Real-time analysis method for torsional vibration of rotating machine driven by cross-domain dynamic excitation and related device

Abstract

The invention discloses a real-time analysis method for torsional vibration of a rotating machine driven by cross-domain dynamic excitation and a related device, and relates to the technical field of monitoring of rotating mechanical performance, and the method comprises the following steps: firstly, constructing a universal node model of a rotating machine shaft system to obtain lumped parameters; meanwhile, real and virtual sampling points are arranged and spliced into virtual and real sampling grid nodes, so that the problems of poor generalization and incomplete sampling information of a traditional model are solved; acquiring actual working condition parameters of different sampling timestamps, and generating a training data set through a finite element method; training a torsional vibration analysis model comprising a multi-physics field distribution parallel learning module, a cross-domain characteristic driven torsional excitation resolving module and a dynamic directed graph-based aggregation output module, wherein the multi-physics field module captures the dynamic change of a temperature and pressure field under a variable working condition, and the torsional excitation resolving module extracts cross-domain coupling characteristics by means of an attention mechanism to improve the precision; the aggregation output module guarantees real-time performance; and finally, shaft system torsional vibration analysis is completed by using real-time working condition parameters and the trained model, and torque and relative torsion angle distribution is output. The method gives consideration to real-time performance and accuracy, and can provide support for dynamic control and predictive maintenance of the rotating machinery.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rotating machinery performance monitoring, in particular to a rotating machinery torsional vibration real-time analysis method driven by cross-domain dynamic excitation and related device. BACKGROUND

[0002] In complex rotating machinery such as gas turbines, aero-engines, and water turbines, torsion is the main motion form of the shaft system. Torsional vibration is caused by torque imbalance of the shaft system due to external excitation during mechanical operation, which can cause instantaneous speed fluctuations of elastic components, and further cause rotor cracks, breakage and other faults, seriously affecting the service life of the unit and causing significant losses. Therefore, real-time analysis of shaft system torsional angle fluctuation and torque distribution is the key to ensuring the safe and stable operation of the unit.

[0003] The existing torsional vibration analysis technology has obvious limitations: first, the method relying on mechanism model needs a large amount of prior knowledge and empirical formula in the field, and the modeling is complicated and has poor generalization, which is difficult to adapt to different rotating machinery systems; second, the simplified model based on idealized assumptions regards temperature, air pressure and other state parameters as constant values, ignoring the dynamic changes of physical state under variable working conditions, resulting in deviation of the analysis results from the actual situation; third, the existing deep learning methods (such as convolutional neural network and recurrent neural network) are difficult to extract local correlation features between components under variable working conditions, and are prone to over-smoothing problem, with low precision; and high-precision models (such as Transformer architecture combined with recurrent and time convolution networks) require a large amount of data and computing resources, with high training cost and poor real-time performance.

[0004] Therefore, there is an urgent need for a rotating machinery torsional vibration analysis method with real-time and accuracy to solve the deficiencies of the existing technology and provide support for dynamic control and predictive maintenance of equipment. SUMMARY

[0005] The purpose of the present application is to provide a rotating machinery torsional vibration real-time analysis method driven by cross-domain dynamic excitation and related device, which can accurately and in real time realize rotating machinery torsional vibration analysis.

[0006] To achieve the above-mentioned purpose, the present application provides the following solutions: In a first aspect, the present application provides a rotating machinery torsional vibration real-time analysis method driven by cross-domain dynamic excitation, comprising the following steps: A shaft system universal node model of the rotating machinery shaft system is constructed, the lumped parameters of the rotating machinery shaft system are obtained, and real sensor sampling points and virtual sampling points are arranged on the rotating machinery shaft system to splice a virtual-real sampling grid node.

[0007] Actual working condition parameters of the rotating machinery shaft system are acquired at different sampling time stamps, and a torsional vibration analysis training data set is generated in batches by a finite element method, and a torsional vibration analysis model driven by cross-domain dynamic excitation is trained; the torsional vibration analysis model comprises a multi-physical field distribution parallel learning module, a torsional excitation solving module driven by cross-domain features, and an aggregated output module based on a dynamic directed graph; the multi-physical field distribution parallel learning module is used to learn the temperature field and the air pressure field distribution of the rotating machinery shaft system as a whole under variable working conditions; the torsional excitation solving module is used to extract the implicit coupling features of the temperature field and the air pressure field respectively, and the implicit coupling features of the temperature field and the air pressure field are cross-domain fused based on an attention mechanism to obtain the torsional excitation; the aggregated output module is used to aggregate the torsional excitation to the rotating machinery shaft system to obtain the torsional vibration analysis result corresponding to the sampling time.

[0008] Actual working condition parameters of the rotating machinery shaft system are acquired, and the trained torsional vibration analysis model is used to perform real-time analysis of the shaft system torsional vibration according to the real-time working condition parameters to obtain a shaft system torsional vibration analysis result; the shaft system torsional vibration analysis result comprises the torque and the relative torsion angle distribution of the shaft system as a whole.

[0009] In a second aspect, the present application provides a rotating machinery torsional vibration real-time analysis system driven by cross-domain dynamic excitation, comprising the following functional modules: A virtual-real sampling grid construction unit is configured to construct a shaft system universal node model of the rotating machinery shaft system, acquire lumped parameters of the rotating machinery shaft system, and arrange real sensor sampling points and virtual sampling points on the rotating machinery shaft system to splice a virtual-real sampling grid node.

[0010] A data set generation and training unit is configured to acquire actual working condition parameters of the rotating machinery shaft system at different sampling time stamps, generate a torsional vibration analysis training data set in batches by a finite element method, and train a torsional vibration analysis model driven by cross-domain dynamic excitation; the torsional vibration analysis model comprises a multi-physical field distribution parallel learning module, a torsional excitation solving module driven by cross-domain features, and an aggregated output module based on a dynamic directed graph; the multi-physical field distribution parallel learning module is used to learn the temperature field and the air pressure field distribution of the rotating machinery shaft system as a whole under variable working conditions; the torsional excitation solving module is used to extract the implicit coupling features of the temperature field and the air pressure field respectively, and the implicit coupling features of the temperature field and the air pressure field are cross-domain fused based on an attention mechanism to obtain the torsional excitation; the aggregated output module is used to aggregate the torsional excitation to the rotating machinery shaft system to obtain the torsional vibration analysis result corresponding to the sampling time.

[0011] A shaft system torsional vibration real-time analysis unit is configured to acquire real-time working condition parameters of the rotating machinery shaft system, and use the trained torsional vibration analysis model to perform real-time analysis of the shaft system torsional vibration according to the real-time working condition parameters to obtain a shaft system torsional vibration analysis result; the shaft system torsional vibration analysis result comprises the torque and the relative torsion angle distribution of the shaft system as a whole.

[0012] In a third aspect, the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the real-time torsional vibration analysis method of rotating machinery driven by cross-domain dynamic excitation described above.

[0013] In a fourth aspect, the present application provides a computer-readable storage medium, having stored thereon a computer program, which, when executed by a processor, implements the steps of the real-time torsional vibration analysis method of rotating machinery driven by cross-domain dynamic excitation described above.

[0014] In a fifth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the real-time torsional vibration analysis method of rotating machinery driven by cross-domain dynamic excitation described above.

[0015] According to the specific embodiments provided by the present application, the following technical effects are disclosed: The present application provides a real-time torsional vibration analysis method of rotating machinery driven by cross-domain dynamic excitation and related devices, which does not need to rely on a large number of field prior knowledge and empirical formulas, and can adapt to rotating machinery shafting of different structures through standardized node segmentation and equivalent transformation, effectively solving the problems of complex modeling and poor generalization of traditional mechanism models. In addition, the virtual sampling points can supplement the sampling blind area caused by the structure and cost limitations of real sensors, and combined with real sampling points, the global coverage of shafting physical field information is realized, providing a comprehensive data basis for subsequent accurate analysis. In the cross-domain dynamic excitation torsional vibration analysis model, the multi-physical field distribution parallel learning module can capture the dynamic changes of temperature and air pressure in variable working conditions in real time, breaking the limitation of traditional simplified models that regard physical state parameters as constant values, avoiding analysis deviation caused by ignoring dynamic changes, and providing accurate multi-physical field basic data for subsequent excitation calculation. The torsional excitation calculation module can focus on key nodes with large temperature field and air pressure field differences through attention mechanism, and accurately mine local correlation characteristics between components. The aggregated output module controls the directional transmission of torsional vibration state and torsional excitation through dynamic directed graph, reduces information redundancy calculation, and greatly reduces the consumption of computing resources compared with high-precision models such as Transformer architecture, improves analysis efficiency while ensuring accuracy, and meets real-time requirements. Finally, the trained torsional vibration analysis model is used for real-time analysis of shafting torsional vibration, and the analysis result has real-time and accuracy, which can directly provide data support for dynamic control strategy optimization and predictive maintenance strategy formulation of rotating machinery, effectively solving the core problem that the existing technology cannot balance real-time and accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below only illustrate some of the embodiments of the present application, and for those skilled in the art, other drawings can be obtained from these drawings without creative labor.

[0017] Figure 1 A flow chart of a cross-domain dynamic excitation driven rotating machinery torsional vibration real-time analysis method provided by an embodiment of the present application.

[0018] Figure 2 A schematic diagram of virtual and real data sampling grid nodes in a cross-domain dynamic excitation driven rotating machinery torsional vibration real-time analysis method provided by an embodiment of the present application.

[0019] Figure 3 A structural schematic diagram of a cross-domain dynamic excitation driven torsional vibration analysis model in a cross-domain dynamic excitation driven rotating machinery torsional vibration real-time analysis method provided by an embodiment of the present application.

[0020] Figure 4 A cross-domain feature fusion conversion mechanism schematic diagram in a cross-domain dynamic excitation driven rotating machinery torsional vibration real-time analysis method provided by an embodiment of the present application.

[0021] Figure 5 A shafting dynamic directed graph snapshot structural schematic diagram in a cross-domain dynamic excitation driven rotating machinery torsional vibration real-time analysis method provided by an embodiment of the present application.

[0022] Figure 6 A torsional vibration analysis result and real sample value comparison schematic diagram in a cross-domain dynamic excitation driven rotating machinery torsional vibration real-time analysis method provided by an embodiment of the present application.

[0023] Figure 7 A functional module schematic diagram of a cross-domain dynamic excitation driven rotating machinery torsional vibration real-time analysis system provided by an embodiment of the present application.

[0024] Figure 8 A structural schematic diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0026] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] This application provides a real-time analysis method for torsional vibration of rotating machinery driven by cross-domain dynamic excitation. In one exemplary embodiment, such as... Figure 1 As shown, it includes the following steps: A1. Construct a generalized node model of the rotating machinery shaft system, obtain the lumped parameters of the rotating machinery shaft system, and arrange real sensor sampling points and virtual sampling points on the rotating machinery shaft system, splicing them together to obtain a virtual and real sampling mesh node. In this embodiment, the generalized node model of the shaft system specifically includes a shaft system geometric model, a shaft system excitation model, and a material damping model.

[0028] The shaft system geometric model is constructed by dividing the nodes according to the structural characteristics of the rotating machinery shaft system, transforming it into several shaft segment torsional vibration elements. The center of mass of the large inertial component is used as the concentrated mass point, and the lumped mass, lumped extreme moment of inertia, and torsional stiffness of the shaft segments are calculated. Specifically, the nodes are divided according to the structural characteristics of the shaft system, transforming it into N shaft segment torsional vibration elements. The center of mass of the large inertial component is selected as the concentrated mass point, and the shaft segment diameter and length are recorded. Parameters on the shaft system nodes are calculated, concentrating the mass and extreme moment of inertia of the shaft segments on both sides of the node onto the node. Finally, the stiffness of the shaft segment connecting two concentrated mass nodes is used as the stiffness of these two nodes, and the stiffness of the node is applied equally to the left and right shaft segments. The lumped mass, lumped extreme moment of inertia, and torsional stiffness of the shaft segments are shown in the following equations: .

[0029] .

[0030] .

[0031] in, L This refers to the length of the shaft segment; μ , j p These are the mass parameter per unit length and the polar moment of inertia parameter on the shaft segment, respectively. GI p Torsional stiffness of the shaft segment; M j , J pj , K αj They are nodes j Lumped mass, lumped pole moment of inertia and nodes j Torsional stiffness on the right-hand shaft segment; superscript () d () represents a node; formula subscriptj -1、 j These represent the left and right axis segments of node j, respectively.

[0032] Specifically, the excitation of the rotating shaft system caused by loads is obtained through motor torque, rated speed, etc., serving as a constant shaft excitation model. The torsional vibration damping caused by components such as bearings and couplings is determined based on the thermal diffusivity and other properties of the component materials and design parameters, serving as a material damping model.

[0033] Choosing a reference speed simplifies calculations and enhances versatility by representing a multi-speed geared shaft system as a single-speed system. The reference speed is typically the rotor speed at the drive end or the rotor speed at the load end, and the speed ratio is defined. i = ω 1 / ω 2, ω 1 is the reference speed (drive end or load end). ω Let 2 be the rotational speed at the other end. Then, at this reference rotational speed, the equivalent polar moment of inertia of the node is... J p1j Equivalent torsional stiffness of shaft segment K α1j They are respectively: .

[0034] Establish a polar coordinate system for the rotating axis system, with the axis as... x Direction, with respect to axis and radial direction r Direction, with circumferential direction as the axis θ Direction. Due to structural and cost constraints, the arrangement of sensors on the axis requires the establishment of virtual and real sampling grid nodes on the axis.

[0035] In this embodiment, the virtual and real sampling grid nodes include a spatial information encoding layer, a stitching layer, and a batch normalization layer. The virtual and real sampling grid nodes are constructed through the following steps: B1. Establish a polar coordinate system, arrange real temperature sensors, air pressure sensors and torque sensors as real sampling points, and encode the spatial position information of each real sampling point.

[0036] B2. Arrange virtual sampling points on the polar coordinate system and encode the spatial position information of each virtual sampling point.

[0037] B3. Perform splicing and batch normalization operations on the spatial information of each real and virtual sampling point to complete the construction of the virtual and real sampling grid nodes.

[0038] like Figure 2The shafting profile in the shafting profile, the real torque sensor sampling point is arranged at the front end of the shafting, and the torque, rotation speed and other data are measured, and the virtual torsional vibration data sampling point is arranged on the universal shafting node model. Real temperature sensor and real air pressure sensor are arranged to measure temperature and air pressure data, which should at least include inlet sampling point and outlet sampling point, such as Figure 2 The shafting inlet and outlet cross section in the shafting profile; considering the thermal effect of high temperature steam, according to the heat transfer mechanism, the grid is generated along the x Direction and r Direction distribution on the shafting model, and the virtual temperature data sampling point is arranged; considering the air pressure distribution caused by steam flow, according to the air pressure mechanism, the grid is generated along the x Direction and θ Direction distribution on the shafting model, and the virtual air pressure data sampling point is arranged.

[0039] The spatial coordinates of the real torque sampling point are set as According to the shafting structure, the virtual torsional vibration data sampling point is arranged N The spatial coordinates are set as .

[0040] The shafting inlet and outlet cross section are evenly arranged with m real temperature sampling points and n real air pressure sampling points, such as the real temperature sampling point and the real air pressure sampling point of the inlet cross section, whose spatial coordinates are set as: .

[0041] .

[0042] Wherein, r 1、 r m The r Directional coordinates of the shaft center and the real temperature sampling point of the outer surface, θ 1、 θ n The n Directional coordinates of the first phase and the θ Phase real air pressure sampling point of the outer surface.

[0043] The x Directional coordinates of the torsional vibration sampling point are scaled by PCHIP linear interpolation to be uniformly arranged in the axial direction, which are used as the x Directional coordinates of the virtual temperature and air pressure sampling points on the shafting, and the spatial coordinates of the temperature virtual sampling point and the air pressure virtual sampling point are set as: .

[0044] .

[0045] Wherein, indicated in x directional coordinates of the axis of the x j cross-section r directional coordinates of the outer surface of the x indicated in x j cross-section r directional coordinates of the outer surface of the r m When the subscript is 2, 3, …, -1, it indicates the distance from the axis from near to far.

[0046] The virtual sampling points of temperature and pressure are spliced with the real sampling points of temperature and pressure in spatial position information to obtain the temperature and pressure virtual-real sampling grid: .

[0047] .

[0048] wherein, is the temperature virtual-real sampling grid, is the pressure virtual-real sampling grid; the superscript T represents transposition.

[0049] The spatial position information of the virtual-real sampling grid nodes obtained is normalized and data-specified: .

[0050] wherein, x max and x min are the maximum value and the minimum value of the x directional coordinates, x j is the coordinate of the given j th sampling point in the x direction, is the coordinate value after normalization, ensuring that the value range is between [0, 1]. r The θ directional coordinates are normalized in the same way.

[0051] ​​A2, actual working condition parameters of the rotating machinery shafting at different sampling timestamps are acquired, and a torsional vibration analysis training data set is batch generated through a finite element method, and a torsional vibration analysis model driven by a cross-domain dynamic excitation is trained; the torsional vibration analysis model comprises a multi-physical field distribution parallel learning module, a torsional excitation solving module driven by cross-domain features, and an aggregated output module based on a dynamic directed graph; the multi-physical field distribution parallel learning module is used to learn the global temperature field and pressure field distribution of the rotating machinery shafting under variable working conditions; the torsional excitation solving module is used to extract the implicit coupling features of the temperature field and the pressure field respectively, and to perform cross-domain fusion on the implicit coupling features of the temperature field and the pressure field based on an attention mechanism to obtain a torsional excitation; and the aggregated output module is used to aggregate the torsional excitation to the rotating machinery shafting to obtain a torsional vibration analysis result corresponding to a sampling time.

[0052] As shown in Figure 3 , the multi-physical field distribution parallel learning module comprises a first extreme learning machine (ELM) neural network and a second extreme learning machine (ELM) neural network, which are respectively used to learn the global distribution of the temperature field and the pressure field of the rotating machinery shafting under variable working conditions. The first extreme learning machine neural network obtains a temperature field loss function Ls T through training, and the second extreme learning machine neural network obtains a pressure field loss function Ls P through training.

[0053] The rotating machinery shafting is regarded as a uniform, isotropic and heat source-free model, the second-order spatial derivative of the temperature around each sampling point is calculated to describe the distribution and propagation of heat in the object. The temperature transfer equation is: .

[0054] wherein, T is the temperature, t is the time, λ , C and ρ 1 are the thermal conductivity, the specific heat capacity and the density of the component material, x , r are the x direction and the r direction of the shafting.

[0055] The temperature field loss function is as follows: Ls T = Res PDE1 + Err Data1 .

[0056] wherein, LsT Let the temperature field loss function be... Res PDE1 This is the residual term of the first mechanism equation. Err Data1 This is the mean absolute error term for the first data.

[0057] The residual term of the first mechanism equation is calculated using the following formula: .

[0058] in, The number of sampling points in the computational domain of the temperature field residual. This represents the number of sampling times for the temperature sensor. j The labels are the sampling points in the computational domain of the temperature field residual. i This is a label for the sampling time of the temperature sensor. For the temperature transfer equation residuals, t i For the first i The time of each sampling moment x j and r j The first j The axial and radial coordinates of each sampling point.

[0059] The mean absolute error of the first data is calculated using the following formula: .

[0060] in, The number of sampling points for temperature observations. The output of the first extreme learning machine neural network is t i Sampling time at ( x j , r j Temperature at location In order to be in t i Sampling time at ( x j , r j Temperature observations at the location.

[0061] The residuals of the temperature transfer equation are calculated using the following formula: .

[0062] The pressure field on the surface of rotating machinery is mainly generated by the flow of steam. Treating the steam as an ideal gas and neglecting viscosity, the continuity equation describing the pressure field distribution in polar coordinates is: .

[0063] Momentum equation along the x-axis and along θ The circumferential momentum equations are as follows: .

[0064] .

[0065] in, P For air pressure, t For a moment, ρ 2 represents the steam density. x , r , θ They are respectively the shaft system x direction, r direction and θ Direction. Assume axial radial velocity. And axial circumferential velocity Defined by the linear velocity of rotation, w ω is the rotational angular velocity.

[0066] The pressure field loss function is shown in the following equation: Ls P = Res PDE2 + Err Data2 .

[0067] in, Ls P The pressure field loss function, Res PDE2 This is the residual term of the second mechanism equation. Err Data2 This is the mean absolute error term for the second data.

[0068] The residual term of the second mechanism equation is calculated using the following formula: .

[0069] in, The number of sampling points in the computational domain of the pressure field residual. This represents the number of sampling times for the barometric pressure sensor. j The labels are the sampling points in the computational domain of the temperature field residual. i This is a label for the sampling time of the temperature sensor. For the residuals of the steam flow equation, θ j For the first j The circumferential coordinates of each sampling point.

[0070] The mean absolute error term for the second data is calculated using the following formula: .

[0071] wherein, is the number of sampling points of the pressure observation value, is the pressure distribution output by the second extreme learning machine neural network model, is the pressure observation value at the sampling time point t i on the position x j , θ j .

[0072] The steam flow equation residual is calculated by the following formula: .

[0073] wherein, is the gas density distribution, which is calculated by , taking the ideal gas density , and the denominator T here is the temperature of the outer surface of the shaft system, v θ is the shaft circumferential velocity, v x is the axial velocity, θ is the shaft circumferential direction of the shaft system, w is the angular velocity of rotation, is the radius of the shaft section corresponding to the position x j , ρ 2 is the steam density.

[0074] In training the extreme learning machine neural network, the first extreme learning machine neural network and the second extreme learning machine neural network are trained in parallel based on the least square method, and the rejection sampling multi-round learning algorithm and the early stopping mechanism are adopted to regulate the training convergence of the first extreme learning machine neural network and the second extreme learning machine neural network.

[0075] The input weights and the bias of the ELM single-hidden layer neural network conform to the Gaussian distribution, with a mean of 0 and a variance of 1. In order to make the temperature field loss function and the pressure field loss function Ls T minimum, the output weight vectors Ls P and are trained respectively, and the extreme learning machine equation based on the least square method is as follows: .

[0076] . ​

[0077] where, , is the representation of the input vector on hidden layer neurons, is the representation of the mechanism equation term on hidden layer neurons, is the representation of the data term on hidden layer neurons; is the weight parameter connecting hidden layer neurons and output, Nh is the number of hidden layer neurons; , is the target value, is the target value of the mechanism equation term, is the target value of the data term; superscript T represents transpose. When the input weight , bias and matrix have been determined, the output weight vector is obtained using pseudo-inverse operation.

[0078] The training conditions are set, and the rejection sampling multi-round learning algorithm and early stopping mechanism are defined. Among them, the initial number of mechanism equation training sampling points is defined as , the number of data term training sampling points is ; the number of mechanism equation sampling points in each round is , the number of data term sampling points is ; the number of mechanism equation sampling points meeting the sampling point acceptance condition in each round is , the number of data term sampling points is , the initial moment ; the maximum number of allowed training points is , the threshold of training error is ; the total number of training sampling points is ; the sampling point acceptance condition of the rejection sampling method is defined as: .

[0079] where, α is the acceptance range, is the error on the sampling point j , and maxLs is the maximum value of the loss function obtained in each round of sampling points. Only the sampling points with errors greater than a certain range can be accepted to generate higher quality training data.

[0080] The total number of training sampling points in each round will increase with the rejection sampling, which is defined as , and the early stopping condition is set as the total number of training sampling points or the loss function . If one of the conditions is met, the training is ended, reducing the consumption of computer resources.

[0081] In an exemplary embodiment, the cross-domain feature-driven torsional excitation solution module employs an attention focusing mechanism and a cross-domain feature fusion and transformation mechanism.

[0082] The attention focusing mechanism employs temperature distribution attention focusing function and air pressure distribution attention focusing function.

[0083] The temperature distribution attention focusing function includes an axial global attention head, a radial local attention head, a normalized activation function, and a weighted output function, focusing on node locations with large temperature field differences. Specifically, the global attention head focuses on the temperature difference changes between different axial segments, while the local attention head focuses on the decay of the radial surface temperature abruptly changing to the axial steady state. The temperature distribution attention focusing function satisfies the following definition: .

[0084] .

[0085] .

[0086] in, For output Location-based temperature attention weight W T This is an adjustable weighting parameter with a value range of (0,1); For the first i Axial global attention weights for each axial sampling point For the first i The axial sampling point and its 1st axial sampling point m The temperature difference between adjacent nodes For the first i The axial sampling point and its 1st axial sampling point m The axial distance between adjacent nodes; For the first j Radial local attention weights for each radial sampling point For the first j The radial sampling point and its i.e., the ... m Temperature difference between adjacent nodes For the first j The radial sampling point and its i.e., the ... m The radial distance between adjacent nodes; This is the normalized activation function.

[0087] The pressure distribution attention focusing function includes an axial global attention head, a circumferential local attention head, a normalized activation function, and a weighted output function, focusing on node locations with large pressure field differences. Specifically, the global attention head focuses on the pressure difference changes along the axial gas flow direction, while the local attention head focuses on the relationship between the rotor speed and circumferential pressure changes. The pressure distribution attention focusing function satisfies the following definition: .

[0088] .

[0089] .

[0090] in, For the output [ i , j [Location, air pressure, attention focus] W P This is an adjustable weighting parameter with a value range of (0,1); For the first i Axial global attention weights for each axial sampling point For the first i The axial sampling point and its 1st axial sampling point m The air pressure difference between adjacent nodes For the first i The axial sampling point and its 1st axial sampling point m The axial distance between adjacent nodes; For the first j Circumferential local attention weights for each circumferential sampling point For the first j The circumferential sampling point and its first m The air pressure difference between adjacent nodes; For the first j The circumferential sampling point and its first m Phase difference between adjacent nodes.

[0091] for and One-hot encoding is performed on each item, setting only the coordinates to 1 and all other items to 0, and then flattening the result into a one-dimensional column vector. This yields a column vector containing position information and corresponding temperature attention weights. and a column vector containing location information and corresponding barometric pressure attention weights. As shown in the following formula: .

[0092] .

[0093] in, represents a one-hot type position encoding, .

[0094] The cross-domain feature fusion conversion mechanism includes: improving the temperature field and the air pressure field feature diversity and merging the feature information through a multi-branch convolution layer. The correlation between the temperature field and the air pressure field is calculated through scaled dot-product attention, and the coupled temperature field information in the air pressure field and the coupled air pressure field information in the temperature field are extracted. The cross-domain fusion information is focused on the shaft system node position according to the attention weight vector. Through temperature-torque conversion and air pressure-torque conversion, the superimposed torsional excitation on the shaft system is obtained.

[0095] As shown in Figure 4 , the temperature field distribution data U T and the air pressure field distribution data U P are normalized, and then multi-branch convolution is used to enhance the feature extraction capability. 3*3 convolution kernel, 5*5 hollow convolution kernel and 1*1 convolution kernel are used in parallel to obtain diverse features, and then Filter Concatenation operation is used to merge the feature information of the temperature field and the air pressure field by channel. Through reshaping and linear transformation of the temperature field feature data F T , the query Q T , the key K T and the value V T of the temperature field are obtained; the air pressure field feature data F P is reshaped and linearly transformed to obtain the query Q P , the key K P and the value V P .

[0096] The coupling relationship between the temperature field and the air pressure field is quantified and cross-domain fusion is performed. When the coupled air pressure field information in the temperature field is extracted, the correlation between Q P and K T is calculated through scaled dot-product attention, and the Softmax activation function is applied to obtain the attention score matrix reflecting the correlation between Q P and K T , and the attention score weighted aggregation V T is obtained, and then 1*1 convolution is performed to obtain the information transmitted from the temperature field to the pressure field​ The above process is expressed as follows: .

[0097] .

[0098] wherein, is a scaling factor, is K T dimension; the superscript T represents transposition.

[0099] Similarly, the temperature field information coupled in the gas pressure field is extracted, and the process is expressed as follows: .

[0100] .

[0101] The temperature field feature data F T and the information transmitted from the pressure field to the temperature field are output together through the fusion unit, and the output result is , as shown in the following formula: .

[0102] .

[0103] wherein, H T is the fusion weight of the temperature field; represents multiplication of the corresponding position elements of the matrix; is a Sigmoid activation function, which ensures that the weight range is (0, 1); represents a 1x1 convolution operation, which performs channel dimension reduction.

[0104] Similarly, the output result of the gas pressure field fusion unit is : .

[0105] .

[0106] The temperature field and the gas pressure field position information and the corresponding attention weight are embedded when solving the torsional excitation. The temperature field fusion result is output as an axial node distribution by vector weighting, and the gas pressure field fusion result is output as an axial node distribution by vector weighting, and then respectively passes through a 1x5 max pooling layer for pooling, and finally respectively passes through a fully connected layer to be mapped to the temperature field additional torque on the axial node and the additional torque of the air pressure field And perform one-hot position encoding.

[0107] The final result is the cross-domain fused torsional excitation superimposed on the shaft system node model, represented as a torque vector. ,Right now Figure 3 Torsional excitation output in the middle: .

[0108] in, pos T For additional torque Location encoding, pos P For additional torque Location encoding.

[0109] Furthermore, in this embodiment, the aggregation output module aggregates the torsional vibration state and torsional excitation onto the shaft system nodes through the adjacency matrix and jump connection control edge information transmission; such as Figure 5 A snapshot of the axis-directed graph at one of the sampling times. The following is an example: ， ， .

[0110] in, V For the set of graph nodes, For the first i The feature vector of a node includes torsional vibration state parameters lumped onto the node, a vector indicating the node type, and spatial location. E G It is a set of edges of a graph, divided into edges that transmit torsional vibration states. and incentive transmission edge Two categories; The adjacency matrix is ​​composed of two sub-adjacency matrices: the torsional vibration state transfer matrix and the excitation transfer matrix, and is defined as follows: .

[0111] by Figure 5 For example, the torsional excitation information is concentrated on the first node. By controlling the information transmission on the edges through the adjacency matrix and jump connections, the torsional vibration state and torsional excitation are aggregated onto the shaft system nodes. The overall shaft system torsional vibration analysis result after aggregation is expressed as follows: .

[0112] in, For the first j Torsional vibration state at the node; τ0 is a real torque value on the first node, α 0 is a real relative torsion angle value on the first node; 0 is an excitation torque value on the first i node; 0 is a torsional vibration state information transfer matrix from the first node to the next node, 0 is a torsional excitation information superposition matrix from the first i node to the first j node, represents matrix multiplication; N 0 is the total number of shafting nodes; Norm {} represents data normalization output, which maps the output value to the [-1, 1] interval.

[0113] The transfer matrix of the torsional vibration state information is calculated by the following formula: .

[0114] The superposition matrix of the torsional excitation information is calculated by the following formula: .

[0115] wherein, 0 is the concentrated rotational inertia on the node, c j 0 is the torsional damping coefficient, s t 0 is the complex frequency of real-time shafting rotation, K j 0 is the torsional stiffness of the shaft section, 0 is the applied stiffness coefficient.

[0116] The working condition parameters of the rotating machinery shafting are collected at different sampling timestamps, including input torque (N·mm), shafting speed (rpm), inlet temperature (℃), outlet temperature (℃), inlet air pressure (kPa), outlet air pressure (kPa), etc. The input working condition parameter set is used to train the target model. When training, the first / second extreme learning machine (ELM) neural network is trained first, and then the parameters of each subsequent module are trained. The torsional vibration analysis model obtained by training can be used to determine the temperature field and air pressure field distribution of the shafting as a whole, and the superimposed torsional excitation of the shafting, and finally aggregate the output as the torque (N·mm) and relative torsion angle (°) distribution of the shafting as a whole.

[0117] In one exemplary embodiment, 2000 sets of sample data are generated in batches through finite element simulation calculations, combining the geometric parameters, material properties, and boundary conditions of the rotating machinery shaft system. Tetrahedral meshes are used, and each set of samples includes the shaft system's operating parameters, temperature and pressure field distributions, and torsional vibration parameters. The 2000 sets of sample data are divided into training, validation, and test sets in an 8:1:1 ratio, and the network weight parameters are trained using a loss function.

[0118] A3. Obtain the real-time operating parameters of the rotating machinery shaft system, and use the trained torsional vibration analysis model to perform real-time analysis of the shaft system torsional vibration based on the real-time operating parameters to obtain the shaft system torsional vibration analysis results; the shaft system torsional vibration analysis results include the distribution of the overall torque and relative torsional angle of the shaft system.

[0119] In one exemplary embodiment, a trained torsional vibration analysis model is used to perform real-time analysis of shaft torsional vibration and compare it with actual torsional vibration sample values, such as... Figure 6 As shown, 150 sampling points are evenly distributed on the curve. At sampling time t1, the average absolute deviation of the calculated relative torsion angle is 0.0014, and the average absolute percentage deviation is 0.55%; the average absolute deviation of the calculated torque at sampling time t1 is 0.0012, and the average absolute percentage deviation is 0.34%. At sampling time t2, the average absolute deviation of the calculated relative torsion angle is 0.0031, and the average absolute percentage deviation is 0.89%; the average absolute deviation of the calculated torque at sampling time t2 is 0.0037, and the average absolute percentage deviation is 1.01%.

[0120] According to the specific embodiments provided in this application, the solution proposed in this application has the following technical effects: 1) By utilizing the real-time global distribution of temperature and pressure fields of rotating machinery shafts under varying working conditions, a multi-head attention mechanism is adopted to extract the axial long-range dependence of temperature and pressure field distributions and the local correlation features between components. Then, cross-domain feature fusion and transformation are performed to obtain the superimposed torsional excitation on the shaft system. This can extract the coupling features of multi-physics fields at a deeper level, thereby improving the calculation accuracy of the torsional vibration analysis model.

[0121] 2) By using the aggregation output module based on dynamic directed graphs, the torsional vibration state and the directional transmission of torsional excitation on the shaft system are controlled, and the torsional vibration analysis results are aggregated and output. This can more comprehensively characterize the influence of cross-domain dynamic torsional excitation on the torsional vibration of rotating machinery, while greatly improving the efficiency of torsional vibration analysis and saving computational resources.

[0122] Based on the same inventive concept, this application also provides a system for implementing the above-described method for real-time analysis of torsional vibration in rotating machinery driven by cross-domain dynamic excitation. The solution provided by this system is similar to the solution described in the above method. In an exemplary embodiment, such as...Figure 7 As shown, a cross-domain dynamic excitation driven rotating machinery torsional vibration real-time analysis system is provided, comprising the following functional modules: A virtual-real sampling grid construction unit is configured to construct a shafting universal node model of a rotating machinery shafting, obtain lumped parameters of the rotating machinery shafting, and arrange real sensor sampling points and virtual sampling points on the rotating machinery shafting, and splice to obtain a virtual-real sampling grid node.

[0123] A data set generation and training unit is configured to obtain actual working condition parameters of the rotating machinery shafting at different sampling timestamps, and generate torsional vibration analysis training data sets in batches through a finite element method, and train a cross-domain dynamic excitation driven torsional vibration analysis model; the torsional vibration analysis model comprises a multi-physical field distribution parallel learning module, a cross-domain feature driven torsional excitation solving module, and an aggregated output module based on a dynamic directed graph; the multi-physical field distribution parallel learning module is configured to learn temperature field and pressure field distributions of the rotating machinery shafting globally under variable working conditions; the torsional excitation solving module is configured to extract implicit coupling features of the temperature field and the pressure field respectively, and perform cross-domain fusion of the implicit coupling features of the temperature field and the pressure field based on an attention mechanism to obtain a torsional excitation; and the aggregated output module is configured to aggregate the torsional excitation to the rotating machinery shafting to obtain a torsional vibration analysis result corresponding to a sampling time.

[0124] A shafting torsional vibration real-time analysis unit is configured to obtain real-time working condition parameters of the rotating machinery shafting, and perform real-time analysis of shafting torsional vibration according to the real-time working condition parameters by using the trained torsional vibration analysis model to obtain a shafting torsional vibration analysis result; the shafting torsional vibration analysis result comprises a distribution of a torsional moment and a relative torsional angle of the shafting as a whole.

[0125] Of course, Figure 7 The architecture shown is only exemplary, and when different functions are implemented, some of the components shown in the system can be omitted Figure 7 according to actual needs.

[0126] In an exemplary embodiment, a computer device is provided, which can be a server or a terminal, and an internal structure diagram thereof can be as shown in Figure 8As shown in the figure. The computer device includes a processor, a memory, an Input / Output (I / O) interface, and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor, which can realize the above-mentioned embodiment of a cross-domain dynamic incentive driven rotating machinery torsional vibration real-time analysis method.

[0127] Those skilled in the art can understand that, Figure 8 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0128] In an exemplary embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to realize the steps in the above method embodiments.

[0129] In an exemplary embodiment, a computer readable storage medium is provided, storing a computer program, which is executed by a processor to realize the steps in the above method embodiments.

[0130] In an exemplary embodiment, a computer program product is provided, including a computer program, which is executed by a processor to realize the steps in the above method embodiments.

[0131] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0132] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to a memory, a database or other medium used in the embodiments provided in the present application can include at least one of a non-volatile and a volatile memory. The non-volatile memory can include a read-only memory (ROM), a magnetic tape, a floppy disk, a flash memory, an optical storage, a high-density embedded non-volatile memory, a resistive random access memory (ReRAM), a magnetoresistive random access memory (MRAM), a ferroelectric random access memory (FRAM), a phase change memory (PCM), a graphene memory, etc. The volatile memory can include a random access memory (RAM) or an external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as a static random access memory (SRAM) or a dynamic random access memory (DRAM), etc.

[0133] The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0134] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0135] The principles and implementation modes of the present application are described by applying specific examples in the present application. The above embodiments are only used to help understand the method and its core idea of the present application; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range can be changed. In conclusion, the content of the present application should not be understood as a limitation.

Claims

1. A real-time analysis method for torsional vibration of rotating machinery driven by cross-domain dynamic excitation, characterized in that, The application relates to a method for realizing real-time torsional vibration analysis of a rotating machinery shaft system. The method comprises the following steps: constructing a shaft system universal node model of the rotating machinery shaft system, obtaining lumped parameters of the rotating machinery shaft system, arranging real sensor sampling points and virtual sampling points on the rotating machinery shaft system, and splicing to obtain virtual-real sampling grid nodes; Actual working condition parameters of the rotating machinery shaft system are obtained at different sampling time stamps, and torsional vibration analysis training data sets are batch-generated through a finite element method, and a torsional vibration analysis model driven by cross-domain dynamic excitation is trained; the torsional vibration analysis model comprises a multi-physical field distribution parallel learning module, a torsional excitation solving module driven by cross-domain features, and an aggregated output module based on a dynamic directed graph; the multi-physical field distribution parallel learning module is used for learning temperature field and air pressure field distributions of the rotating machinery shaft system under variable working conditions; the torsional excitation solving module is used for extracting implicit coupling features of the temperature field and the air pressure field respectively, and performing cross-domain fusion on the implicit coupling features of the temperature field and the air pressure field based on an attention mechanism to obtain torsional excitation; and the aggregated output module is used for aggregating the torsional excitation to the rotating machinery shaft system to obtain torsional vibration analysis results corresponding to sampling time. Real-time working condition parameters of the rotating machinery shaft system are obtained, and the trained torsional vibration analysis model is used to perform real-time analysis on shaft system torsional vibration according to the real-time working condition parameters to obtain shaft system torsional vibration analysis results; the shaft system torsional vibration analysis results comprise distributions of torsional moment and relative torsional angle of the shaft system as a whole.

2. The cross-coupled dynamic excitation driven torsional vibration real-time analysis method of claim 1, wherein, The shaft system universal node model specifically comprises a shaft system geometric model, a shaft system excitation model and a material damping model; the shaft system geometric model is segmented according to structural features of the rotating machinery shaft system, equivalent transformation is performed into a plurality of shaft segment torsional vibration units, a mass center of a large inertia component is taken as a concentrated mass point, and node lumped mass, lumped polar moment of inertia and shaft segment torsional stiffness are calculated; the shaft system excitation model determines a constant excitation caused by a load according to motor torque and rated speed; and the material damping model determines torsional vibration damping caused by bearings and couplings according to material thermal diffusivity and design parameters of components.

3. The method of claim 1, wherein the method is a real-time analysis method of torsional vibration of a rotating machine driven by a dynamic excitation across domains, characterized by, The virtual-real sampling grid nodes comprise a space information coding layer, a splicing layer and a batch normalization layer, and the virtual-real sampling grid nodes are constructed through the following steps: A shaft system polar coordinate system is established, real temperature sensors, air pressure sensors and torsional moment sensors are arranged as real sampling points, and space position information of the real sampling points is coded; Virtual sampling points are arranged on the shaft system polar coordinate system, and space position information of the virtual sampling points is coded; Space information of the real sampling points and the virtual sampling points is spliced and batch-normalized to complete construction of the virtual-real sampling grid nodes.

4. The cross-coupled dynamic excitation driven torsional vibration real-time analysis method of claim 1, wherein, The multi-physics field distribution parallel learning module comprises a first extreme learning machine neural network and a second extreme learning machine neural network; the first extreme learning machine neural network is obtained by training a temperature field loss function Ls T The second extreme learning machine neural network is obtained by training a gas pressure field loss function Ls P The first extreme learning machine neural network and the second extreme learning machine neural network are trained in parallel based on a least square method, and a rejection sampling multi-round learning algorithm and an early stop mechanism are adopted to regulate network training convergence. A temperature field loss function is shown in the following formula: Ls T = Res PDE1 + Err Data1 ； wherein, Ls T is a temperature field loss function, Res PDE1 is a first mechanism equation residual term, Err Data1 is a first data mean absolute error term; A first mechanism equation residual term is calculated through the following formula: ； wherein, is the number of sampling points of the temperature field residual calculation domain, is the number of sampling instants of the temperature sensor, j is the index of the sampling point of the temperature field residual calculation domain, i is the index of the sampling instant of the temperature sensor, is the temperature transfer equation residual, t i is the time of the i sampling instant, x j and r j are the axial coordinate and the axial radial coordinate, respectively, of the j sampling point. A first data average absolute error term is calculated through the following formula: ； wherein, is the number of sampling points of temperature observations, is the output of the first extreme learning machine neural network at t i temperature at the sampling time at x j , r j position, is the output of the second extreme learning machine neural network at t i temperature observation at the sampling time at x j , r j position. A temperature transfer equation residual is calculated through the following formula: ； wherein, Lambda , C and Rho 1are the thermal conductivity, the specific heat capacity and the density of the component material, respectively; An air pressure field loss function is shown in the following formula: Ls P = Res PDE2 + Err Data2 ； wherein, Ls P is a barometric field loss function, Res PDE2 is a second mechanism equation residual term, Err Data2 is a second data mean absolute error term; A second mechanism equation residual term is calculated through the following formula: ； wherein, is the number of sampling points of the pressure field residual calculation domain, is the number of sampling instants of the pressure sensor, j is the index of the sampling point of the temperature field residual calculation domain, i is the index of the sampling instant of the temperature sensor, is the residual of the steam flow equation, Theta j is the axial coordinate of the j th sampling point; A second data average absolute error term is calculated through the following formula: ； wherein, is the number of sampling points of the pressure observation value, is the pressure distribution output by the second extreme learning machine neural network model, is the pressure observation value at the sampling time point in the position of t i , x j , Theta j the position A steam flow equation residual is calculated through the following formula: ； wherein is the gas density distribution, v θ is the shaft circumferential velocity, v x is the axial velocity, Theta is the shaft circumferential direction of the shaft system, w is the angular velocity of rotation, is the corresponding x j is the radius of the shaft section at the position, Rho 2 is the vapor density.

5. The cross-coupled dynamic excitation driven torsional vibration real-time analysis method of claim 1, wherein, The cross-domain feature driven torsional excitation solving module adopts an attention focusing mechanism and a cross-domain feature fusion conversion mechanism; The temperature distribution attention focusing function and the air pressure distribution attention focusing function are adopted in the attention focusing mechanism; the temperature distribution attention focusing function includes an axial global attention head, a radial local attention head, a normalization activation function and a weighted output function, and focuses on the node position with a temperature field difference value greater than a threshold value; the air pressure distribution attention focusing function includes an axial global attention head, a circumferential local attention head, a normalization activation function and a weighted output function, and focuses on the node position with an air pressure field difference value greater than a threshold value; The cross-domain feature fusion conversion mechanism includes: The temperature field and the air pressure field feature diversity are improved and the feature information is combined through a multi-branch convolution layer; The correlation is calculated through a scaled dot-product attention, and the coupled temperature field information in the air pressure field and the coupled air pressure field information in the temperature field are extracted; The cross-domain fusion information is focused on the shafting node position according to the attention weight vector; The superimposed torsional excitation on the shafting is obtained through temperature-torque conversion and air pressure-torque conversion.

6. The cross-coupled dynamic excitation driven torsional vibration real-time analysis method of claim 1, wherein, The torsional vibration state and the torsional excitation are aggregated to the shafting node through the adjacency matrix and the jump connection to control the information transmission on the edge in the aggregation output module; and the overall shafting torsional vibration analysis result after aggregation is represented as: ； wherein, is the first j torsional state at the node; τ 0 is the true torque value at the first node, α 0 is the true relative torsion angle value at the first node; is the first i exciting torque value at the node; is the first transfer matrix of the torsional state information at the node to the next node, is the first i superposition matrix of the torsional excitation information at the node to the first j node, denotes matrix multiplication; N is the total number of shafting nodes; Norm {} denotes data normalization output, mapping the output value to the [-1, 1] interval; The transfer matrix is calculated by the following formula: ； The superimposition matrix is calculated by the following formula: ； wherein, is the lumped moment of inertia on the node, c j is the torsional damping coefficient, s t is the complex frequency of the real-time shafting rotation, K j is the torsional stiffness of the shaft section, is the added stiffness coefficient.

7. A cross-domain dynamic excitation driven rotating machinery torsional vibration real-time analysis system, characterized in that, It includes: A virtual-real sampling grid construction unit is configured to construct a shafting universal node model of a rotating machinery shafting, obtain lumped parameters of the rotating machinery shafting, and arrange real sensor sampling points and virtual sampling points on the rotating machinery shafting to splice a virtual-real sampling grid node. A data set generation and training unit is configured to obtain actual working condition parameters of the rotating machinery shafting at different sampling time stamps, and generate torsional vibration analysis training data sets in batches through a finite element method to train a cross-domain dynamic excitation driven torsional vibration analysis model; the torsional vibration analysis model includes a multi-physical field distribution parallel learning module, a cross-domain feature driven torsional excitation solving module and an aggregation output module based on a dynamic directed graph; the multi-physical field distribution parallel learning module is configured to learn temperature field and air pressure field distributions of the rotating machinery shafting under variable working conditions; the torsional excitation solving module is configured to extract implicit coupling features of the temperature field and the air pressure field respectively, and perform cross-domain fusion of the implicit coupling features of the temperature field and the air pressure field based on an attention mechanism to obtain a torsional excitation; and the aggregation output module is configured to aggregate the torsional excitation to the rotating machinery shafting to obtain a torsional vibration analysis result corresponding to a sampling time. A shafting torsional vibration real-time analysis unit is configured to obtain real-time working condition parameters of the rotating machinery shafting, and perform real-time analysis of shafting torsional vibration according to the real-time working condition parameters by using the trained torsional vibration analysis model to obtain a shafting torsional vibration analysis result; the shafting torsional vibration analysis result includes a torque and a relative torsional angle distribution of the shafting as a whole.

8. A computer device comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the cross-domain dynamic excitation driven real-time torsional vibration analysis method of any one of claims 1-6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the cross-domain dynamic excitation driven real-time torsional vibration analysis method of any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the cross-domain dynamic excitation driven real-time torsional vibration analysis method of any one of claims 1-6.

Images