Transient calculation method and device for liquid-solid coupling of liquid-filled rotor of centrifugal machine

Abstract

The invention discloses a liquid-solid coupling transient calculation method and device for a liquid-filled rotor of a centrifugal machine. The method comprises the implementation steps that (1) dynamic characteristics of fluid in a rotary drum of the centrifugal machine under multiple working conditions, namely different liquid filling amounts and different rotating speeds are obtained; 2) inputting the characteristic point displacement, the liquid filling amount and the rotation speed obtained by simulation into a neural network for training, and constructing a neural network finite element dynamic model; and 3) through a neural network and finite element coupling method, carrying out finite element modeling on a linear part, describing a nonlinear part by using a neural network, coupling a finite element model part and a neural network part according to a boundary condition and a connection condition, and carrying out linear-nonlinear transient dynamic solution analysis on the coupled dynamic model. The method can solve the problems that the calculation speed is slow and time is consumed due to the fact that the freedom degree of the model is too high in a traditional finite element modeling method, and the non-linear dynamic characteristic representation of the liquid-carrying rotor system model is inaccurate.

Description

Technical Field

[0001] The invention relates to the field of finite element simulation of rotor dynamics computer-aided calculation, and specifically designs a method and device for liquid-solid coupling transient calculation of a centrifuge liquid-filled rotor based on a neural network-finite element collaborative model. Background Art

[0002] Centrifugal separation machines such as spent fuel centrifuges and super-heavy centrifuges have extremely high performance requirements. The coupling effect between the fluid and the rotor in the drum has a significant effect on the stable and reliable operation of the rotor system and the centrifugal separation machine. The liquid-solid coupling effect must be included in the rotor dynamics analysis. However, when modeling complex rotor-fluid models, there are problems such as slow calculation, multiple dimensions, and large errors. Therefore, finite element software can be used to analyze the fluid, and neural networks can be used for mutual coupling after training, which can not only accurately describe the dynamic characteristics of the fluid itself, but also accurately describe the dynamic characteristics of the rotor system under liquid-solid coupling.

[0003] On the basis of parameterization, the dynamic characteristics of the fluid in the centrifuge drum at different filling volumes and different speeds are obtained. The dynamic characteristics are reflected in the changes in the fluid exciting force. At the same time, the nonlinear factors between the structures make the transient solution of the system difficult and time-consuming. In order to improve the calculation speed, the method of separating the linear-nonlinear nodes of the rotor system is adopted, which greatly reduces the dimension of the iterative solution calculation and improves the calculation of the transient response. Summary of the invention

[0004] The present invention provides a method and device for calculating the liquid-solid coupling transient state of a centrifuge liquid-filled rotor based on a neural network-finite element collaborative model, which is designed to solve the problems in the current technology that the calculation dimension of the fluid-rotor liquid-solid coupling finite element model is too large, the simplified model cannot truly reflect the coupling characteristics, and there are difficulties in solving the transient problem of liquid-solid coupling. The present invention adopts a method based on neural network training, which not only saves the modeling process of the rotor system, but also can get closer to the real liquid-solid coupling characteristics. At the same time, the linear-nonlinear node separation calculation method is used to accelerate the solution speed of the liquid-solid coupling model and improve the transient calculation.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] S1. Obtain the dynamic characteristics of the fluid in the centrifuge drum at different filling volumes and different rotation speeds, wherein the dynamic characteristics are reflected in the fluid exciting force obtained under different working conditions and the vibration changes of the characteristic points on the axis in the time domain;

[0007] S2, input the obtained vibration displacement, liquid filling volume and rotation speed of the nonlinear node on the shaft into the neural network for training, and construct a neural network finite element dynamics model;

[0008] S3. Through the method of coupling neural network and finite element, the linear part is modeled by finite element, and the nonlinear part is described by neural network. Then, the finite element model part and the neural network part are coupled and solved according to the boundary conditions and connection conditions.

[0009] S1 includes the steps:

[0010] S1.1 Perform finite element analysis on the centrifuge rotor, i.e. use simulation software to perform fluid analysis on the liquid in the centrifuge drum and analyze the fluid excitation force function on the coupling wall;

[0011] S1.2 simulates the liquid in the drum at different filling volumes and different rotation speeds to obtain multiple sets of exciting force data;

[0012] S1.3 performs transient structural simulation based on the obtained fluid exciting force to obtain multiple sets of displacements of characteristic points on the axis.

[0013] S2 includes the steps:

[0014] S2.1 Input the vibration displacement, liquid filling volume and rotation speed of the nonlinear node on the axis into the neural network for training, and output the time domain exciting force F l , we can get the mapping function expression of the neural network from input to output, and then we can build the neural network model;

[0015] S2.2 For the neural network model, a large number of experiments are carried out to measure the nonlinear part, and a neural network dynamics model based on nonlinear grid mapping driven by experimental data is constructed to verify the accuracy of the model.

[0016] S3 includes the following steps:

[0017] S3.1 Use finite element method to model the rotor system, replace the rotor with beam element, regard the disk element as concentrated mass element, and combine the matrix of the rotor system, including mass, stiffness and damping matrix;

[0018] S3.2 Use a force model to describe the bearing unit used to connect the rotor to the centrifuge main structure;

[0019] S3.3 separates the linear and nonlinear nodes of the rotor and uses the Newmark-β method to obtain the integral expression of the rotor motion equation;

[0020] S3.4 calculates the vibration displacement of each node by solving the coupling equation based on the coupling relationship between the nonlinear neural network model and the nonlinear nodes of the rotor.

[0021] In a second aspect, the present application discloses a liquid-solid coupling calculation device for a centrifuge liquid-filled rotor based on a neural network-finite element collaborative model, comprising:

[0022] Simulation module: configured to execute the dynamic characteristics of the fluid, wherein the dynamic characteristics are embodied in the linear node displacement of the rotor at different fluid filling volumes and different rotation speeds.

[0023] Training module: configured to execute the training of the vibration displacement, filling volume and rotation speed of the nonlinear node on the input shaft in the neural network to obtain the fluid excitation force function.

[0024] Coupling module: configured to couple the finite element model part and the neural network part through boundary conditions and connection conditions to form a dynamic model.

[0025] The computing module is configured to execute a transient computing method based on a neural network-finite element collaborative model to solve the coupled dynamic model.

[0026] The device comprises a storage unit and a calculation unit, wherein the storage unit stores time domain results calculated by a simulation module, so that the calculation unit executes the liquid-solid coupling transient calculation steps of the centrifuge liquid-filled rotor of the neural network-finite element collaborative model.

[0027] The present invention has the following advantages:

[0028] The present invention provides a method and device for calculating the liquid-solid coupling transient state of a centrifuge liquid-filled rotor based on a neural network-finite element collaborative model. The finite element software is used to quickly solve the dynamic characteristics of the fluid. In order to make the simulation results more consistent with the actual situation, a parameterized method is adopted to take different filling amounts and different rotation speeds, that is, the dynamic characteristics of the fluid in the centrifuge drum under multiple working conditions. The present invention adopts a method of training the nonlinear nodes through a neural network, and the vibration displacement, filling amount and rotation speed of the nonlinear nodes on the shaft are input into the neural network for training, and the output is the time domain exciting force F l , which provides the premise for the transient analysis of the entire rotor system. By separating the nonlinear nodes of the rotor, the dimension of iterative calculation is reduced, and the time for calculating transient response is reduced. At the same time, the Newmark-β calculation method is used to integrate the rotor motion differential equation to ensure the accuracy of the calculation. Based on the steady-state finite element, the huge amount of calculation and convergence problems of transient calculation fluid are avoided. Only neural network training is needed to conveniently solve the transient dynamic response of the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 The figure is a flowchart of an implementation of an embodiment of the present invention.

[0030] Figure 22 is a diagram of a rotor system according to an embodiment of the present invention.

[0031] Figure 3 This is a neural network training diagram in an embodiment of the present invention.

[0032] Figure 4 Schematic diagram of a module of a liquid-solid coupling transient computing device for a centrifuge liquid-filled rotor based on a neural network in an embodiment of the present invention.

[0033] Figure 5 Graph showing transient simulation results of fluid in an embodiment of the present invention.

[0034] Figure 6 It is a spectrum diagram of the rotor system in an embodiment of the present invention. DETAILED DESCRIPTION

[0035] like Figure 1 As shown, a method and device for calculating the liquid-solid coupling transient state of a centrifuge liquid-filled rotor based on a neural network-finite element collaborative model according to an embodiment of the present invention:

[0036] This embodiment adopts a coupling analysis model of the rotor and the fluid. The schematic diagram of the coupling model is shown in FIG. Figure 2 The implementation steps of the centrifuge liquid-solid coupling transient calculation method, device, computer equipment and storage medium based on the neural network-finite element collaborative model in this embodiment are as follows:

[0037] The protection scope of the present invention is not limited to the description of this implementation method.

[0038] Step 1: Obtain the dynamic characteristics of the fluid in the centrifuge bowl at different filling volumes and different speeds, where the dynamic characteristics are reflected in the fluid excitation force obtained under different working conditions and the vibration changes of the characteristic points on the axis in the time domain;

[0039] 1) Perform finite element analysis on the centrifuge rotor, that is, use simulation software to perform fluid analysis on the liquid in the centrifuge drum and analyze the change in the amplitude of the fluid exciting force on the coupling wall;

[0040] 2) Simulate the liquid in the drum at different filling volumes and different rotation speeds to obtain multiple sets of exciting force data;

[0041] 3) Perform transient structural simulation based on the obtained fluid exciting force to obtain multiple sets of displacements of characteristic points on the axis.

[0042] Step 2: According to the vibration displacement of the nonlinear nodes on the axis, input it into the neural network for training and build a neural network dynamic model

[0043] 1) The vibration displacement q of the nonlinear node on the axis l =[q1,q2,q3,…,qn ], fluid pan speed The drum filling volume Q is input into the neural network for training, and the output is the time domain exciting force F l , we can get the mapping function expression of the neural network from input to output as follows, and the neural network structure is as follows Figure 3 shown.

[0044] First write the expression of the neural network model:

[0045]

[0046] where q l , ,Q,t are functions of the liquid plate displacement, velocity, liquid filling volume and time respectively.

[0047] According to the assumption, the displacement of the liquid plate is the displacement of the rotor nonlinear node, so the formula can be rewritten as:

[0048]

[0049] 2) Conduct a large number of experiments on the neural network model to measure the nonlinear part, build a neural network dynamics model based on nonlinear grid mapping driven by experimental data, conduct large-scale dynamics characteristic simulation prediction based on a high-reliability model, obtain sufficient training data samples to improve the credibility of the reverse identification mapping grid and verify the accuracy of the model;

[0050] Step 3: Through the method of coupling neural network and finite element, the linear part is modeled by finite element, and the nonlinear part is described by neural network. Then, the finite element model part and the neural network part are coupled and solved according to the boundary conditions and connection conditions.

[0051] 1) It is difficult and time-consuming to directly solve the system control equations with mixed linear and nonlinear nodes. The equations are preprocessed by the "linear-nonlinear" node separation method, which divides them into linear and nonlinear parts to reduce the dimension of the nonlinear coefficient matrix, which can greatly reduce the calculation time.

[0052] 2) Perform finite element modeling on the rotor system and combine the mass, stiffness and damping matrix of the rotor system. The dynamic equation of the rotor system is obtained as follows:

[0053]

[0054] Where [M] is the mass matrix, [C] is the damping matrix, [K] is the stiffness matrix, [F] is the fluid excitation force vector, q, They are displacement vector, velocity vector and acceleration vector respectively.

[0055] To rearrange linear and nonlinear nodes, a transformation matrix [T] is introduced, which satisfies [T] T =[T] -1 , the nonlinear nodes are moved to the nonlinear part in the lower right corner through the unitary matrix [T], and the positions of the linear nodes remain basically unchanged. The rotor system equation after processing becomes the following form:

[0056]

[0057] The subscripts i and s represent nodes related to linear or nonlinearity, respectively, and q i is the displacement of the linear node; q s is the displacement of the nonlinear node; F e is the fluid exciting force vector; It is a function of the nonlinear force node displacement, velocity, and acceleration.

[0058] Based on the Newmark-β integration method, the velocities at time t and t+Δt are And the displacement q is expressed as follows:

[0059]

[0060] Rewrite it as follows

[0061]

[0062] in, To ensure the iteration convergence, let α=1 / 4 and β=1 / 2.

[0063] At time t+Δt, substituting the above formula into the equation of the rotor system, the nonlinear node iterative calculation form is

[0064]

[0065] where Π pq =α1M pq +β1M pq +K pq ,p,q=i,s

[0066]

[0067] Rewrite the above formula as:

[0068]

[0069] Prepare for the subsequent coupled displacement, where

[0070] 3) First write the expression of the neural network model:

[0071]

[0072] in It is a function of the liquid plate displacement, speed, liquid filling volume and time.

[0073] According to the assumption, the displacement of the liquid plate is the displacement of the rotor nonlinear node, so the formula can be rewritten as:

[0074]

[0075] By coupling this formula with the formula derived in the previous stage, the coupling formula between the nonlinear node and the rotor system is obtained:

[0076] A s q s (t+Δt)=F l (t+Δt)+f s

[0077] The rotor vibration displacement coupled with the liquid exciting force can be obtained by iterative coupling using the Newton-Raphson method.

[0078] Define a function Γ(q s ) t+Δt

[0079]

[0080] Therefore, the iterative calculation formula is written as

[0081] q s [n+1]=q s [n]-{J Γ q s [n]} -1 ·Γ(q s [n]),n=1,2,...

[0082] Where n is the number of iterations, the Jacobian matrix J can be written as

[0083]

[0084] The partial derivative of a vector with respect to displacement is written as:

[0085]

[0086] Where I is the identity matrix

[0087] During the Newton-Raphson iteration, the convergence criterion for the displacement is

[0088]

[0089] where ε set is the set relative error tolerance. After multiple iterations, when q s [n] When the convergence criteria are met, use q s [n] instead of q s (t+Δt), as the nonlinear nodal displacement of the rotor system.

[0090] This application also discloses a centrifuge liquid-solid coupling calculation device based on a neural network-finite element collaborative model, see Figure 4 ,include:

[0091] Simulation module: configured to execute the dynamic characteristics of the fluid, wherein the dynamic characteristics are embodied in the linear node displacement of the rotor at different fluid filling volumes and different rotation speeds.

[0092] Training module: configured to execute the training of the vibration displacement, filling volume and rotation speed of the nonlinear node on the input shaft in the neural network to obtain the fluid excitation force function.

[0093] Coupling module: configured to couple the finite element model part and the neural network part through boundary conditions and connection conditions to form a dynamic model.

[0094] The computing module is configured to execute a transient computing method based on a neural network-finite element collaborative model to solve the coupled dynamic model.

[0095] The device comprises a storage unit and a calculation unit, wherein the storage unit stores time domain results calculated by a simulation module, so that the calculation unit executes the liquid-solid coupling transient calculation steps of the centrifuge liquid-filled rotor of the neural network-finite element collaborative model.

[0096] In this example, the transient simulation results of the fluid in the drum are as follows: Figure 5 As shown, it can be seen that during the rotor startup stage, the fluid in the bowl is in a turbulent form. After stabilization, under the action of centrifugal force, it is distributed along the inner wall of the bowl, forming a cylindrical cavity in the middle, which is symmetrical to the connecting line of the supporting bearings.

[0097] In order to reflect the influence of unbalanced force on the vibration characteristics of the entire rotor, a frequency spectrum diagram of the rotor system is drawn, such as Figure 6 shown.

[0098] An accurate and efficient neural network method is provided to solve the problem of nonlinear node calculation in the fluid-rotor coupling process.

[0099] The above is a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Those skilled in the art can make appropriate adjustments without departing from the principles of the present invention, and these adjustments should be included in the protection scope of the present invention.

Claims

1. A method for calculating the liquid-solid coupling transient state of a centrifuge liquid-filled rotor, characterized in that: The method comprises the steps of: S1. Obtain the dynamic characteristics of the fluid in the centrifuge drum at different filling volumes and different rotation speeds, wherein the dynamic characteristics are reflected in the fluid exciting force obtained under different working conditions and the vibration changes of the characteristic points on the axis in the time domain; S2, according to the vibration displacement of the nonlinear node on the shaft, the filling volume and the rotation speed are input into the neural network for training, and a neural network finite element dynamics model is constructed; S3. Through the method of coupling neural network and finite element, the linear part is modeled by finite element, and the nonlinear part is described by neural network. Then, the finite element model part and the neural network part are coupled and solved according to the boundary conditions and connection conditions.

2. The method for calculating the liquid-solid coupling transient state of a centrifuge liquid-filled rotor according to claim 1, characterized in that: S1 includes the steps: S1.1 Perform finite element analysis on the centrifuge rotor, i.e. use simulation software to perform fluid analysis on the liquid in the centrifuge drum and analyze the fluid excitation force function on the coupling wall; S1.2 simulates the liquid in the drum at different filling volumes and different rotation speeds to obtain multiple sets of exciting force data; S1.3 performs transient structural simulation based on the obtained fluid exciting force to obtain multiple sets of displacements of characteristic points on the axis.

3. The method for calculating the liquid-solid coupling transient state of a centrifuge liquid-filled rotor according to claim 1, characterized in that: S2 includes the steps: S2.1 Input the vibration displacement, liquid filling volume and rotation speed of the nonlinear node on the axis into the neural network for training, and output the time domain exciting force F l , we can get the mapping function expression of the neural network from input to output, and then we can build the neural network model; S2.2 Conduct experiments on the nonlinear part of the neural network model, build a neural network dynamics model based on nonlinear grid mapping driven by experimental data, and verify the accuracy of the model.

4. The method for calculating the liquid-solid coupling transient state of a centrifuge liquid-filled rotor according to claim 1, characterized in that: S3 includes the following steps: S3.1 Use finite element method to model the rotor system, replace the rotor with beam element, regard the disk element as concentrated mass element, and combine the matrix of the rotor system, including mass, stiffness and damping matrix; S3.2 Use a force model to describe the bearing unit used to connect the rotor to the centrifuge main structure; S3.3 separates the linear and nonlinear nodes of the rotor and uses the Newmark-β method to obtain the integral expression of the rotor motion equation; S3.4 calculates the vibration displacement of each node by solving the coupling equation based on the coupling relationship between the nonlinear neural network model and the nonlinear nodes of the rotor.

5. A liquid-solid coupling transient computing device for a centrifuge liquid-filled rotor for implementing the method as claimed in any one of claims 1 to 4, characterized in that: include: A simulation module, a training module, a coupling module and a calculation module; the simulation module, the training module, the coupling module and the calculation module are connected in sequence; A simulation module is configured to execute the dynamic characteristics of the fluid; wherein the dynamic characteristics are reflected in the linear node displacement of the rotor at different fluid filling amounts and different rotation speeds; Training module: configured to execute the training of the vibration displacement, filling volume and rotation speed of the nonlinear node on the input shaft in the neural network to obtain the fluid excitation force function; Coupling module: configured to couple the finite element model part and the neural network part through boundary conditions and connection conditions to form a dynamic model; The computing module is configured to execute a transient computing method based on a neural network-finite element collaborative model to solve the coupled dynamic model.

6. The liquid-solid coupling transient computing device for a centrifuge liquid-filled rotor according to claim 5, characterized in that: The device also includes: a storage unit and a calculation unit; the storage unit stores the time domain results calculated by the simulation module, so that the calculation unit executes the liquid-solid coupling transient calculation steps of the centrifuge liquid-filled rotor of the neural network-finite element collaborative model.

Images