An edge cloud cooperative electro-hydraulic actuator digital twin modeling method and system

Abstract

The application discloses an edge-cloud collaborative electro-hydraulic actuator digital twin modeling method and system, wherein the edge-cloud collaborative electro-hydraulic actuator digital twin modeling method comprises the following steps: constructing a high-fidelity simulation model of a target electro-hydraulic actuator by using simulation software, and generating simulation data by using the simulation model; constructing a proxy model of the running characteristics of the target electro-hydraulic actuator by using a reduced-order model and deep learning, and deploying the proxy model on an edge device and a cloud server according to the computing power requirement and the input parameter quantity; collecting the running state data of the target electro-hydraulic actuator in real time; inputting the running state data into the proxy model to perform real-time model reasoning on the edge device and the cloud server, so as to predict the physical field running characteristics of the target electro-hydraulic actuator under the current actual working condition; and performing data fusion and visual processing on the physical field running characteristics, so as to construct a digital twin reflecting the running state of the target electro-hydraulic actuator. The method has real-time performance and high modeling precision.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of electro-hydraulic actuator health monitoring, and particularly relates to an edge-cloud collaborative electro-hydraulic actuator digital twin modeling method and system. BACKGROUND

[0002] The electro-hydraulic actuator is a key execution unit in a hydraulic system, has advantages of high response speed, high-precision control and high dynamic performance, and has been widely applied in fields of aerospace, engineering machinery, robots and the like, and its health state will directly affect the working efficiency and safety of the hydraulic system.

[0003] The traditional health monitoring method relies on offline data or a single sensor, and has problems of insufficient real-time performance and low data utilization rate. SUMMARY

[0004] The application aims to provide an edge-cloud collaborative electro-hydraulic actuator digital twin modeling method and system to efficiently integrate sensor data under an edge-cloud collaborative architecture and realize virtual-real real-time interaction.

[0005] According to a first aspect of an embodiment of the application, an edge-cloud collaborative electro-hydraulic actuator digital twin modeling method is provided, which can include:

[0006] A high-fidelity simulation model of a target electro-hydraulic actuator is constructed by using simulation software, and simulation data is generated by using the simulation model;

[0007] A proxy model of the running characteristics of the target electro-hydraulic actuator is constructed by using a reduced-order model and deep learning, and is deployed on an edge device and a cloud server according to the computing power requirement and the input parameter quantity;

[0008] The running state data of the target electro-hydraulic actuator is collected in real time;

[0009] The running state data is input into the proxy model to perform real-time model reasoning on the edge device and the cloud server to predict the physical field running characteristics of the target electro-hydraulic actuator under the current actual working condition;

[0010] The physical field running characteristics are subjected to data fusion and visual processing to construct a digital twin reflecting the running state of the target electro-hydraulic actuator.

[0011] In some optional embodiments of the application, the high-fidelity simulation model of the target electro-hydraulic actuator is constructed by using simulation software, including:

[0012] A structural stress simulation model and an electromagnetic simulation model of the target electro-hydraulic actuator are constructed by using simulation software;

[0013] A fluid simulation model of the target electro-hydraulic actuator is constructed by using simulation software;

[0014] A system-level simulation model of the target electro-hydraulic actuator is constructed by using simulation software.

[0015] In some optional embodiments of the present application, a proxy model of the operating characteristics of the target electro-hydraulic actuator is constructed by using a reduced-order model and deep learning, comprising:

[0016] Simulation result sample points of the simulation working condition variables are generated by using a Latin hypercube sampling method, and simulation results of the simulation working condition variable sample points are output by using the simulation model;

[0017] The simulation results are used to construct a reduced-order model by using an intrinsic orthogonal decomposition method, so as to obtain low-dimensional outputs corresponding to each sample and form a training data set;

[0018] The proxy model is trained by using the training data set, and a trained proxy model is obtained.

[0019] In some optional embodiments of the present application, the simulation results are used to construct a reduced-order model by using an intrinsic orthogonal decomposition method, so as to obtain low-dimensional outputs corresponding to each sample and form a training data set, comprising:

[0020] The simulation results are subjected to intrinsic orthogonal decomposition;

[0021] The singular values and corresponding singular vectors are selected to construct a reduced-order model;

[0022] The simulation results are subjected to dimension reduction by using the reduced-order model to obtain low-dimensional outputs;

[0023] The simulation results and the corresponding low-dimensional outputs constitute training data pairs to obtain a training data set.

[0024] In some optional embodiments of the present application, the operating state data of the target electro-hydraulic actuator is collected in real time, comprising:

[0025] The original time step used in training of the proxy model is obtained;

[0026] The operating state data of the target electro-hydraulic actuator is collected in real time by using physical sensors at the original time step.

[0027] In some optional embodiments of the present application, the operating state data is input into the proxy model to perform real-time model inference on the edge device and the cloud server, comprising:

[0028] The electro-hydraulic actuator system performance model inference is performed by using the edge device;

[0029] The piston rod structure stress model inference is performed by using the edge device;

[0030] The force motor electromagnetic model inference is performed by using the edge device;

[0031] Performing a sealing leakage model inference using a cloud server

[0032] Performing an electromagnetic valve fluid model inference using a cloud server.

[0033] In some optional embodiments of the present application, data fusion and visualization processing are performed on physical field operation characteristics to construct a digital twin reflecting the operation state of the target electro-hydraulic actuator, including:

[0034] Loading a three-dimensional model file and a corresponding vertex information file;

[0035] Loading inference results of a surrogate model and a corresponding node information file;

[0036] Using a spatial matching algorithm to pair and combine the vertices in the three-dimensional model with the nodes in the inference results of the surrogate model;

[0037] Mapping and normalizing the inference results of the surrogate model to the model vertices to obtain result values;

[0038] Mapping the result values to a color gradient and using vertex coloring to recolor the model.

[0039] According to a second aspect of the embodiments of the present application, an edge-cloud collaborative electro-hydraulic actuator digital twin modeling system is provided, characterized by comprising a processor, a memory, and a program or instructions stored in the memory and executable on the processor, and the program or instructions are executed by the processor to implement the steps of the edge-cloud collaborative electro-hydraulic actuator digital twin modeling method according to any one of the first aspect embodiments.

[0040] The above technical solutions of the present application have the following beneficial technical effects:

[0041] The method of the embodiments of the present application has real-time performance and high modeling accuracy. Through edge-cloud collaborative computing technology, combined with the low-latency processing of edge devices and the powerful computing capability of cloud servers, real-time high-precision modeling of electro-hydraulic actuator state information is realized, so that health monitoring is completed in a shorter time, and the response speed and prediction accuracy are improved. The method is efficient, realizes resource optimization, reduces the dependence on traditional simulation models by using model reduction and surrogate model technology, and reduces the computational complexity. At the same time, the edge device and the cloud server work collaboratively to optimize the allocation of computing resources and improve the overall operation efficiency of the monitoring system. In addition, the method is intuitive, realizes comprehensive digital twin construction and visualization of the electro-hydraulic actuator, and intuitively displays the operation state of the device, which is convenient for monitoring and analysis. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1is a flowchart of a method for modeling a digital twin of an electro-hydraulic actuator in edge cloud cooperation in an example embodiment of the present application;

[0043] Figure 2 is a flowchart of a method for modeling a digital twin in another example embodiment of the present application;

[0044] Figure 3 is a flowchart of a method for training a surrogate model in an example embodiment of the present application;

[0045] Figure 4 is a flowchart of a method for reasoning of a surrogate model in an example embodiment of the present application;

[0046] Figure 5 is a flowchart of a method for three-dimensional visualization of a reasoning result in an example embodiment of the present application;

[0047] Figure 6 is a block diagram of a digital twin running system in an example embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the objectives, technical solutions and advantages of the present application clearer, further detailed description will be made to the present application with reference to the specific embodiments and the accompanying drawings. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present application. In addition, in the following description, the description of the known structures and technologies is omitted to avoid unnecessary confusion of the concepts of the present application.

[0049] In the accompanying drawings, schematic diagrams of layer structures according to embodiments of the present application are shown. These diagrams are not drawn to scale, in which certain details are exaggerated for the purpose of clarity, and certain details may be omitted. The shapes of various regions, layers, and their relative sizes and positional relationships shown in the diagrams are only exemplary, and in actuality, they may deviate due to manufacturing tolerances or technical limitations, and a person skilled in the art can additionally design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0050] Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without making creative efforts fall within the scope of protection of the present application.

[0051] In the description of the present application, it should be noted that the terms “first”, “second”, “third” are only for the purpose of description, and cannot be understood or implied as indicating or suggesting relative importance.

[0052] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as there is no conflict.

[0053] Digital twin technology can realize accurate simulation and prediction by constructing virtual digital twin, and can reflect the running state of equipment in real time, and become an important means of fault warning. However, constructing and updating digital twin needs to process a large amount of real-time data, and it is difficult for edge terminal or cloud platform to meet the requirements of low delay and high performance at the same time.

[0054] Edge-cloud collaborative computing technology can integrate the low delay advantage of edge terminal and the powerful data processing capability of cloud terminal, realize real-time high-precision modeling of electro-hydraulic actuator state information, and provide a new technical path for health monitoring, fault prediction and system optimization. At present, how to efficiently integrate sensor data under the edge-cloud collaborative architecture, realize real-time interaction between virtual and real, and use artificial intelligence algorithm for fault warning still faces challenges.

[0055] The edge-cloud collaborative electro-hydraulic actuator digital twin modeling method and system provided by the embodiments of the present application will be described in detail below in combination with the drawings and specific embodiments and their application scenarios.

[0056] As shown in Figure 1 In a first aspect of the embodiments of the present application, an edge-cloud collaborative electro-hydraulic actuator digital twin modeling method is provided, which can include:

[0057] S110: Constructing a high-fidelity simulation model of the target electro-hydraulic actuator by using simulation software, and generating simulation data by using the simulation model;

[0058] S120: Constructing a proxy model of the running characteristics of the target electro-hydraulic actuator by using a reduced-order model and deep learning, and deploying it on edge devices and cloud servers according to the computing power requirement and the input parameter quantity;

[0059] S130: Real-time acquisition of the running state data of the target electro-hydraulic actuator;

[0060] S140: Inputting the running state data into the proxy model to perform real-time model reasoning on the edge devices and cloud servers to predict the physical field running characteristics of the target electro-hydraulic actuator under the current actual working condition;

[0061] S150: Data fusion and visualization processing of the physical field running characteristics to construct a digital twin reflecting the running state of the target electro-hydraulic actuator.

[0062] The method has real-time performance, high modeling accuracy, and through edge-cloud collaborative computing technology, combines the low-latency processing of edge devices and the powerful computing power of cloud servers to realize real-time high-precision modeling of electro-hydraulic actuator state information, so that health monitoring is completed in a shorter time, improving response speed and prediction accuracy; and the method has high efficiency, realizes resource optimization, uses model reduction and proxy model technology to reduce dependence on traditional simulation models and reduce computational complexity; at the same time, edge devices and cloud servers work collaboratively to optimize the allocation of computing resources and improve the overall operating efficiency of the monitoring system; in addition, the method has intuitiveness, realizes comprehensive digital twin construction and visualization of the electro-hydraulic actuator, and intuitively displays the running state of the device, facilitating monitoring and analysis.

[0063] Specifically, as shown in Figure 2 The digital twin modeling method can include:

[0064] An electro-hydraulic actuator simulation model is constructed using simulation software;

[0065] Simulation data is generated using the simulation model, an electro-hydraulic actuator operating characteristic proxy model is constructed using model reduction and deep learning technology, and is selectively deployed on edge devices and cloud servers according to its computing power requirements and input parameter quantity;

[0066] Real-time collection of electro-hydraulic actuator entity operating state data is performed using physical sensors;

[0067] Sensor data is input into the proxy model, real-time model inference is performed on the edge device or cloud server, and the operating characteristics of each physical field of the electro-hydraulic actuator under the current actual working condition are predicted;

[0068] The inference results of the proxy model are data fused and visualized on the cloud server side, a comprehensive digital twin reflecting the key operating state of the electro-hydraulic actuator is constructed, and is embedded into the monitoring platform.

[0069] Finally, the entire operating system is deployed in the field monitoring environment to realize edge-cloud collaborative real-time synchronization of the digital twin and the physical device.

[0070] In some embodiments, a high-fidelity simulation model of the target electro-hydraulic actuator is constructed using simulation software, including:

[0071] A structural stress simulation model and an electromagnetic simulation model of the target electro-hydraulic actuator are constructed using simulation software;

[0072] A fluid simulation model of the target electro-hydraulic actuator is constructed using simulation software;

[0073] A system-level simulation model of the target electro-hydraulic actuator is constructed using simulation software.

[0074] For the structural stress simulation model and the electromagnetic simulation model of the electro-hydraulic actuator, ANSYS, ABAQUS and other simulation software can be used to establish. For the fluid simulation model of the electro-hydraulic actuator, PumpLinx, ANSYS Fluent and other fluid simulation software or modules can be used to establish. For the system-level simulation model of the electro-hydraulic actuator, AMESim simulation software can be used to establish. The geometric characteristics and parameter settings of each simulation model must be consistent with the actual electro-hydraulic actuator to ensure the high fidelity and reliability of the model.

[0075] As shown in Figure 3 In some embodiments, a proxy model of the target electro-hydraulic actuator operating characteristics is constructed using a reduced-order model and deep learning, including:

[0076] Using Latin hypercube sampling method to generate simulation working condition variable sample points, using simulation model to output simulation results of simulation working condition variable sample points;

[0077] Using intrinsic orthogonal decomposition method to construct a reduced-order model from the simulation results, thereby obtaining a low-dimensional output corresponding to each sample to form a training data set;

[0078] And using the training data set to train the proxy model to obtain a trained proxy model.

[0079] In this embodiment, sample points are divided, simulation working condition variable sample points are generated using Latin hypercube sampling method, and simulation results under these working condition sample points are output using high-fidelity simulation model. Then, model reduction and neural network training are performed, intrinsic orthogonal decomposition method is used to construct a reduced-order model from the original output data, thereby obtaining a low-dimensional output corresponding to each sample to form a training data set, a suitable proxy model structure is selected, and the proxy model is trained. Next, model accuracy verification is performed, the accuracy of the trained proxy model is verified, and the error is calculated. Finally, model deployment is performed, and the proxy model is selected to run on the edge device or cloud server according to the complexity, input parameter amount and real-time requirement of the proxy model.

[0080] To ensure that the simulation data is complete and available, the simulation working condition sample points need to be evenly distributed in the multi-dimensional space, so Latin hypercube sampling method is used to generate simulation working condition variable sample points. The value range of a working condition variable is divided, and the width of the divided sub-interval is defined using the following formula:

[0081] Δ=(b-a) / N

[0082] In the formula, Δ is the width of each sub-interval, N is the total number of samples, a and b are the minimum and maximum values of the working condition variable.

[0083] A number is randomly selected in each sub-interval to obtain the sampling point of the working condition variable in this sub-interval:

[0084] x i = a + ((i - 1) + u i )Δ

[0085] where x i is the sampling point of the parameter in the i-th sub-interval, u i is a random number uniformly distributed in 0 to 1.

[0086] Similarly, for multi-dimensional working condition variables, the sampling points are defined by the following formula:

[0087] x ij = a j + [(π j (i) - 1) + u ij ](b j - a j ) / N

[0088] where x ij is the sampling point of the i-th sample in the j-th dimension, a j and b j are the minimum value and the maximum value of the working condition variable in the j-th dimension, π j is a randomly generated permutation (i.e. π j (1), π j (2), …, π j (N) is a random permutation of 1 to N), u ij is a random number uniformly distributed in 0 to 1.

[0089] Finally, the sampling points of all dimensions are combined into a multi-dimensional sample vector:

[0090] X i = (x i1 , x i2 , …, x im )

[0091] where X i is the combined multi-dimensional sample vector, and m is the dimension of the sample.

[0092] In some embodiments, the simulation results are used to construct a reduced-order model using the proper orthogonal decomposition method, thereby obtaining a low-dimensional output corresponding to each sample, forming a training data set, including:

[0093] Performing proper orthogonal decomposition on the simulation results;

[0094] Selecting singular values and corresponding singular vectors to construct a reduced-order model;

[0095] Using the reduced-order model to reduce the dimension of the simulation results to obtain a low-dimensional output;

[0096] The simulation result and the corresponding low-dimensional output form a training data pair to obtain a training data set.

[0097] Specifically, the sample vector is used to obtain the original output data of the simulation model to form an original output matrix:

[0098] Y = [y1 y2... y N ] T ∈R N×d

[0099] In the formula, Y is the original output matrix, y is the original output data of the simulation model, and d is the dimension of the high-dimensional output.

[0100] Eigen-orthogonal decomposition is performed on the original output, that is, singular value decomposition is performed on the matrix Y:

[0101] Y = U∑V T

[0102] In the formula, U∈R N×N is an orthogonal matrix,∑∈R N×d is a singular value matrix, and V∈R d×d is a right singular vector matrix.

[0103] A plurality of main singular values and corresponding singular vectors are selected to construct a reduced-order model, so as to reduce the high-dimensional output data to a low-dimensional representation, and the low-dimensional output of each sample is represented as:

[0104] y i,reduced = [U k ∑ k ] i,: ∈R k ,k=d

[0105] In the formula, y i,reduced is the low-dimensional output of the i-th sample, k is the number of selected main singular values, U k ∈R N×k is a left singular vector matrix,∑ k ∈R k×k is a singular value matrix, and V k ∈R d×k is a right singular vector matrix.

[0106] The input sample and the corresponding low-dimensional output form a training data pair, and a suitable proxy model structure is selected to construct a mapping relationship and perform interpolation.

[0107] In some embodiments, a long short-term memory network can be used to construct the mapping relationship, which is particularly suitable for real-time prediction occasions.

[0108] In other embodiments, a radial basis function network can be employed to construct the mapping relationship, which is suitable for cases with large amounts of data and low real-time requirements.

[0109] As shown in FIG. 8, the agent model training process can select one or more representative working condition samples that do not belong to the sample vectors, input these samples into the trained reduced-order agent model to obtain the corresponding predicted output, input the same working condition samples into the original simulation model to obtain the original simulation results thereof, and compare the predicted output of the agent model with the simulation results of the original simulation model to calculate the mean square error and the average error. Figure 3

[0110] In some embodiments, for an agent model with a large number of input parameters or high real-time requirements, the agent model is deployed in an edge device, and for an agent model with high computational complexity, large computing power requirements, or relying on a complex cloud technology stack, the agent model is deployed in a cloud server.

[0111] In some embodiments, the running state data of the target electro-hydraulic actuator is collected in real time, including:

[0112] The original time step used in the training of the agent model is obtained.

[0113] The running state data of the target electro-hydraulic actuator is collected in real time by a physical sensor at the original time step.

[0114] To meet the input requirements of the agent model, for the collection of an actual running state data signal of the electro-hydraulic actuator, the sampling frequency should be consistent with the reciprocal of the original time step used in the training of the corresponding agent model:

[0115] f i = 1 / Δt i

[0116] In the formula, f i is the sampling frequency of the i-th state data signal, and Δt i is the time step used in the training of the agent model corresponding to the state data, with the unit of s.

[0117] As shown in FIG. 9, in some embodiments, the running state data is input into the agent model for real-time model inference on the edge device and the cloud server, including: Figure 4 The performance model of the electro-hydraulic actuator system is inferred by the edge device.

[0118] The piston rod structure stress model is inferred by the edge device.

[0119] The force motor electromagnetic model is inferred by the edge device.

[0120]

[0121] ​​Seal leakage model reasoning is performed by using a cloud server.

[0122] Electromagnetic valve fluid model reasoning is performed by using a cloud server.

[0123] The embodiment collects key operating state data of the electro-hydraulic actuator, such as oil pressure, oil temperature, pipeline flow, controller current and voltage, according to the sampling frequency of the physical sensor in the foregoing embodiment. Meanwhile, the connection with the MQTT Broker is established, and current working condition data such as control parameters are received.

[0124] For the agent model deployed on the edge device, the collected data is directly used as the input parameter for model reasoning. For the agent model deployed on the cloud server, the corresponding agent model is used for reasoning after the collected data is reported to the MQTT Broker. For the agent model with input and output mutual coupling and dependence, the reasoning sequence needs to be determined.

[0125] The system performance model is derived from the AMESim simulation software, the agent model structure is based on the long short-term memory network, and is saved in the onnx format and is used for reasoning in the edge device through the onnxruntime.

[0126] The input of the model is the motion amplitude, frequency, cycle number and external load force in the control parameters, and the output is the actuator cylinder displacement, electromagnetic spool displacement, electromagnetic force motor driving current, inner cavity pressure, outer cavity pressure and force acting on the electromagnetic spool.

[0127] The piston rod structure stress model is derived from the ANSYS simulation software, the agent model structure is based on the radial basis function network, and is saved in the pkl format and is used for reasoning in the edge device through the joblib library.

[0128] The input of the model is the external load force in the control parameters, and the output is the piston rod internal node stress, cylinder wall two-point stress and piston rod ear stress.

[0129] The force motor electromagnetic model is derived from the electromagnetic module of the ANSYS simulation software, the agent model structure is based on the radial basis function network, and is saved in the pkl format and is used for reasoning in the edge device through the joblib library.

[0130] The input of the model is the linear force motor permanent magnet magnetic flux and driving current, the external load force in the control parameters, and the electro-hydraulic actuator actuator cylinder displacement, and the output is the linear force motor internal node magnetic induction intensity and linear force motor output force.

[0131] The seal leakage model is derived from MATLAB Simulink simulation software, the structure of the surrogate model is based on a radial basis function network, and is saved in fmu format, and inference is performed in the cloud server through the fmpy library.

[0132] The input of the model is the amplitude, frequency, cycle number and external load force of the electro-hydraulic actuator motion in the control parameters, and the output of the surrogate model is the external cavity pressure, and the output is the dynamic seal wear amount and leakage rate.

[0133] The electromagnetic valve fluid model is derived from the ANSYS Fluent simulation module, the structure of the surrogate model is based on a radial basis function network, and the weight matrix, coordinate matrix and correction matrix are saved in csv format, and inference is performed in the cloud server through a specific JavaScript script.

[0134] The input of the model is the amplitude, frequency, cycle number and collected oil temperature of the electro-hydraulic actuator motion in the control parameters, and the output is the internal node pressure and internal node temperature of the electromagnetic valve.

[0135] As shown in Figure 5 In some embodiments, data fusion and visualization processing are performed on the physical field running characteristics to construct a digital twin that reflects the running state of the target electro-hydraulic actuator, including:

[0136] Load the three-dimensional model file and the corresponding vertex information file;

[0137] Load the inference results of the surrogate model and the corresponding node information file;

[0138] Use a spatial matching algorithm to pair and combine the vertices in the three-dimensional model with the nodes in the inference results of the surrogate model;

[0139] Map and normalize the inference results of the surrogate model to the model vertices to obtain result values;

[0140] Map the result values to a color gradient and use vertex shading to recolor the model.

[0141] In this embodiment, model files in formats such as fbx, stl and gltf can be used, the corresponding vertex information files can be saved in formats such as csv and json, and the vertex information file contains the spatial three-dimensional coordinates of all vertices of the three-dimensional model.

[0142] For three-dimensional model files with too sparse vertices, subdivision surface processing is also required to ensure that the information can be effectively mapped. Specifically, the number of vertices of the three-dimensional model file should be greater than the number of nodes corresponding to the output results of the surrogate model.

[0143] The agent model inference result can be derived from an agent model deployed on an edge device or from an agent model deployed on a cloud server. Corresponding node information files can be saved in formats such as csv and json. The node information files contain the spatial three-dimensional coordinates of all grid nodes divided during simulation.

[0144] If the three-dimensional model vertex coordinates and the grid node coordinates are stored in different length units, all coordinate values in the corresponding node information file need to be multiplied by the corresponding scale factor to convert them to the same length unit.

[0145] For vertex information files and node information files with the same length unit, all coordinate values are extracted to create point cloud objects. The node point cloud is aligned to the vertex point cloud using the iterative closest point algorithm.

[0146] Using the nearest neighbor search algorithm, the nearest several nodes of each vertex are matched.

[0147] In some embodiments, the number of matched nodes is 1.

[0148] In other embodiments, the number of matched nodes is defined using the following formula:

[0149]

[0150] In the formula, n is the number of matched nodes, V is the total number of vertices, N is the total number of nodes, is a rounding up operation.

[0151] According to the matched nodes, the agent model inference result is mapped to all vertices of the three-dimensional model, that is, all vertices of the three-dimensional model are assigned a calculated value. The specific value is defined using the following formula:

[0152]

[0153] In the formula, Vertex i is the calculated value assigned to the vertex of the three-dimensional model, Node j is the value of the corresponding node of the agent model inference result, n is the number of matched nodes, and i and j are the indices of the vertex and the node.

[0154] The calculated values assigned to all vertices are mapped to a color gradient, for example, low values are mapped to cool colors and high values are mapped to warm colors, and the model is recolored using a vertex shading method.

[0155] The digital twin running system is composed of monitoring sensors, edge computing devices, and cloud servers in the electro-hydraulic actuator system. The digital twin running system of the electro-hydraulic actuator is built as shown in Figure 6 .

[0156] In the edge computing node, the edge computing device has a CPU, a network port communication, a WLAN communication and the like to realize a communication function between the edge device and the AD conversion module, a monitoring signal processing and a proxy model inference function, and a communication function between the edge device and the cloud server. The cloud server and the edge computing device are connected through an MQTT protocol.

[0157] It should be noted that the edge-cloud collaborative electro-hydraulic actuator digital twin modeling method provided in the embodiments of the present application can be an edge-cloud collaborative electro-hydraulic actuator digital twin modeling device, or a control module in the edge-cloud collaborative electro-hydraulic actuator digital twin modeling device for executing the method of edge-cloud collaborative electro-hydraulic actuator digital twin modeling. In the embodiments of the present application, the edge-cloud collaborative electro-hydraulic actuator digital twin modeling device is taken as an example to illustrate the edge-cloud collaborative electro-hydraulic actuator digital twin modeling device provided in the embodiments of the present application.

[0158] In a second aspect of the embodiments of the present application, an edge-cloud collaborative electro-hydraulic actuator digital twin modeling system is provided, which includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, and the program or instructions implement the steps of the edge-cloud collaborative electro-hydraulic actuator digital twin modeling method according to any one of the first aspect embodiments when executed by the processor.

[0159] The edge-cloud collaborative electro-hydraulic actuator digital twin modeling system in the embodiments of the present application can be a device, or a component, an integrated circuit, or a chip in a terminal. The system can be a mobile electronic device, or a non-mobile electronic device. Illustratively, the mobile electronic device can be a mobile phone, a tablet computer, a notebook computer, a palm computer, a vehicle-mounted electronic device, a wearable device, an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA), etc., and the non-mobile electronic device can be a server, a network attached storage (NAS), a personal computer (PC), a television (TV), a teller machine, or a self-service machine, etc., and the embodiments of the present application are not limited specifically.

[0160] The edge-cloud collaborative electro-hydraulic actuator digital twin modeling system in the embodiments of the present application can be a device with an operating system. The operating system can be an Android operating system, an ios operating system, or other possible operating systems, and the embodiments of the present application are not limited specifically.

[0161] The edge cloud cooperative electro-hydraulic actuator digital twin modeling system provided by the embodiment of the application can realize Figure 1 The method embodiment realizes various processes, and details are not described herein to avoid repetition.

[0162] The embodiments of the application are described above in combination with the drawings, but the application is not limited to the specific embodiments described above, and the specific embodiments described above are only illustrative but not restrictive. Those skilled in the art can make many forms under the inspiration of the application without departing from the scope of the application and the protection scope of the claims.

Claims

1. An edge cloud collaborative electro-hydraulic actuator digital twin modeling method, characterized in that, The method comprises the following steps: constructing a high-fidelity simulation model of the target electro-hydraulic actuator by using simulation software, and generating simulation data by using the simulation model; the simulation data is used to construct a reduced-order model, and the low-dimensional output of the reduced-order model is used to train a proxy model; a proxy model of the operating characteristics of the target electro-hydraulic actuator is constructed by using the reduced-order model and deep learning, and is deployed on an edge device and a cloud server according to the requirement of computing power and the amount of input parameters; real-time acquisition of the operating state data of the target electro-hydraulic actuator; the operating state data is input into the proxy model to perform real-time model inference on the edge device and the cloud server, so as to predict the physical field operating characteristics of the target electro-hydraulic actuator under the current actual working condition; data fusion and visualization processing are performed on the physical field operating characteristics to construct a digital twin reflecting the operating state of the target electro-hydraulic actuator.

2. The edge cloud coordinated electro-hydraulic actuator digital twin modeling method of claim 1, wherein, The method of constructing a high-fidelity simulation model of the target electro-hydraulic actuator by using simulation software comprises the following steps: constructing a structural stress simulation model and an electromagnetic simulation model of the target electro-hydraulic actuator by using simulation software; constructing a fluid simulation model of the target electro-hydraulic actuator by using simulation software; constructing a system-level simulation model of the target electro-hydraulic actuator by using simulation software.

3. The edge cloud coordinated electro-hydraulic actuator digital twin modeling method of claim 1, wherein, The method of constructing a proxy model of the operating characteristics of the target electro-hydraulic actuator by using the reduced-order model and deep learning comprises the following steps: simulation result of the simulation working condition variable sample points is output by using the simulation model; the simulation result is used to construct a reduced-order model by using the intrinsic orthogonal decomposition method, so as to obtain the low-dimensional output corresponding to each sample and form a training data set; and the training data set is used to train a proxy model, so as to obtain a trained proxy model.

4. The edge cloud coordinated electro-hydraulic actuator digital twin modeling method of claim 3, wherein, The method of constructing a reduced-order model by using the simulation result by using the intrinsic orthogonal decomposition method, so as to obtain the low-dimensional output corresponding to each sample and form a training data set comprises the following steps: the simulation result is subjected to intrinsic orthogonal decomposition; singular values and corresponding singular vectors are selected to construct a reduced-order model; the simulation result is subjected to dimension reduction by using the reduced-order model to obtain low-dimensional output; the simulation result and the corresponding low-dimensional output constitute a training data pair to obtain a training data set.

5. The edge cloud coordinated electro-hydraulic actuator digital twin modeling method of claim 1, wherein, The method of real-time acquisition of the operating state data of the target electro-hydraulic actuator comprises the following steps: obtaining the original time step adopted during training of the proxy model; the operating state data of the target electro-hydraulic actuator is real-time acquired by using a physical sensor at the original time step.

6. The edge cloud coordinated electro-hydraulic actuator digital twin modeling method of claim 1, wherein, The method of inputting the operating state data into the proxy model to perform real-time model inference on the edge device and the cloud server comprises the following steps: electro-hydraulic actuator system performance model inference is performed by using the edge device; piston rod structure stress model inference is performed by using the edge device; force motor electromagnetic model inference is performed by using the edge device; seal leakage model inference is performed by using the cloud server; electromagnetic valve fluid model inference is performed by using the cloud server.

7. The edge cloud coordinated electro-hydraulic actuator digital twin modeling method of claim 1, wherein, The method of performing data fusion and visualization processing on the physical field operating characteristics to construct a digital twin reflecting the operating state of the target electro-hydraulic actuator comprises the following steps: loading a three-dimensional model file and a corresponding vertex information file; loading the inference result of the proxy model and the corresponding node information file; using a spatial matching algorithm to pair and combine the vertices in the three-dimensional model with the nodes in the inference result of the proxy model; mapping and normalizing the inference result of the proxy model to the model vertices to obtain a result value; mapping the result value to a color gradient and using a vertex coloring method to recolor the model.

8. An edge cloud coordinated electro-hydraulic actuator digital twin modeling system, comprising: comprise: a processor, a memory, and a program or instructions stored on the memory and executable on the processor, the program or instructions being executed by the processor to implement the steps of the edge cloud collaborative electro-hydraulic actuator digital twin modeling method according to any one of claims 1-7.

Images