Digital twin and neural network three-cylinder pump pressure measurement method, system and medium

Abstract

This invention discloses a method, system, and medium for measuring the pressure of a three-cylinder pump using digital twins and neural networks. The method inputs time-series data of the pump parameters during operation into a trained physical information neural network model to obtain the pressure field information of the outlet manifold. The training process of the physical information neural network model is as follows: a simulation model mapped to the physical three-cylinder pump is constructed using digital twin technology to obtain time-series data of pump parameters under normal and abnormal conditions. After preprocessing, a training set is constructed. Partial differential equation terms are embedded in the physical information neural network model as physical loss functions for physical constraints. The physical information neural network model is trained using the training set, and trainable parameters are adjusted through back-up updates. This method, system, and medium for measuring the pressure of a three-cylinder pump not only conforms to the constraints of physical knowledge but also reflects the real-time changes in field data, enhancing the reliability and stability of pressure prediction.

Description

Technical Field

[0001] This invention relates to the field of three-cylinder pump pressure measurement technology, and in particular to a three-cylinder pump pressure measurement method, system, and medium based on digital twins and neural networks. Background Technology

[0002] As a key fluid machinery in the industrial field, the pressure field prediction of three-cylinder pumps is crucial for efficient equipment operation and fault early warning. Traditional methods mainly rely on physical modeling or data-driven techniques: physical modeling solves for pressure distribution based on fluid dynamics equations, but requires precise knowledge of fluid parameters and boundary conditions, resulting in high computational complexity and difficulty in adapting to complex operating conditions such as multiphase flow; data-driven methods fit the relationship between pressure and operating parameters through machine learning, but rely on a large amount of high-quality data, and the model lacks physical interpretability, so the prediction results may deviate from actual physical laws. In addition, in practical applications, the measured data of three-cylinder pumps is often limited by sensor deployment, resulting in data gaps or noise issues, while simulation data generated solely by digital twins faces the challenge of insufficient accuracy in physical modeling, making it difficult for traditional techniques to achieve high-precision real-time prediction in multivariate coupled scenarios.

[0003] To address the aforementioned issues, the disconnect between data-driven and physical modeling in existing technologies leads to a contradiction between accuracy and real-time performance in three-cylinder pump pressure prediction: physical models are computationally time-consuming and difficult to meet online monitoring requirements; while lightweight data models improve speed, the lack of physical constraints can easily cause prediction biases. How to effectively integrate prior physical knowledge with data characteristics to achieve high-precision, interpretable prediction of the three-cylinder pump pressure field has become a key technical challenge in the field of fluid machinery. Summary of the Invention

[0004] Based on the technical problems existing in the background technology, this invention proposes a three-cylinder pump pressure measurement method, system and medium based on digital twin and neural network, which not only conforms to the constraints of physical knowledge, but also can reflect the changing patterns of field data in real time, thereby enhancing the reliability and stability of pressure prediction.

[0005] The digital twin and neural network-based three-cylinder pump pressure measurement method proposed in this invention inputs the time-series data of the pump parameters during the operation of the three-cylinder pump into a trained physical information neural network model to obtain the pressure field information of the outlet manifold cavity.

[0006] The training process of the physical information neural network model is as follows:

[0007] A simulation model mapping to a physical three-cylinder pump was constructed using digital twin technology to obtain time-series data of pump parameters under normal and abnormal conditions. After preprocessing, a training set was constructed.

[0008] Partial differential equation terms are embedded in the physical information neural network model as physical loss functions for physical constraints. The physical information neural network model is trained using a training set, and the trainable parameters are adjusted by inverse updating.

[0009] Furthermore, the time-series data of the pump parameters includes the pressure of the inlet manifold cavity, the pressure of the outlet manifold cavity, the motor current, the liquid mass flow rate of the inlet manifold cavity, and the time data at the corresponding moment.

[0010] Furthermore, after obtaining the time-series data, the preprocessing procedure for the time-series data is as follows:

[0011] Remove outliers and duplicate values ​​of adjacent data points from the time series data, and divide it into training set, validation set and test set according to time order;

[0012] Standardize the data in the time series data, excluding time data.

[0013] The time data values ​​in the time series data are reduced by a set factor so that the reduced time data is used only for partial differential equation terms.

[0014] Furthermore, the partial differential equation terms include partial derivative terms, and the physical loss function The formula is as follows:

[0015] ;

[0016] in, It is the volume of the outlet manifold lumen. It is the equivalent density of the fluid mixture inside the three-cylinder pump. The pressure in the outlet manifold lumen. For time values, It is the mass flow rate of the liquid in the inlet manifold. For sample index, This represents the total number of samples.

[0017] Furthermore, when the three-cylinder pump is in pure oil mode, among which The calculation formula is as follows:

[0018] ;

[0019] in, It is the density of the oil at atmospheric pressure. This refers to the bulk modulus of the oil. The current pressure in the outlet manifold lumen, Atmospheric pressure.

[0020] Furthermore, when the three-cylinder pump is in oil-air mixture mode, where The calculation formula is as follows:

[0021] ;

[0022] in, For the density of the mixture, This refers to the bulk modulus of the oil. Atmospheric pressure The air variability index, The initial mixing density, This represents the normalized percentage of air volume.

[0023] Furthermore, the aforementioned The calculation formula is as follows:

[0024] ;

[0025] in, It is the density of the oil at atmospheric pressure. The density of the gas at atmospheric pressure. Air fraction represents the percentage of air mixed in with a liquid.

[0026] Furthermore, the physical information neural network model is trained using a total loss function, which includes a physical loss function and a mean absolute error loss function. Specifically:

[0027] ;

[0028] in, Let the mean absolute error loss function be . For physical loss function, This represents the weight of physical loss.

[0029] A computer system includes a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the method described above by the computer program.

[0030] A computer-readable storage medium storing a plurality of classification programs, the plurality of classification programs being invoked by a processor to execute the method described above.

[0031] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as ROM, RAM, magnetic disk, or optical disk.

[0032] The advantages of the digital twin and neural network-based three-cylinder pump pressure measurement method, system, and medium provided by this invention are as follows: A simulation model is constructed using digital twin technology to obtain time-series data, alleviating the data acquisition challenge. Partial differential equations conforming to the modeling object are embedded into a physical information neural network model. This model integrates prior physical knowledge with data features, improving the accuracy and real-time performance of pressure field prediction while ensuring physical interpretability. Furthermore, the embedded partial differential equation terms serve as physical constraints to train and adjust the hyperparameter optimization model. Ultimately, a three-cylinder pump pressure measurement method that conforms to physical constraints and can reflect real-time changes in field data is obtained, enhancing the reliability and stability of pressure prediction. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the process of the present invention;

[0034] Figure 2 This is a flowchart of the training process for a physical information neural network model. Detailed Implementation

[0035] The technical solution of the present invention will now be described in detail through specific embodiments. Many specific details are set forth in the following description to provide a thorough understanding of the invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0036] like Figures 1 to 2 As shown, the digital twin and neural network-based three-cylinder pump pressure measurement method proposed in this invention inputs the pump parameter time series data during the operation of the three-cylinder pump into a trained physical information neural network model to obtain the pressure field information of the outlet manifold cavity.

[0037] The training process of the physical information neural network model is as follows:

[0038] Step 1: Construct a simulation model that maps to the physical three-cylinder pump using digital twin technology, thereby obtaining time-series data of pump parameters under normal and abnormal conditions, and constructing a training set after preprocessing.

[0039] Step 2: Embed partial differential equation terms as physical loss functions for physical constraints into the physical information neural network model, train the physical information neural network model using the training set, and adjust the trainable parameters by back-updating.

[0040] This embodiment addresses the problems in existing technologies, such as the complexity of physical modeling calculations, difficulty in adapting to complex working conditions, reliance on large amounts of high-quality data for data-driven approaches, and the contradiction between accuracy and real-time performance caused by the separation of these two aspects. It utilizes digital twin technology to construct a simulation model to acquire time-series data, alleviating the data acquisition difficulties. Partial differential equations conforming to the modeling object are embedded into a Physics-informed Neural Network (PINN) model. By leveraging the PINN model to integrate prior physical knowledge and data features, the accuracy and real-time performance of pressure field prediction are improved while ensuring physical interpretability. Furthermore, the embedded partial differential equation terms serve as physical constraints to train and adjust the hyperparameter optimization model. Ultimately, a three-cylinder pump pressure measurement method is obtained that conforms to physical constraints and can reflect real-time changes in on-site data, enhancing the reliability and stability of pressure prediction.

[0041] In one embodiment, step one involves using digital twin technology to obtain time-series data of pump parameters under normal and abnormal conditions through a simulation model mapped to a physical three-cylinder pump. After preprocessing, a training set is constructed, specifically as follows:

[0042] (a1) Construct a simulation model of the three-cylinder pump using Simulink. Various existing methods can be used for the simulation model, such as the Simulink simulation model of the Matlab series, the multiphysics coupling model constructed by general finite element analysis software such as ANSYS, and the flow field simulation model built by professional CFD software such as CFX and Fluent.

[0043] (a2) By running the simulation model, time series data of pump parameters under normal and abnormal conditions are obtained, and the data are divided in sequence to obtain training set, validation set and test set;

[0044] Abnormal conditions include leakage, blockage, and bearing wear. The timing data of pump parameters during the operation of the three-cylinder pump include the pressure of the inlet manifold cavity, the pressure of the outlet manifold cavity, the motor current, the liquid mass flow rate of the inlet manifold cavity, and the time data at the corresponding moment.

[0045] (a3) Standardize the data in the time series data, excluding the time data;

[0046] (1)

[0047] in, For standardized data, For data before standardization, They are respectively The mean and standard deviation of the attribute set in which it is located;

[0048] (a4) Reduce the value of time data in the time series data by a set factor so that the reduced time data is only used for partial differential equation terms;

[0049] The time data value is reduced by a set factor (e.g., 10,000 times) to reduce its weight in the physical information neural network model, so that the time data is only used for the differentiation of partial differential equation terms without affecting the training of the physical information neural network model.

[0050] (a5) After data preprocessing through (a1) to (a4), the training set is fed into the physical information neural network model for model training.

[0051] This embodiment uses a simulation model to obtain time-series data, which can solve the problem of missing abnormal data for three-cylinder pumps. The training set, validation set, and test set can be divided as needed. In this embodiment, it is preferred that the size of the training set accounts for more than 70% of the total dataset, so that the training set can reflect the characteristics of the total dataset.

[0052] In one embodiment, step two involves embedding partial differential equation terms as physical loss functions for physical constraints into the physical information neural network model, training the physical information neural network model using the training set, and adjusting the trainable parameters through reverse updates. Specifically:

[0053] (b1) Embed partial differential equation terms in the physical information neural network model as the physical loss function of physical constraints;

[0054] First, a theoretical physical model for predicting the pressure of a three-cylinder pump is constructed, with the following formula:

[0055] (2);

[0056] in, It is the volume of the outlet manifold lumen. It is the equivalent density of the fluid mixture inside the three-cylinder pump. The current pressure in the outlet manifold lumen, For time values, It is the mass flow rate of the liquid in the inlet manifold. For sample index, This represents the total number of samples.

[0057] The value of is known and constant. The value is calculated using PyTorch's automatic differentiation. PyTorch is an open-source deep learning framework developed by Facebook AI Research (FAIR). These are the input features of the model, and they are also known, i.e., the training set. The calculation formula is as follows:

[0058] When the three-cylinder pump is in pure oil mode:

[0059] (3);

[0060] in, It is the density of the oil at atmospheric pressure. This refers to the bulk modulus of the oil. The current pressure in the outlet manifold lumen, Atmospheric pressure.

[0061] When the three-cylinder pump is in oil-air mixture mode:

[0062] (4);

[0063] in, For mixed density, This refers to the bulk modulus of the oil. Atmospheric pressure The air variability index, The initial mixing density, This represents the normalized percentage of air volume.

[0064] against The calculation formula is as follows:

[0065] (5);

[0066] in, It is the density of the oil at atmospheric pressure. The density of the gas at atmospheric pressure. Air fraction represents the percentage of air mixed in with a liquid.

[0067] Normalized air volume percentage The calculation formula is as follows:

[0068] (6);

[0069] in, For air fraction, Atmospheric pressure The current pressure in the outlet manifold lumen, This represents the proportion of air entrained in the liquid.

[0070] (b2) Construct a physical information neural network model;

[0071] The training set used as input to the physical information neural network model includes the pressure in the inlet manifold, the pressure in the outlet manifold, the motor current, the liquid mass flow rate in the inlet manifold, and the time data at the corresponding time points.

[0072] The physical information neural network model consists of an input layer, a fully connected layer, an output layer, and a loss function layer. The input layer receives time data from the operation of the three-cylinder pump. Motor current Pressure of the inlet manifold Current pressure in the outlet manifold Liquid mass flow rate in the inlet manifold The hidden layers in the fully connected layer use Swish as the activation function; the output layer is used for output. Pressure in the outlet manifold of the three-cylinder pump at all times In contrast The derivative terms of t are calculated automatically, making it easier to incorporate them into the loss function. The physical loss function in backpropagation is calculated using the method of Sino-Israeli.

[0073] It should be noted that the key idea of ​​the physical information neural network model in this embodiment is to embed physical constraints into the neural network, thereby enabling the network to learn the behavior of the physical system and satisfy the physical equations. In this embodiment, the fully connected layer consists of 6 hidden layers, each containing 258 neurons, and uses Swish as the activation function.

[0074] because When the physical information is fed into the neural network model, it is scaled by a set factor (e.g., 10,000 times). According to the following formula, the automatically differentiated value needs to be divided by 10,000 to obtain the actual value. :

[0075] The actual time for the input physical information neural network model is ,and According to the chain rule:

[0076] (7);

[0077] And because:

[0078] (8);

[0079] therefore:

[0080] (9);

[0081] Based on the above description, this embodiment proposes a total loss function for the physical information neural network model. as follows:

[0082] (10);

[0083] in, Let the mean absolute error loss function be . For physical loss function, This represents the weight of physical loss.

[0084] in,

[0085] ,(11)

[0086] (12);

[0087] In the formula, For the first The pressure prediction result for each sample, i.e., the predicted value (the pressure field information of the outlet manifold lumen predicted by the physical information neural network model). The pressure simulation data for the i-th sample is the actual value (the actual pressure field information of the outlet manifold). This represents the total number of samples.

[0088] It can intuitively reflect the data error between the actual value and the predicted value of the physical information neural network model, and can be used to judge whether the accuracy of the prediction result meets the requirements of actual use. The training of the physical information neural network model can be constrained, enabling the physical information neural network model to learn the behavior of the physical system and satisfy the physical equation (Equation (2)) as much as possible.

[0089] The physical information neural network model is trained based on the above total loss function and the input training set. By adjusting the trainable parameters, a physical information neural network model with better prediction performance can be found. The trainable parameters include batch size, hidden layer size, number of hidden layers, learning rate, and weights of physical loss.

[0090] During the forward propagation of the physical information neural network model, the model outputs the magnetic induction intensity and the pressure in the outlet manifold of the three-cylinder pump. Its relative and The derivative terms at each time step are calculated using automatic differentiation, allowing them to be substituted into the physical loss function (Equation (11)) for backpropagation during model optimization, ensuring that the model satisfies the constraints of the physical equations. Automatic differentiation is performed using the PyTorch framework's Autograd function. Its principle is to dynamically construct a computation graph and automatically calculate the gradient of each parameter during backpropagation using the chain rule, thus ensuring that the model considers both data error (data-driven loss) and physical loss when updating parameters. Also consider the error in the physical equations (the loss driven by physics, i.e.) ).

[0091] When training a physical information neural network model, the total loss function can be optimized through multiple iterations. The value of the total loss function approaches 0. When the value of the total loss function approaches 0, the model can be considered an approximation of the three-cylinder pump. Thus, the PINN network training is complete. During training, the validation set is used to verify whether the trained physical information neural network model meets the required performance requirements. Finally, the test set is used to test the trained physical information neural network model, and finally, the trained physical information neural network model is used to predict the pressure of the three-cylinder pump.

[0092] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for measuring the pressure of a three-cylinder pump using digital twins and neural networks, characterized in that, The time-series data of the pump parameters when the three-cylinder pump is working are input into the trained physical information neural network model to obtain the pressure field information of the outlet manifold cavity. The training process of the physical information neural network model is as follows: A simulation model mapping to a physical three-cylinder pump was constructed using digital twin technology to obtain time-series data of pump parameters under normal and abnormal conditions. After preprocessing, a training set was constructed. Partial differential equation terms are embedded in the physical information neural network model as physical loss functions for physical constraints. The physical information neural network model is trained using the training set, and the trainable parameters are adjusted by back-updating. The partial differential equation terms include partial derivative terms, and the physical loss function The formula is as follows: in, It is the volume of the outlet manifold lumen. It is the equivalent density of the fluid mixture inside the three-cylinder pump. The current pressure in the outlet manifold lumen, For time values, It is the mass flow rate of the liquid in the inlet manifold. For sample index, This represents the total number of samples.

2. The method for measuring the pressure of a three-cylinder pump according to claim 1, characterized in that, The time-series data of the pump parameters include the pressure in the inlet manifold, the pressure in the outlet manifold, the motor current, the liquid mass flow rate in the inlet manifold, and the time data at the corresponding moment.

3. The method for measuring the pressure of a three-cylinder pump according to claim 1, characterized in that, After obtaining the time series data, the preprocessing process for the time series data is as follows: Remove outliers and duplicate values ​​of adjacent data points from the time series data, and divide it into training set, validation set and test set according to time order; Standardize the data in the time series data, excluding time data. The time data values ​​in the time series data are reduced by a set factor so that the reduced time data is used only for partial differential equation terms.

4. The method for measuring the pressure of a three-cylinder pump according to claim 1, characterized in that, When the three-cylinder pump is in pure oil mode, among which The calculation formula is as follows: in, It is the density of the oil at atmospheric pressure. This refers to the bulk modulus of the oil. Atmospheric pressure.

5. The method for measuring the pressure of a three-cylinder pump according to claim 1, characterized in that, When the three-cylinder pump is in oil-air mixture mode, The calculation formula is as follows: in, For the density of the mixture, This refers to the bulk modulus of the oil. Due to current pressure, Atmospheric pressure The air variability index, The initial mixing density, This represents the normalized percentage of air volume.

6. The method for measuring the pressure of a three-cylinder pump according to claim 5, characterized in that, The The calculation formula is as follows: in, It is the density of the oil at atmospheric pressure. The density of the gas at atmospheric pressure. Air fraction represents the percentage of air mixed in with a liquid.

7. The method for measuring the pressure of a three-cylinder pump according to claim 1, characterized in that, The physical information neural network model is trained using a total loss function, which includes a physical loss function and a mean absolute error loss function. Specifically: in, Let the mean absolute error loss function be . For physical loss function, This represents the weight of physical loss.

8. A computer system comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the method according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of classification programs, which are used by a processor to execute the method as described in any one of claims 1-7.

Images