (4x3x3x2=72) of oil film thickness field-pressure field reference solutions. The obtained oil film thickness field-pressure field reference solutions are divided into groups of training reference solutions and groups of verification reference solutions . In this example, is 60, is 12.

[0031] Step S23 is configured to discretely sample the oil film thickness field to obtain oil film thickness field sampling point data. In this embodiment, the specific method of obtaining the oil film thickness field sampling point data in step S23 can be various forms known to those skilled in the art, and in an example of the present application, step S23 can include: According to formula (7), the oil film thickness field is discretely sampled, and the corresponding , , (7) wherein, is a fixed sampling point, is the number of sampling points, is a radial coordinate, is an angular coordinate, is the radius of the oil chamber of the sliding shoe, is the outer circle radius of the sliding shoe bottom surface. All oil film thickness fields and share the same sampling point position to ensure consistency of the input dimension. In this example, may be 256.

[0032] Step S24 is configured to discretely sample the oil film pressure field to obtain oil film pressure field sampling point data. In this embodiment, the specific method of obtaining the oil film pressure field sampling point data in step S24 can be various forms known to those skilled in the art, and in an example of the present application, step S24 can include: According to formula (8), the oil film pressure field is discretely sampled, and the corresponding , , (8) wherein, is a random sampling point, is a uniform probability distribution, is a uniform probability distribution, is the radius of the oil chamber of the sliding shoe, is the outer circle radius of the sliding shoe bottom surface.

[0033] By the probability distribution , a total of spatial coordinates are randomly generated, and for any pressure field and , the spatial coordinates are not necessarily the same. In this example, may be 100.

[0034] Step S25 obtains a training data set according to the oil film thickness field sampling point data and the oil film pressure field sampling point data. Specifically, in this example, the training data set is: ; Validation dataset for: ; The number of samples in the training data set is and The product of , the number of samples in the validation dataset is and In this example, the number of samples in the training dataset is 6000, and the number of samples in the validation dataset is 1200.

[0035] Step S11 is used to construct a DeepONet-based oil film thickness field-oil film pressure field calculation model. In this embodiment, the specific method for constructing the oil film thickness field-oil film pressure field calculation model in step S11 can be various forms known to those skilled in the art. In one example of the present invention, step S11 may include: Figure 4 The steps shown in Figure 4 In the step S11, the following steps may be included: In step S30, a branch network and a backbone network are constructed.

[0036] In step S31, a global feature vector and a position feature vector are obtained; In step S32, the pressure value at a specified position of the oil film field is calculated.

[0037] In this Figure 4 In the method shown, step S30 is used to construct a branch network and a trunk network. Specifically, in this embodiment, the constructed oil film thickness field-pressure field DeepONet is used Indicated by the branch network and backbone networks It is used to approximate the global functional mapping relationship from the oil film thickness field to the oil film pressure field. The branch network and the backbone network of DeepONet were originally composed of multi-layer perceptrons, which can only capture local features and are difficult to model global features such as radial gradient pressure, axial symmetry and boundary conditions of the oil film field. Figure 6 As shown, the branch network of DeepONet in the present invention adopts a series structure of attention mechanism and multi-layer perceptron. The attention mechanism is used to capture global features, and the multi-layer perceptron is used to process local features. The backbone network retains the multi-layer perceptron structure to encode spatial coordinates and operating conditions. Both the branch network and the backbone network output F-dimensional feature vectors. In this example, F can be 16, and the key matrix of the branch network attention mechanism is , query matrix Sum Matrix are all 256x256 matrices, the multi-layer perceptron of the branch network comprises an input layer , a hidden layer , a hidden layer and an output layer , the number of neurons are 256, 64, 16, 16 respectively; the MLP of the backbone network comprises an input layer , a hidden layer , a hidden layer and an output layer , the number of neurons are 4, 8, 16, 16 respectively.

[0038] Step S31 is configured to input the oil film thickness values on the fixed sampling points into the branch network and output the global feature vector, and the backbone network is configured to input the randomly generated spatial coordinates and operating conditions and calculate the position feature vector. Specifically, the input of the branch network can be the oil film thickness values on the fixed sampling points , and the output is an F-dimensional global feature vector : ; The input of the backbone network is the randomly generated spatial coordinates , and the output is an F-dimensional position feature vector : .

[0039] Step S32 is configured to calculate the pressure value at the specified position of the oil film field. Specifically, in this example, the global feature vector output by the branch network and the position feature vector output by the backbone network can be dot producted to calculate the pressure value at the spatial coordinates : ; The input oil film thickness values of the fixed branch network are input into the backbone network in turn to calculate the pressure value at the specified position of the oil film field .

[0040] Step S12 is configured to train the oil film thickness field-oil film pressure field calculation model. In this embodiment, the specific method of training the oil film thickness field-oil film pressure field calculation model in this step S12 can be various forms known by those skilled in the art, and in one example of the present application, this step S12 can include the steps shown in Figure 5 . In this Figure 5 , the step S12 can include: In step S40, the training data set is input into the oil film thickness field-oil film pressure field calculation model, and the predicted pressure value is output; ​​In step S41, the loss function value is calculated; In step S42, it is judged whether the loss function value is greater than a threshold value; In step S43, in the case where the loss function value is greater than the threshold value, the parameters of the oil film thickness field-oil film pressure field calculation model are updated, and the pressure field is recalculated; In the case where the loss function value is less than or equal to the threshold value, the training is completed.

[0041] In the method as Figure 4 shown, steps S40 to S43 are used to train the oil film thickness field-pressure field DeepONet. In this embodiment, the training and accuracy evaluation of the oil film thickness field-pressure field DeepONet can be carried out in a Python environment. First, the parameters of the oil film thickness field-oil film pressure field DeepONet are initialized, including the key matrix , the query matrix , the value matrix , the weights of each neuron, the bias of each neuron of the branch network, the weights of each neuron of the trunk network, and the bias of each neuron of the trunk network. Then, the loss function threshold is set to . In this embodiment, is set to . Step S40 is used to obtain a predicted pressure value. The training data set is input into the DeepONet, and the pressure field is calculated according to a given oil film thickness field. Step S41 is used to calculate the loss function value. Although the DeepONet has the ability to learn functional mapping compared with the machine learning model, it is still essentially a data-driven black box model, which lacks the consistency of the corresponding oil film field physical information and boundary conditions, and limits the further improvement of the solving accuracy. Therefore, the present application adds the Reynolds equation in the form of physical constraints to the training process of the DeepONet. The pressure field calculated by the DeepONet is compared with the reference solution, and the loss function value is calculated according to the following formula: , (1) , (2) , (3) , (4) , (9) , (10) wherein, is the loss function, is the Reynolds equation constraint term, is the data constraint term, is the boundary condition constraint term, is the total number of training dataset samples, is the number of pressure field coordinate points in each sample, is the spatial coordinate in the polar coordinate system, is the oil film thickness, is the kinematic viscosity of the hydraulic oil, is the predicted pressure value of the oil film thickness field-oil film pressure field calculation model for the th sample at the coordinate , is the tangential velocity of the sliding shoe bottom surface relative to the swash plate, is the radial flow velocity caused by the change in oil film thickness, is the reference solution of the th sample at the coordinate , is the boundary pressure value, is the pressure solution of the th sample at the boundary position . Wherein, and can be obtained by automatic differentiation in Python.

[0042] Step S42 is used to determine whether the loss function value is greater than the threshold value. Step S43, in the case where the loss function value is greater than the threshold value, updates the parameters of the oil film thickness field-oil film pressure field calculation model according to the gradient descent, and recalculates the pressure field. In this example, the gradient descent update algorithm can be Adam. In the case where the loss function value is less than or equal to the threshold value, the training is completed.

[0043] Step S13 is used to calculate the pressure value at the randomly generated spatial coordinate using the trained oil film thickness field-oil film pressure field calculation model.

[0044] The present application also evaluates the precision of the oil film thickness field-pressure field DeepONet. Specifically, the validation dataset is input into the trained DeepONet and the oil film field is calculated, and the evaluation index is calculated according to formula (11): The smaller the evaluation index, the higher the precision of the DeepONet in calculating the oil film field: , (11) In this example, the trained DeepONet calculates the oil film field under two groups of working conditions: main shaft speed 1000 rpm, load pressure 9 MPa, and main shaft speed 5000 rpm, load pressure 21 MPa, and converts to the spatial rectangular coordinate system, as shown in Figure 7 (main shaft speed 1000 rpm, load pressure 9 MPa) and Figure 8(5000 rpm of main shaft rotation speed, 21 MPa of load pressure) is shown. The average time of DeepONet to calculate a single oil film field is 116.6 ms, and the evaluation index , and fast and accurate calculation of the oil film field of the plunger pump is realized. Meanwhile, after the DeepONet training is completed, according to the oil film thickness value of the input branch network, only the spatial coordinates input into the trunk network need to be changed, and the oil film pressure value at any position can be calculated. Therefore, the resolution of the oil film field calculated by DeepONet is only related to the number of spatial coordinates input into the trunk network, and high-precision and high-resolution fast calculation of the oil film field can be realized by increasing the number of spatial coordinates without retraining the network.

[0045] On the other hand, the application provides a DeepONet-based oil film field calculation system for a plunger pump, which comprises a processor configured to perform the method according to any one of the above.

[0046] The application has the following beneficial effects: The application avoids the process of repeatedly solving partial differential equations by constructing the oil film thickness field-pressure field DeepONet, and the DeepONet model can quickly realize pressure field calculation through only one forward propagation after training. Compared with the prior art, the application avoids complex grid division and iterative calculation, and the pressure field calculation is faster.

[0047] The application directly learns the global functional mapping relationship from the oil film thickness field to the oil film pressure field by constructing the oil film thickness field-pressure field DeepONet, the branch network is a series structure of an attention mechanism and an MLP, the oil film thickness field is discretized and global features are extracted, the trunk network encodes spatial coordinates and working conditions, and outputs position features, the global features and the position features are dot multiplied and summed to calculate the pressure value at any position in the pressure field domain, and the application has strong generalization ability and can still maintain high solving accuracy on data not participating in training. Meanwhile, the loss function proposed in the application contains a Reynolds equation constraint term, a data constraint term and a boundary condition constraint term, the consistency of the oil film field physical information and the boundary condition is fully considered, and the solving accuracy is further improved.

[0048] The oil film thickness field-pressure field DeepONet constructed by the application decouples the oil film thickness value and the spatial coordinates through the branch network and the trunk network, and only the spatial coordinates input into the trunk network need to be changed to calculate the oil film pressure value at any position. Therefore, the resolution of the oil film pressure field is only related to the number of spatial coordinates input into the trunk network, and high-resolution fast calculation of the oil film pressure field can be realized by increasing the number of spatial coordinates without retraining the network.

[0049] Those skilled in the art will appreciate that embodiments of the application can be readily used as software, hardware, or a combination of software and hardware. In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0050] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

[0051] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

[0052] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

[0053] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0054] The memory can include non-persistent memory, random access memory (RAM), and / or non-volatile memory, such as read only memory (ROM) or flash memory, among others. The memory is an example of computer-readable media.

[0055] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0056] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusions, so that a process, method, article or apparatus that includes a list of elements does not only include those elements, but also includes other elements not explicitly listed, or further includes elements inherent in such a process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0057] The above only is an embodiment of the present application, and is not used to limit the present application. For those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the scope of claims of the present application.

Claims

1. A plunger pump oil film field calculation method based on DeepONet, characterized in that: The calculation method includes: Get the training dataset; Construct an oil film thickness field-oil film pressure field calculation model based on DeepONet; The oil film thickness field-oil film pressure field calculation model is trained using the training data set, wherein the loss function of the oil film thickness field-oil film pressure field calculation model includes formulas (1) to (4): ，（1） ，（2） ，（3） ，（4） in, is the loss function, is the constraint term of the Reynolds equation, is the data constraint item, is the boundary condition constraint, is the total number of samples in the training dataset, is the number of pressure field coordinate points in each sample, is the spatial coordinate in the polar coordinate system, is the oil film thickness, is the kinematic viscosity of the hydraulic oil, The calculation model of oil film thickness field-oil film pressure field is Samples at coordinates The predicted pressure value at is the tangential velocity of the bottom surface of the sliding shoe relative to the swash plate, is the radial flow velocity caused by the change of oil film thickness, For the Samples at coordinates The reference solution at is the boundary pressure value, For the Samples at the boundary The pressure solution at ; The pressure values ​​at randomly generated spatial coordinates are calculated based on the trained oil film thickness field-oil film pressure field calculation model.

2. The calculation method according to claim 1, characterized in that Obtaining a training dataset includes: Establish the spatial polar coordinates of the oil film thickness field; Construct oil film thickness field model; Construct the Reynolds equation for the plunger pump slipper pair and solve the pressure field reference solution; Discretely sample the oil film thickness field to obtain oil film thickness field sampling point data; Discretely sample the oil film pressure field to obtain oil film pressure field sampling point data; A training data set is acquired based on the oil film thickness field sampling point data and the oil film pressure field sampling point data.

3. The calculation method according to claim 2, characterized in that Establishing the oil film thickness field spatial polar coordinates includes: The projection of the center of the bottom surface of the sliding shoe on the swash plate is taken as the origin , establish a contact surface between the oil film and the swash plate Axis and Shaft, establish the normal direction of the contact surface between the oil film and the swash plate axis.

4. The calculation method according to claim 2, characterized in that Constructing the oil film thickness field model includes: The oil film thickness model is established according to formula (5): ，（5） in, is the oil film thickness field, is the base oil film thickness, is the center oil film thickness, is the sliding shoe overturning angle, is the sliding shoe overturning azimuth, is the radial coordinate, is the angle coordinate, is the radius of the oil chamber of the skid shoe, is the outer radius of the bottom surface of the sliding shoe.

5. The calculation method according to claim 4, characterized in that Constructing the Reynolds equation for the plunger pump slipper pair and solving the pressure field reference solution includes: According to formula (6), the Reynolds equation of the plunger pump slipper pair is established. ，（6） in, is the oil film pressure field; The given oil film thickness field is input into the Reynolds equation to obtain the corresponding pressure field data as the reference solution of the oil film thickness field-oil film pressure field.

6. The calculation method according to claim 2, characterized in that: Discrete sampling of the oil film thickness field to obtain oil film thickness field sampling point data includes: According to formula (7), the oil film thickness field is discretized and the corresponding , ，（7） in, is a fixed sampling point, is the number of sampling points, is the radial coordinate, is the angle coordinate, is the radius of the oil chamber of the skid shoe, is the outer radius of the bottom surface of the sliding shoe.

7. The calculation method according to claim 2, characterized in that: Discrete sampling of the oil film pressure field to obtain oil film pressure field sampling point data includes: According to formula (8), the oil film pressure field is discretized and the corresponding , ，（8） in, are random sampling points, is a uniform probability distribution, is a uniform probability distribution, is the radius of the oil chamber of the skid shoe, is the outer radius of the bottom surface of the sliding shoe.

8. The calculation method according to claim 1, characterized in that The construction of the DeepONet-based oil film thickness field-oil film pressure field calculation model includes: Constructing a branch network and a backbone network, wherein the branch network adopts a series structure of an attention mechanism and a multi-layer perceptron, the attention mechanism is used to capture global features, and the multi-layer perceptron is used to process local features. The branch network is used to input oil film thickness values ​​and output global feature vectors; the backbone network is a multi-layer perceptron structure, which is used to input spatial coordinates and operating conditions and calculate position feature vectors; A pressure value is calculated according to the global eigenvector and the position eigenvector.

9. The calculation method according to claim 1, characterized in that: Using the training data set to train the oil film thickness field-oil film pressure field calculation model includes: Inputting the training data set into the oil film thickness field-oil film pressure field calculation model and outputting a predicted pressure value; Calculate the loss function value; Determine whether the loss function value is greater than a threshold; When it is determined that the loss function value is greater than a threshold value, updating the parameters of the oil film thickness field-oil film pressure field calculation model and recalculating the pressure field; When it is determined that the loss function value is less than or equal to the threshold, the training is completed.

10. A plunger pump oil film field calculation system based on DeepONet, characterized in that: The system comprises a processor configured to execute the method according to any one of claims 1 to 9.

Images