CN114925558A - Digital twinning modeling method for bearing test bed - Google Patents

Digital twinning modeling method for bearing test bed Download PDF

Info

Publication number
CN114925558A
CN114925558A CN202210407846.1A CN202210407846A CN114925558A CN 114925558 A CN114925558 A CN 114925558A CN 202210407846 A CN202210407846 A CN 202210407846A CN 114925558 A CN114925558 A CN 114925558A
Authority
CN
China
Prior art keywords
model
bearing
deep learning
digital twin
rotor system
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210407846.1A
Other languages
Chinese (zh)
Inventor
马磊明
姜斌
肖玲斐
郭勤涛
陆宁云
于乾坤
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Aeronautics and Astronautics
Original Assignee
Nanjing University of Aeronautics and Astronautics
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Aeronautics and Astronautics filed Critical Nanjing University of Aeronautics and Astronautics
Priority to CN202210407846.1A priority Critical patent/CN114925558A/en
Publication of CN114925558A publication Critical patent/CN114925558A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/23—Design optimisation, verification or simulation using finite element methods [FEM] or finite difference methods [FDM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Geometry (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Hardware Design (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Graphics (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

The invention discloses a bearing test bed digital twinning modeling method which comprises the steps of firstly, utilizing a finite element model to modify in a layered mode and fitting bearing contact rigidity based on a neural network to establish a bearing-rotor system finite element model based on ADAMS, fusing the bearing-rotor system finite element model with a bearing-rotor system model based on deep learning obtained by utilizing experimental data, and constructing a digital twinning model based on structural dynamics-deep learning combination; then, establishing two-phase flow fields, bearing inner and outer ring temperature fields and fluid-solid heat transfer Fluent simulation agent models respectively by utilizing deep learning, fusing the agent models with a deep learning-based bearing lubrication model obtained by utilizing experimental data, and establishing a fluid-solid coupling-deep learning combined digital twin model; finally, performing relevance analysis on the two digital twin models by using a characteristic distribution counteradaptation depth migration theory; the invention realizes the simulation analysis of the complex working condition of the bearing test bed in the lubricating state and can obtain the running parameters with high reliability in real time.

Description

Digital twinning modeling method for bearing test bed
Technical Field
The invention relates to the technical field of digital twin modeling, and mainly relates to a bearing test bed digital twin modeling method.
Background
Based on the one-to-many characteristics of the digital twin model, a plurality of twin models can be established for a single bearing test bed, and each model corresponds to the specific physical characteristics of the model, such as a structural dynamics model, a thermal management model, a fluid-solid coupling model considering lubrication and the like. The more common digital twin modeling methods include Finite Element Method (FEM), Computational Fluid Dynamics (CFD), lumped parameter physical model, kalman filter, and deep learning modeling method based on neural network. The digital twin modeling method based on the effective fusion of the data drive and the physical model can realize the fusion and complementation of the data drive and the physical model on the principle level, reasonably and effectively combine the high-precision sensing data statistical characteristics with the system mechanism model, and obtain the high-precision bearing test bed digital twin model.
The domain adaptation method in the transfer learning aims at solving the distribution difference between the source domain and the target domain, and establishes knowledge transfer from the source domain to the target domain. According to different sources of the source field and the target field in the field adaptation method, the fault migration diagnosis research can be roughly divided into three categories, including: migration between different working conditions of the same equipment, migration between different equipment and migration of virtual equipment to a physical entity. The design and implementation of simulation models are usually highly abstract, and data obtained through simulation does not contain environmental noise, so that differences exist between the results of system simulation and experimental measurement. The simulation sample is not true enough, and the fact that the simulation sample is directly used for training a fault diagnosis model leads a network to learn the characteristics existing in the simulation sample only, and the generalization capability of the model on a real sample is difficult to guarantee. The transfer learning and domain adaptation methods are also applicable to solving the domain differences between the simulation data and the real data.
Deep learning aims to build a deep model by simulating the learning process of the brain, and learn the implicit characteristics in data by combining massive training data, so that the rich internal information of the data is described, and the classification or prediction precision is finally improved. The deep learning shows unprecedented application prospects in the fields of image recognition, voice recognition, industrial robots and the like.
Disclosure of Invention
The invention aims to: aiming at the problems in the prior art, the invention provides a digital twin modeling method for a bearing test bed, which is characterized in that a digital twin model is built by respectively combining the advantages of multidisciplinary mechanism analysis and deep learning, a multi-body dynamics model and a fluid-solid coupling model are deeply fused by using transfer learning, the simulation analysis of complex working conditions of the bearing test bed in a lubrication state can be accurately and comprehensively realized, and information such as deformation, vibration, stress, speed, lubrication and the like of each part of a bearing can be monitored in real time.
The technical scheme is as follows: in order to achieve the purpose, the invention adopts the technical scheme that:
a bearing test bed digital twin modeling method comprises the following steps:
step S1, establishing a three-dimensional geometric model and a finite element model of the bearing test bed, wherein the finite element model of the bearing test bed comprises a bearing system finite element model and a rotor system finite element model;
s2, building a digital twin model based on the combination of structure dynamics and deep learning;
s3, building a digital twin model based on fluid-solid coupling-deep learning combination;
and step S4, performing relevance analysis on the digital twin model based on the structure dynamics-deep learning combination and the digital twin model based on the fluid-solid coupling-deep learning combination by using a feature distribution counteradaptation deep migration theory.
Further, the specific steps of building a digital twin model based on the combination of structural dynamics and deep learning in step S2 include:
s2.1, calculating the nonlinear contact stiffness of the bearing roller under different loads and rotating speeds by using Nastran finite element analysis software based on a bearing system finite element model, and fitting the nonlinear contact stiffness of the bearing roller through a radial basis function neural network; specifically, the combined radial basis functions are selected for modeling as follows:
Figure BDA0003602517040000021
wherein x is R An input vector representing a neural network;
Figure BDA0003602517040000022
an output vector representing the neural network; b m The mth central cell representing the hidden layer; h is the number of neural units; v. of m Is a weight coefficient between the hidden layer and the output layer; n is a radical of an alkyl radical m Is the width of the basis function;
Figure BDA0003602517040000023
and
Figure BDA0003602517040000024
is a weight coefficient;
s2.2, carrying out simulation modal shape analysis by using Patran based on a finite element model of the rotor system;
s2.3, performing a modal knock experiment on the bearing test bed, and performing test modal shape analysis by using N-modal software;
step S2.4, comparing the simulation modal shape analysis obtained in the step S2.2 with the test modal shape analysis obtained in the step S2.3, and performing layered correction on the finite element model of the bearing-rotor system by adopting a proxy model method;
s2.5, fusing the bearing roller nonlinear contact stiffness model obtained in the step S2.1 and the bearing-rotor system finite element model obtained in the step S2.4, and respectively building an ADAMS-based bearing-rotor system rigid-flexible coupling model and a Workbench-based bearing-rotor system fully-flexible coupling model;
s2.6, confirming the ADAMS-based rigid-flexible coupling model of the bearing-rotor system by using the Workbench-based fully-flexible coupling model of the bearing-rotor system; when the frequency errors of the two models do not exceed a preset threshold value in each order, the ADAMS-based rigid-flexible coupling model of the bearing-rotor system is accurate; when the frequency error of each order of the two models is larger than the preset threshold value, repeating the step S2.4;
s2.7, under the condition that lubrication of the bearing test bed is not considered, acquiring vibration, rotating speed, temperature, stress strain and load signals of the bearing test bed in real time, establishing a bearing-rotor system proxy model based on deep learning, and testing the accuracy of the proxy model by utilizing online data to obtain the bearing-rotor system model based on deep learning;
and S2.8, fusing the ADAMS-based rigid-flexible coupling model of the bearing-rotor system obtained in the step S2.6 and the deep learning-based model of the bearing-rotor system obtained in the step S2.7 by using a weighted Softmax loss function to obtain a digital twin model based on the combination of structure dynamics and deep learning.
Further, the specific steps of building a digital twin model based on the fluid-solid coupling-deep learning combination in the step S3 include:
s3.1, respectively establishing a numerical calculation model of an oil-gas two-phase flow field in the bearing, a numerical calculation model of a temperature field of an inner ring and an outer ring of the bearing and a fluid-solid heat transfer numerical calculation model containing a solid domain in oil injection lubrication and under-ring lubrication modes by using Fluent;
s3.2, respectively establishing an oil-gas two-phase flow field numerical model in the bearing, a bearing inner-outer ring temperature field numerical model and a fluid-solid heat transfer numerical model agent model in the step S3.1 based on a deep learning method;
s3.3, acquiring vibration, rotating speed, temperature, stress strain, load and oil supply quantity signals of the bearing test bed in real time under the condition that the bearing test bed considers lubrication, establishing a bearing lubrication proxy model based on deep learning, and testing the accuracy of the proxy model by utilizing online data to obtain a bearing lubrication model based on deep learning;
and S3.4, fusing the two-phase flow field agent model based on deep learning, the bearing inner and outer ring temperature field agent model based on deep learning, the fluid-solid heat transfer agent model based on deep learning and the bearing lubrication model based on deep learning, which are obtained in the step S3.2, by using a weighted Softmax loss function, and obtaining a digital twin model based on fluid-solid coupling-deep learning combination.
Further, in the step S2.8 and the step S3.4, the weighted Softmax loss function is adopted to balance the contribution degrees of the models obtained in the step S2.6, the step S2.7, the step S3.2 and the step S3.3 to the modeling error, so that the knowledge of the models under the imbalance of the monitored data is fully learned; the weighted Softmax loss function is specifically as follows:
the training data sets based on the various models obtained in step S2.6, step S2.7, step S3.2 and step S3.3 are: { (x) i ,y i )|i＝1,2,…,M tr }; wherein x is i Is the ith training sample; y is i Is a label of the sample, and satisfies y i The method is characterized in that the method belongs to {1,2, …, R }, wherein R is different mark numbers and corresponds to different types of models; m tr The number of training data sets; the number of samples for each type of model in the dataset is as follows:
Figure BDA0003602517040000041
wherein n is c The number of training samples in the c state; i (-) is an indicator function when y i When the output of the I (-) is 1, otherwise, the output of the I (-) is 0;
designing a weighted Softmax loss function
Figure BDA0003602517040000042
Satisfies the following conditions:
Figure BDA0003602517040000043
wherein p is i,c The prediction probability of the ith sample belonging to the c model; omega c The weight of the diagnostic error of the c model is specifically expressed as follows:
Figure BDA0003602517040000044
where max (·) is the maximum value of the variable.
Further, the fluid-solid heat transfer model in step S3.1 includes:
the bearing friction torque M is calculated as follows:
M＝M 0 +M 1 (5)
wherein, M 0 Is the torque associated with the bearing type, speed and lubrication; m 1 Is the friction torque associated with the bearing load;
the forced convection heat transfer coefficient Nu _ f of the fixed wall surface meets the following conditions:
Figure BDA0003602517040000045
Figure BDA0003602517040000046
wherein R is e Is the Reynolds number; p r Is the prandtl number; u is the end face vicinity air flow velocity; d is the bearing equivalent diameter; v. of 1 Is the air kinematic viscosity;
the convection heat transfer coefficient Nu _ r of the rotating wall surface meets the following conditions:
Figure BDA0003602517040000047
wherein C is a correction coefficient; n is a correction index.
Further, the depth migration theory based on feature distribution confrontation adaptation in the step S4 includes a deep convolutional neural network shared by the domains, a feature distribution adaptation module, and a domain discriminator; the method specifically comprises the following steps:
s4.1, utilizing the digital twin model of the structure dynamics-deep learning combination obtained in the step S2 and the digital twin model construction field shared deep convolutional neural network sample space based on the fluid-structure interaction-deep learning combination obtained in the step S3;
respectively taking the digital twin model based on the structure dynamics-deep learning combination obtained in the step S2 as a source domain, taking the digital twin model based on the fluid-solid coupling-deep learning combination obtained in the step S3 as a target domain, extracting deep migration characteristics from original signals of the source domain and the target domain by utilizing a deep convolutional neural network shared by the fields, and constructing a sample space
Figure BDA0003602517040000051
The following were used:
Figure BDA0003602517040000052
wherein the content of the first and second substances,
Figure BDA0003602517040000053
is the ith source domain sample;
Figure BDA0003602517040000054
is the jth target domain sample; n is the number of source domain samples; m is the number of target domain samples;
step 4.2, calculating the migration characteristic distribution difference of the source domain and the target domain;
estimation of neural network F using maximum mean difference of Gaussian kernel implantation 2 Difference in migration characteristic distribution of layers
Figure BDA0003602517040000055
The following were used:
Figure BDA0003602517040000056
wherein the content of the first and second substances,
Figure BDA0003602517040000057
as a neural network F 2 A layer source domain sample space;
Figure BDA0003602517040000058
as a neural network F 2 A layer target domain sample space; k (·, ·) is a Gaussian kernel function; phi (-) is the nonlinear mapping from the sample space to the deep feature space;
4.3, judging whether the migration features come from a source domain or a target domain by a design domain discriminator;
taking the deep migration features extracted in the step S4.1 as the input of a neighborhood discriminator, wherein an output layer only contains a single neuron, and the loss function of the neighborhood discriminator is designed as follows:
Figure BDA0003602517040000059
wherein D (-) is a discriminator; when the loss function of the neighborhood discriminator obtains the maximum value, the migration characteristic can be judged to come from a source domain or a target domain; the output of the neighborhood arbiter can be expressed as:
Figure BDA0003602517040000061
wherein z is a deep feature extracted from the source domain and target domain samples; p is s (z) is the probability density of the source domain sample features; p t (z) is the probability density of the target domain sample features; it can be seen that when the output of the neighborhood discriminator is closer to 1, the probability that the input migration feature comes from the source domain is higher; conversely, the greater the probability from the target domain.
Has the advantages that:
the invention provides a bearing test bed digital twinning modeling method based on multidisciplinary mechanism analysis and deep learning, aiming at the problem of bearing test bed digital twinning modeling. Compared with the prior art, the advantages are that:
(1) a nonlinear contact stiffness model of the bearing roller under the working conditions of variable load and variable rotating speed is established by utilizing a neural network, and the finite element simulation precision of the bearing based on structural dynamics is improved;
(2) under the condition that lubrication is considered by a bearing test bed, respectively establishing a bearing-rotor system model and a bearing lubrication model based on deep learning by using collected bearing test data, and combining various finite element models by using a weighted Softmax loss function to improve the precision of a digital twin model; the defects that the model precision is limited and the model mechanism cannot be effectively explained due to the fact that digital twin research is carried out based on a single finite element simulation model or a deep learning model are overcome;
(3) the method comprises the following steps of researching the relevance of a digital twin model based on the combination of structure dynamics and deep learning and a digital twin model based on the combination of fluid-structure interaction and deep learning by utilizing a characteristic distribution antagonistic adaptation deep migration theory, taking the digital twin model based on the combination of structure dynamics and deep learning as a source domain, taking the digital twin model based on the combination of fluid-structure interaction and deep learning as a target domain for carrying out characteristic migration, fully utilizing the advantages of simulation analysis of different disciplines, and breaking through the limitations of simulation software of different disciplines;
(4) and constructing a digital twin model capable of truly and comprehensively reflecting the complex working condition of the bearing test bed in the lubrication state, monitoring information such as deformation, vibration, stress, speed, lubrication and the like of each part of the bearing in real time, and acquiring running parameters with high reliability.
Drawings
FIG. 1 is a flow chart of a digital twinning modeling method for a bearing test bed provided by the invention;
FIG. 2 is a finite element model of a bearing system according to an embodiment of the present invention;
FIG. 3 is a finite element model of a rotor system according to an embodiment of the present invention;
FIG. 4 is a graph showing the Patran simulation mode shape according to the embodiment of the present invention;
FIG. 5 shows the mode shape of the N-modal test in an embodiment of the present invention;
FIG. 6 is a schematic diagram of finite element model layer-by-layer modification according to an embodiment of the present invention;
FIG. 7 is a schematic diagram of a finite element model modification method based on a proxy model method according to an embodiment of the present invention;
FIG. 8 is a calculation result of a proxy model of a bearing test stand support according to an embodiment of the present invention;
FIG. 9 shows the results of the model validation based on the rigid-flexible coupling of the ADAMS bearing-rotor system in the embodiment of the present invention;
FIG. 10 is a simulation model of an oil-gas two-phase flow field inside a bearing according to an embodiment of the present invention;
fig. 11 is a schematic diagram of the depth migration theory based on feature distribution counteradaptation in the embodiment of the present invention.
Detailed Description
The present invention will be further described with reference to the accompanying drawings. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
The invention provides a bearing test bed digital twin modeling method based on the fusion of multidisciplinary mechanism analysis and deep learning, and the flow is shown in figure 1. The method mainly comprises the steps of establishing a three-dimensional geometric model and a finite element model of the bearing test bed, establishing a digital twinning model based on the combination of structure dynamics and deep learning, establishing a digital twinning model based on the combination of fluid-structure interaction and deep learning, and analyzing the relevance of the models. The method comprises the following specific steps:
and step S1, establishing a three-dimensional model of the bearing test bed by adopting SolidWorks, dividing grids, and respectively establishing finite element models of a bearing system (shown in figure 2) and a rotor system (shown in figure 3).
And step S2, constructing a digital twin model based on the combination of structural dynamics and deep learning. The method specifically comprises the following steps:
and S2.1, calculating the nonlinear contact stiffness of the bearing roller under different loads and rotating speeds by using Nastran based on a finite element model of the bearing system. The specific settlement results of the bearing stiffness under different loads are shown in the following table 1:
TABLE 1 calculation of contact stiffness of roller and inner and outer races under different loads
Figure BDA0003602517040000071
Figure BDA0003602517040000081
And modeling the nonlinear contact stiffness of the bearing roller through a radial basis function neural network. In the present embodiment, the combined radial basis functions are selected for modeling as follows:
Figure BDA0003602517040000082
wherein x is R An input vector representing a neural network;
Figure BDA0003602517040000083
an output vector representing the neural network; b m The mth central cell representing the hidden layer; h is the number of neural units; v. of m Is the weight coefficient between the hidden layer and the output layer; n is m Is the width of the basis function;
Figure BDA0003602517040000084
and
Figure BDA0003602517040000085
are the weight coefficients.
And S2.2, carrying out simulation mode vibration mode analysis by using Patran based on the finite element model of the rotor system, wherein the specific analysis result is shown in FIG. 4.
And S2.3, carrying out a modal tapping experiment on the bearing test bed, and carrying out test modal shape analysis by using N-modal software, wherein the specific analysis result is shown in figure 5.
And S2.4, comparing the simulation modal shape analysis obtained in the step S2.2 with the test modal shape analysis obtained in the step S2.3, and performing layered correction on the bearing-rotor finite element model by adopting a proxy model method.
In a complex system, a plurality of factors affecting a certain characteristic of the system need to be decomposed into a plurality of substructures to be respectively inspected so as to distinguish the influence condition of each factor. The idea of the layered correction of the complex system is as follows: the whole system is divided into four levels of a whole system, a subsystem, a component assembly and a unit, and the lower the level is, the less factors are influenced, and the lower the coupling degree is. The complex system layering process is shown in fig. 6.
The proxy model method is a basic method integrating multiple disciplines such as experimental design and analysis, mathematical statistics, linear algebra and the like, and aims to establish a relational model capable of reflecting variables and responses and analyze the conversion relationship between the variables and the responses. On the basis of reasonably arranging the sampled data obtained by the experiment, a proper regression or interpolation technology is selected for data processing, a proper function which can be expressed clearly is selected to approximate an implicit function or a function which cannot be expressed clearly, and the principle of the proxy model method is shown in fig. 7. The calculation result of the bearing test stand support proxy model in this embodiment is shown in fig. 8. The modal shape correction results are shown in table 2 below:
TABLE 2 Modal vibration shape correction results
Figure BDA0003602517040000086
Figure BDA0003602517040000091
And S2.5, fusing the bearing roller nonlinear contact rigidity model obtained in the step S2.2 and the bearing-rotor system finite element model obtained in the step S2.3, and respectively building an ADAMS-based rigid-flexible coupling model of the bearing-rotor system and a Workbech-based fully-flexible coupling model of the bearing-rotor system.
S2.6, confirming the ADAMS-based rigid-flexible coupling model of the bearing-rotor system by using the Workbench-based fully-flexible coupling model of the bearing-rotor system; when the frequency errors of the two models do not exceed a preset threshold value in each order, the ADAMS-based rigid-flexible coupling model of the bearing-rotor system is accurate; and when the frequency error of each order of the two models is larger than the preset threshold value, repeating the step S2.4 and carrying out hierarchical correction. The embodiment of the invention sets the threshold value of the modal frequency matching error to be 1%. Table 3 gives the results of the bearing test stand bracket model confirmation based on modal frequency comparison, and the finite element model pair is as shown in fig. 9:
TABLE 3 bearing test bench support model confirmation value results based on modal frequency contrast
Figure BDA0003602517040000092
As can be seen from table 3, the model matching errors of the bearing test bed support based on the modal frequency comparison are all less than 1% of the set threshold, and the step S2.4 does not need to be returned to for re-correction.
And S2.7, under the condition that lubrication of the bearing test bed is not considered, acquiring vibration, rotating speed, temperature, stress strain and load signals of the bearing test bed in real time, establishing a bearing-rotor system proxy model based on deep learning, and testing the accuracy of the proxy model by utilizing online data to obtain the bearing-rotor system model based on deep learning.
And S2.8, fusing the ADAMS-based rigid-flexible coupling model of the bearing-rotor system obtained in the step S2.6 and the deep learning-based model of the bearing-rotor system obtained in the step S2.7 by using a weighted Softmax loss function to obtain a digital twin model based on the combination of structure dynamics and deep learning.
And step S3, constructing a digital twin model based on fluid-solid coupling-deep learning combination.
And S3.1, respectively establishing a numerical calculation model of an oil-gas two-phase flow field inside the bearing, a numerical calculation model of a temperature field of an inner ring and an outer ring of the bearing and a fluid-solid heat transfer numerical model containing a solid domain in oil injection lubrication and under-ring lubrication modes by using Fluent. The oil-gas two-phase flow field simulation model in the bearing is shown in fig. 10 and comprises solid domains such as a main shaft and a bearing seat and fluid domains such as lubricating oil and air.
The bearing friction torque M calculation formula:
M＝M 0 +M 1 (2)
wherein, M 0 Is the torque related to the bearing type, speed and lubrication knowledge; m 1 Is the friction torque associated with the load to which the bearing is subjected.
The forced convection heat transfer coefficient Nu _ f of the fixed wall surface meets the following conditions:
Figure BDA0003602517040000101
Figure BDA0003602517040000102
wherein R is e Is the Reynolds number; p r Is the prandtl number; u is the end face vicinity air flow velocity; d is the bearing equivalent diameter; v. of 1 Is the air kinematic viscosity.
The convection heat exchange coefficient Nu _ r of the rotating wall surface meets the following conditions:
Figure BDA0003602517040000103
wherein C is a correction coefficient; n is a correction index.
S3.2, respectively establishing an oil-gas two-phase flow field numerical model, a bearing inner-outer ring temperature field numerical model and a fluid-solid heat transfer numerical model in the bearing in the step S3.1 based on a deep learning method;
s3.3, acquiring vibration, rotating speed, temperature, stress strain, load and oil supply quantity signals of the bearing test bed in real time under the working condition that the bearing test bed considers lubrication, establishing a bearing lubrication proxy model based on the signals, and testing the accuracy of the proxy model by utilizing online data to obtain a deep learning-based bearing lubrication model;
and S3.4, fusing the two-phase flow field agent model based on deep learning, the bearing inner and outer ring temperature field agent model based on deep learning, the fluid-solid heat transfer agent model based on deep learning and the bearing lubrication model based on deep learning, which are obtained in the step S3.2, by using a weighted Softmax loss function, and obtaining a digital twin model based on fluid-solid coupling-deep learning combination.
The weighted Softmax loss function in the step S2.8 and the step S3.4 can balance the contribution degree of each type of model obtained in the step S2.6, the step S2.7, the step S3.2 and the step S3.3 to the modeling error, and sufficiently learn the knowledge of each type of model under the condition of monitoring data imbalance; the weighted Softmax loss function is specifically:
the training data sets of the various models obtained based on step S2.6, step S2.7, step S3.2 and step S3.3 are: { (x) i ,y i )|i＝1,2,...,M tr }; wherein x is i Is the ith training sample; y is i Is a mark of the sample, and satisfies y i E.g. {1, 2., R }, wherein R is different mark numbers and corresponds to different types of models; m is a group of tr The number of training data sets; the number of samples for each type of model in the dataset is as follows:
Figure BDA0003602517040000111
wherein n is c The number of training samples in the c state; i (-) is an indicator function when y i When the output is c, the output of I (·) is 1, and conversely, the output of I (·) is 0; when the number of samples of the training data set is distributed in a balanced manner, the modeling error weight of each model is 1; when the number of samples of the training data set is distributed in an unbalanced manner, the model with a small number of samples has a larger modeling error weight, and the model with a large number of samples has a smaller modeling error weight;
design weighted Softmax loss function
Figure BDA0003602517040000112
Satisfies the following conditions:
Figure BDA0003602517040000113
wherein p is i,c The predicted probability of belonging to the c model for the ith sample; omega c The weight of the diagnostic error of the c model is specifically expressed as follows:
Figure BDA0003602517040000114
where max (·) is the maximum value of the variable.
And step S4, performing relevance analysis on the digital twin model based on the structure dynamics-deep learning combination and the digital twin model based on the fluid-solid coupling-deep learning combination by using a feature distribution counteradaptation deep migration theory. The depth migration theory based on feature distribution confrontation adaptation comprises a deep convolutional neural network shared by fields, a feature distribution adaptation module and a field discriminator, and the specific principle is shown in fig. 11; the method specifically comprises the following steps:
s4.1, constructing a deep convolutional neural network sample space shared by the field by using the structural dynamics-deep learning combined digital twin model obtained in the step S2 and the fluid-solid coupling-deep learning combined-based digital twin model obtained in the step S3;
respectively taking the digital twin model based on the structure dynamics-deep learning combination obtained in the step S2 as a source domain, taking the digital twin model based on the fluid-solid coupling-deep learning combination obtained in the step S3 as a target domain, extracting deep migration characteristics from original signals of the source domain and the target domain by utilizing a deep convolutional neural network shared by the fields, and constructing a sample space
Figure BDA0003602517040000121
The following were used:
Figure BDA0003602517040000122
wherein the content of the first and second substances,
Figure BDA0003602517040000123
is the ith source domain sample;
Figure BDA0003602517040000124
is the jth target domain sample; n is the number of source domain samples; m is the number of target domain samples;
step 4.2, calculating the migration characteristic distribution difference of the source domain and the target domain;
maximum mean difference estimation neural network F using gaussian kernel implantation 2 Difference in migration characteristic distribution of layers
Figure BDA0003602517040000125
The following were used:
Figure BDA0003602517040000126
wherein the content of the first and second substances,
Figure BDA0003602517040000127
as a neural network F 2 A layer source domain sample space;
Figure BDA0003602517040000128
as a neural network F 2 A layer target domain sample space; k (·,. cndot.) is a Gaussian kernel function; phi (-) is the nonlinear mapping from the sample space to the deep feature space;
4.3, judging whether the migration features come from a source domain or a target domain by a design domain discriminator;
taking the deep migration features extracted in the step S4.1 as the input of a neighborhood discriminator, wherein an output layer only contains a single neuron, and the loss function of the neighborhood discriminator is designed as follows:
Figure BDA0003602517040000129
wherein D (-) is a discriminator. When the loss function of the neighborhood discriminator obtains the maximum value, the accurate judgment of whether the migration characteristic comes from the source domain or the target domain can be realized; the output of the neighborhood arbiter can be expressed as:
Figure BDA00036025170400001210
wherein z is a deep feature extracted from the source domain and target domain samples; p is s (z) is the probability density of the source domain sample features; p t (z) is the probability density of the target domain sample features; it can be seen that when the output of the neighborhood discriminator is closer to 1, the probability that the input migration feature comes from the source domain is higher; conversely, the greater the probability from the target domain.
The above description is only of the preferred embodiments of the present invention, and it should be noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the invention and these are intended to be within the scope of the invention.

Claims (6)

1. A bearing test bed digital twin modeling method is characterized by comprising the following steps:
step S1, establishing a three-dimensional geometric model and a finite element model of the bearing test bed, wherein the finite element model of the bearing test bed comprises a bearing system finite element model and a rotor system finite element model;
s2, building a digital twin model based on the combination of structure dynamics and deep learning;
step S3, constructing a digital twin model based on fluid-structure interaction and deep learning combination;
and step S4, performing relevance analysis on the digital twin model based on the structure dynamics-deep learning combination and the digital twin model based on the fluid-structure interaction-deep learning combination by using a feature distribution counteradaptation deep migration theory.
2. The bearing test bed digital twin modeling method according to claim 1, wherein the specific steps of building a digital twin model based on a combination of structure dynamics and deep learning in the step S2 include:
s2.1, calculating the nonlinear contact stiffness of the bearing roller under different loads and rotating speeds by using Nastran finite element analysis software based on a bearing system finite element model, and fitting the nonlinear contact stiffness of the bearing roller through a radial basis function neural network; specifically, the combined radial basis functions are selected for modeling as follows:
Figure FDA0003602517030000011
wherein x is R An input vector representing a neural network;
Figure FDA0003602517030000012
an output vector representing the neural network; b is a mixture of m The mth central cell representing the hidden layer; h is the number of neural units; v. of m Is a weight coefficient between the hidden layer and the output layer; n is m Is the width of the basis function;
Figure FDA0003602517030000013
and
Figure FDA0003602517030000014
is a weight coefficient;
s2.2, carrying out simulation modal shape analysis by using Patran based on a finite element model of the rotor system;
s2.3, performing a modal knock experiment on the bearing test bed, and performing test modal shape analysis by using N-modal software;
step S2.4, comparing the simulation modal shape analysis obtained in the step S2.2 with the test modal shape analysis obtained in the step S2.3, and performing layered correction on the finite element model of the bearing-rotor system by adopting a proxy model method;
s2.5, fusing the bearing roller nonlinear contact stiffness model obtained in the step S2.1 and the bearing-rotor system finite element model obtained in the step S2.4, and respectively building an ADAMS-based bearing-rotor system rigid-flexible coupling model and a Workbench-based bearing-rotor system fully-flexible coupling model;
s2.6, confirming the ADAMS-based rigid-flexible coupling model of the bearing-rotor system by using the Workbench-based fully-flexible coupling model of the bearing-rotor system; when the frequency errors of the two models in each order do not exceed a preset threshold, the ADAMS-based rigid-flexible coupling model of the bearing-rotor system is accurate; when the frequency error of each order of the two models is larger than the preset threshold value, repeating the step S2.4;
s2.7, under the condition that lubrication of the bearing test bed is not considered, acquiring vibration, rotating speed, temperature, stress strain and load signals of the bearing test bed in real time, establishing a bearing-rotor system proxy model based on deep learning, and testing the accuracy of the proxy model by utilizing online data to obtain the bearing-rotor system model based on deep learning;
and S2.8, fusing the ADAMS-based rigid-flexible coupling model of the bearing-rotor system obtained in the step S2.6 and the deep learning-based model of the bearing-rotor system obtained in the step S2.7 by using a weighted Softmax loss function to obtain a digital twin model based on the combination of structure dynamics and deep learning.
3. The bearing test bed digital twin modeling method according to claim 1, wherein the specific steps of building a digital twin model based on fluid-solid coupling-deep learning combination in the step S3 include:
s3.1, respectively establishing a numerical calculation model of an oil-gas two-phase flow field in the bearing, a numerical calculation model of a temperature field of an inner ring and an outer ring of the bearing and a fluid-solid heat transfer numerical calculation model containing a solid domain in oil injection lubrication and under-ring lubrication modes by using Fluent;
s3.2, respectively establishing an oil-gas two-phase flow field numerical model, a bearing inner-outer ring temperature field numerical model and a fluid-solid heat transfer numerical model proxy model in the step S3.1 based on a deep learning method;
s3.3, acquiring vibration, rotating speed, temperature, stress strain, load and oil supply quantity signals of the bearing test bed in real time under the condition that the bearing test bed considers lubrication, establishing a bearing lubrication proxy model based on deep learning, and testing the accuracy of the proxy model by utilizing online data to obtain a bearing lubrication model based on deep learning;
and S3.4, fusing the two-phase flow field agent model based on deep learning, the bearing inner and outer ring temperature field agent model based on deep learning, the fluid-solid heat transfer agent model based on deep learning, and the bearing lubrication model based on deep learning, which are obtained in the step S3.2, by using a weighted Softmax loss function, and obtaining a digital twin model based on fluid-solid coupling-deep learning combination.
4. A bearing test bed digital twin modeling method according to any one of claims 2-3, wherein in step S2.8 and step S3.4, the contribution degree of each type of model obtained in step S2.6, step S2.7, step S3.2 and step S3.3 to the modeling error is balanced by a weighted Softmax loss function, so that the knowledge of each type of model under the imbalance of the monitored data is fully learned; the weighted Softmax loss function is specifically as follows:
the training data sets of the various models obtained based on step S2.6, step S2.7, step S3.2 and step S3.3 are: { (x) i ,y i )|i＝1,2,...,M tr }; wherein x is i Is the ith training sample; y is i Is a label of the sample, and satisfies y i The method is characterized in that the method belongs to the field of modeling, belongs to {1, 2., R }, wherein R is different mark numbers and corresponds to different types of models; m tr The number of training data sets; the number of samples for each type of model in the dataset is as follows:
Figure FDA0003602517030000031
wherein the content of the first and second substances,n c the number of training samples in the c state; i (-) is an indicator function when y i When the output of the I (-) is 1, otherwise, the output of the I (-) is 0;
designing a weighted Softmax loss function
Figure FDA0003602517030000032
Satisfies the following conditions:
Figure FDA0003602517030000033
wherein p is i,c The predicted probability of belonging to the c model for the ith sample; omega c The weight of the diagnostic error of the c model is specifically expressed as follows:
Figure FDA0003602517030000034
where max (·) is the maximum value of the variable.
5. The bearing test stand digital twinning modeling method of claim 3, wherein the fluid-solid heat transfer model in step S3.1 includes:
the bearing friction torque M is calculated as follows:
M＝M 0 +M 1 (5)
wherein, M 0 Is the torque associated with the bearing type, speed and lubrication; m 1 Is the friction torque associated with the bearing load;
the forced convection heat transfer coefficient Nu _ f of the fixed wall surface meets the following conditions:
Figure FDA0003602517030000035
Figure FDA0003602517030000036
wherein R is e Is the Reynolds number; p r Is the prandtl number; u is the end face vicinity air flow velocity; d is the bearing equivalent diameter; v. of 1 Is the air kinematic viscosity;
the convection heat transfer coefficient Nu _ r of the rotating wall surface meets the following conditions:
Figure FDA0003602517030000041
wherein C is a correction coefficient; n is a correction index.
6. The bearing test bed digital twin modeling method according to claim 1, wherein the depth migration theory based on feature distribution counteradaptation in step S4 includes a deep convolutional neural network shared by domains, a feature distribution fitting module and a domain discriminator; the method specifically comprises the following steps:
s4.1, utilizing the digital twin model of the structure dynamics-deep learning combination obtained in the step S2 and the digital twin model construction field shared deep convolutional neural network sample space based on the fluid-structure interaction-deep learning combination obtained in the step S3;
respectively taking the digital twin model based on the structure dynamics-deep learning combination obtained in the step S2 as a source domain, taking the digital twin model based on the fluid-solid coupling-deep learning combination obtained in the step S3 as a target domain, extracting deep migration characteristics from original signals of the source domain and the target domain by utilizing a deep convolutional neural network shared by the fields, and constructing a sample space
Figure FDA0003602517030000042
The following:
Figure FDA0003602517030000043
wherein, the first and the second end of the pipe are connected with each other,
Figure FDA0003602517030000044
is the ith source domain sample;
Figure FDA0003602517030000045
is the jth target domain sample; n is the number of source domain samples; m is the number of target domain samples;
step 4.2, calculating the migration characteristic distribution difference of the source domain and the target domain;
estimation of neural network F using maximum mean difference of Gaussian kernel implantation 2 Difference in migration characteristic distribution of layers
Figure FDA0003602517030000046
The following were used:
Figure FDA0003602517030000047
wherein the content of the first and second substances,
Figure FDA0003602517030000048
as a neural network F 2 A layer source domain sample space;
Figure FDA0003602517030000049
as a neural network F 2 A layer target domain sample space; k (·,. cndot.) is a Gaussian kernel function; phi (-) is the nonlinear mapping from the sample space to the deep feature space;
4.3, judging whether the migration features come from a source domain or a target domain by a design domain discriminator;
taking the deep migration features extracted in the step S4.1 as the input of a neighborhood discriminator, wherein an output layer only contains a single neuron, and the loss function of the neighborhood discriminator is designed as follows:
Figure FDA0003602517030000051
wherein D (-) is a discriminator; when the loss function of the neighborhood discriminator obtains the maximum value, the migration characteristic can be judged to come from a source domain or a target domain; the output of the neighborhood arbiter can be expressed as:
Figure FDA0003602517030000052
wherein z is a deep feature extracted from the source domain and target domain samples; p is s (z) is the probability density of the source domain sample features; p t (z) is the probability density of the target domain sample features; when the output of the neighborhood discriminator is closer to 1, the probability that the input migration feature comes from the source domain is higher; conversely, the greater the probability from the target domain.
CN202210407846.1A 2022-04-19 2022-04-19 Digital twinning modeling method for bearing test bed Pending CN114925558A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210407846.1A CN114925558A (en) 2022-04-19 2022-04-19 Digital twinning modeling method for bearing test bed

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210407846.1A CN114925558A (en) 2022-04-19 2022-04-19 Digital twinning modeling method for bearing test bed

Publications (1)

Publication Number Publication Date
CN114925558A true CN114925558A (en) 2022-08-19

Family

ID=82807660

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210407846.1A Pending CN114925558A (en) 2022-04-19 2022-04-19 Digital twinning modeling method for bearing test bed

Country Status (1)

Country Link
CN (1) CN114925558A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115859837A (en) * 2023-02-23 2023-03-28 山东大学 Fan blade dynamic impact detection method and system based on digital twin modeling

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115859837A (en) * 2023-02-23 2023-03-28 山东大学 Fan blade dynamic impact detection method and system based on digital twin modeling

Similar Documents

Publication Publication Date Title
Fu et al. A state-of-the-art review on uncertainty analysis of rotor systems
Xu et al. A machine learning-based design representation method for designing heterogeneous microstructures
CN112765748B (en) Rotary mechanical digital twin modeling method for mechanism-data heterogeneous information fusion
CN116380166A (en) Equipment abnormality monitoring method, electronic equipment and storage medium
CN116150897A (en) Machine tool spindle performance evaluation method and system based on digital twin
CN112668227A (en) Thin-wall part cutter relieving deformation error prediction model establishing method and application thereof
CN116448419A (en) Zero sample bearing fault diagnosis method based on depth model high-dimensional parameter multi-target efficient optimization
Alnowaiser et al. Digital twin: current research trends and future directions
Guo et al. Real-time prediction of remaining useful life and preventive maintenance strategy based on digital twin
Liu et al. Digital Twin‐Driven Machine Condition Monitoring: A Literature Review
CN114528666A (en) Complex structure system reliability method based on multi-level distributed cooperative agent model
CN115238394B (en) Multi-source uncertainty hybrid reliability digital twin modeling method for composite material structure
CN114925558A (en) Digital twinning modeling method for bearing test bed
Kulkarni et al. Sensor selection framework for designing fault diagnostics system
Li et al. Thermal error modeling of feed axis in machine tools using particle swarm optimization-based generalized regression neural network
Li et al. A review on physics-informed data-driven remaining useful life prediction: Challenges and opportunities
Suryawanshi et al. Empirical-based DA and ANN to diagnose misalignment in rotor-bearing system
Zeng et al. Milling surface roughness prediction based on physics-informed machine learning
Hausmann et al. Systematic approaches for sensor selection and integration–A systematic literature review
Groensfelder et al. Application of machine learning procedures for mechanical system modelling: capabilities and caveats to prediction-accuracy
CN112990601B (en) Worm wheel machining precision self-healing system and method based on data mining
CN113849576A (en) Specific group analysis method and system based on knowledge graph
CN115879355A (en) Temperature compensation method of piezoelectric sensor
CN117454530B (en) Digital twinning-based automobile body part modeling and detecting method and system
CN115630334B (en) CBAM-based cross-domain mechanical fault diagnosis method based on multi-channel feature fusion and application

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination