CN115186558A - Thin-wall part stress-strain real-time prediction method based on digital twinning - Google Patents
Thin-wall part stress-strain real-time prediction method based on digital twinning Download PDFInfo
- Publication number
- CN115186558A CN115186558A CN202210884474.1A CN202210884474A CN115186558A CN 115186558 A CN115186558 A CN 115186558A CN 202210884474 A CN202210884474 A CN 202210884474A CN 115186558 A CN115186558 A CN 115186558A
- Authority
- CN
- China
- Prior art keywords
- thin
- wall part
- strain
- stress
- finite element
- 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
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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
Abstract
The invention discloses a digital twin-based thin-wall part stress-strain real-time prediction method, which comprises the following steps of: (1) establishing a finite element model in the thin-wall part installation process for finite element simulation to obtain finite element simulation data; (2) establishing a BP neural network, rearranging finite element data according to coordinates, dividing the finite element data into a training set, a verification set and a test set, and transmitting the finite element data into the BP neural network for network training to obtain an agent model for stress-strain prediction of the thin-wall part; (3) deploying a customer service end on the production line site to realize sensor data collection and installation process control, deploying a thin-wall part stress-strain real-time prediction agent model at the service end and realizing visualization, and establishing a stress-strain digital twin model in the thin-wall part installation process; (4) after actual deployment, the existing agent model is optimized according to data collected by the sensors in the actual installation process, so that the existing agent model continuously approaches the actual working condition. The invention solves the problem of stress-strain real-time measurement of thin-wall parts which are complex in structure and inconvenient to measure like array antennas.
Description
Technical Field
The invention belongs to the technical field of thin-wall part stress-strain prediction, and particularly relates to a thin-wall part stress-strain real-time prediction method driven by a digital twinning technology.
Background
The circuit board surface element on the new generation array antenna has high integration level and high compactness, and is extremely important to the control of the installation process. However, there are few mounting locations where it can be used for real-time measurement of equipment, resulting in the inability to measure and monitor its stress-strain state in real-time during actual use. The traditional measurement method is used for carrying out real-time strain measurement on the substrate, or the substrate is required to be directly contacted, and the substrate can be damaged by partial measurement modes; or the manufacturing cost is too expensive, a large installation space is needed, the method is not suitable for the installation process of the daughter board substrate with the array antenna function, and the real-time requirement cannot be well met.
In view of the fact that the functional daughter board is expensive in manufacturing cost, the strain measurement process is tedious, and a large amount of manpower, material resources and financial resources are consumed in practical experiments, the current research on real-time monitoring and control means in the installation process is difficult and serious.
The formation and rapid development of the finite element method make the finite element method become a numerical analysis method which can solve the problems of irregular boundaries, complex shapes, material nonlinearity and the like, and is convenient, efficient and high-precision. A BP (Back Propagation) neural network, which is a relatively representative algorithm in an Artificial Neural Network (ANN), can theoretically fit an arbitrary formula. The installation process can be monitored and controlled remotely by matching with a digital twin technology, and the installation process can be controlled timely, effectively and accurately.
According to the invention, the proxy model is constructed based on the BP neural network technology, and the stress-strain prediction and feedback control field of the surface of the thin-wall part can be completed within seconds by the remote end through the screw pretightening force data measured in real time by the sensor deployed on the thin-wall part mounting field, and cloud image visualization is realized.
Disclosure of Invention
The invention aims to solve the problems of the prior art, and the conception is as follows:
1. the finite element technology is utilized in the early stage to solve the problems that the functional daughter board is expensive in manufacturing cost, the strain measurement process is complicated, and a large amount of manpower, material resources and financial resources are consumed in an actual experiment, so that a data set is obtained.
2. And solving the problem of instantaneity of stress-strain prediction by using a proxy model technology.
3. And a digital twin system is established, and the remote monitoring and control decision of the installation process are realized.
4. And collecting data in the actual use process, and continuously training the deployed model to continuously approach the actual working condition.
The technical scheme of the invention is as follows:
a thin-wall part stress-strain real-time prediction method based on digital twinning is provided, and comprises the following steps:
step 1, establishing a finite element model for carrying out finite element simulation in the thin-wall part installation process to obtain finite element simulation data, wherein the finite element simulation data comprises pretightening force of all screws and corresponding surface stress-strain data.
Step 2, establishing a BP neural network, rearranging finite element data according to coordinates, dividing the finite element data into a training set, a verification set and a test set, transmitting the finite element data into the BP neural network for network training, and obtaining a proxy model for thin-wall part stress-strain prediction:
step 2.1: establishing a BP neural network to realize the input-output prediction of an agent model, wherein the BP neural network is based on a TensorFlow framework, an input layer is a one-dimensional vector [ F1, F2, F3, F4, E ], and an output layer is a one-dimensional vector [ epsilon 1, epsilon 2, \8230 ]; the middle of the model comprises 7 hidden layers, so that a large number of node outputs can be predicted by using a small number of input features;
and 2.2, selecting the screw pretightening force as input and the stress-strain on the surface of the thin-wall part as output by the proxy model, training the neural network by using the data set in the step 1 until the error requirement is met, and obtaining a real-time prediction proxy model which can take the screw pretightening force as input and the stress-strain on the surface of the thin-wall part as output.
Step 3, deploying a customer service end on the production line site, realizing sensor data collection and installation process control, deploying a thin-wall part stress-strain real-time prediction agent model at the service end, realizing visualization, and establishing a stress-strain digital twin model of the thin-wall part installation process:
step 3.1: based on a C/S framework, a customer service end is deployed in a production field, and a remote deployment service end performs data exchange based on a TCP/IP protocol in the middle;
step 3.2: the sensor is deployed on the production line site and is responsible for measuring pre-tightening force and stress-strain in the actual installation process, the sensor is connected with a customer service end through a Modbus control bus, and the sensor data collection and installation process control are realized at the customer service end and are transmitted to a remote service end;
step 3.3: and (3) packaging the thin-wall part stress-strain real-time prediction agent model established in the step (2) into a server, performing prediction and decision according to data transmitted by a field customer service terminal, realizing visualization based on unity, and establishing a stress-strain digital twin model in the thin-wall part installation process.
Step 4, after actual deployment, optimizing the existing agent model according to data collected by the sensor in the actual installation process: and (3) based on the digital twin system built in the step (3), carrying out optimization training on the thin-wall part stress-strain real-time prediction agent model through data collected in the actual installation process, so that the thin-wall part stress-strain real-time prediction agent model continuously approaches the actual working condition.
The invention has the following advantages and beneficial effects:
the invention discloses a digital twin-based thin-wall part stress-strain real-time prediction method, which is used for solving the problems that a thin-wall part similar to a new generation of array antenna functional daughter board has high integration level and high compactness of a surface element part, and the stress-strain state of the thin-wall part is difficult to measure and monitor in real time by using a traditional measuring method.
The innovation of the method lies in patent requirement 1, aiming at thin-wall parts which are expensive in manufacturing cost and complicated in strain measurement process, and the actual experiment needs to consume a large amount of manpower, material resources and financial resources, so that a finite element simulation technology is selected to obtain a data set; according to patent claim 2, a proxy model technology is utilized to train a BP neural network, so that a proxy model which can predict the stress-strain state of tens of thousands of nodes according to several pretightening force characteristics and can predict the stress-strain state in the installation process in real time is obtained; according to the patent claim 3, a digital twin system is established, so that the remote monitoring and control decision of the installation process are realized, and the stress-strain state of the thin-wall part in the installation process can be controlled in real time; according to patent claim 4, after the system is deployed, training can be continuously carried out on the model in the later period according to data actually collected in the installation process, so that the agent model continuously approaches to the physical model under the actual working condition.
Compared with the traditional measuring method, the method can realize the remote prediction of the stress-strain state of the whole surface of the thin-wall part in real time and accurately and realize feedback control.
Drawings
FIG. 1 is the overall architecture of the overall prediction method of the present invention;
FIG. 2 is a flow chart of agent model training of the present invention;
FIG. 3 is a BP neural network structure established by the present invention;
FIG. 4 is an error test curve of the stress-strain prediction model constructed in accordance with the present invention;
FIG. 5 is a cloud visualization of stress-strain data according to the present invention;
FIG. 6 is an overall schematic view of a thin-walled part according to a preferred embodiment of the present invention;
Detailed Description
The technical solutions in the embodiments of the present invention will be described in detail and clearly with reference to the accompanying drawings. The described embodiments are only some of the embodiments of the present invention.
The technical scheme for solving the technical problems is as follows:
a thin-wall part stress-strain real-time prediction method based on digital twinning. The method comprises the following steps:
step 1, establishing a finite element model in the thin-wall part installation process to perform finite element simulation, performing finite element simulation, and deriving finite element simulation data to obtain the pretightening force of all screws and the corresponding surface stress-strain data.
Step 2, establishing a BP neural network, rearranging finite element data according to coordinates, dividing the finite element data into a training set, a verification set and a test set, transmitting the finite element data into the BP neural network for network training, and obtaining a proxy model for thin-wall part stress-strain prediction:
step 2.1: establishing a BP neural network to realize the input-to-output prediction of an agent model, wherein the BP neural network is based on a TensorFlow framework, an input layer is a one-dimensional vector [ F1, F2, F3, F4, E ], and an output layer is a one-dimensional vector [ epsilon 1, epsilon 2, \ 8230 ]; the middle of the model comprises 7 hidden layers, so that a large number of node outputs can be predicted by using a small number of input features;
and 2.2, selecting the screw pretightening force as input and the stress-strain on the surface of the thin-wall part as output by the proxy model, training the neural network by using the data set in the step 1 until the error requirement is met, and obtaining a real-time prediction proxy model which can take the screw pretightening force as input and the stress-strain on the surface of the thin-wall part as output.
Step 3, deploying a customer service end on the production line site, realizing sensor data collection and installation process control, deploying a thin-wall part stress-strain real-time prediction agent model at the service end, realizing visualization, and establishing a stress-strain digital twin model of the thin-wall part installation process:
step 3.1: based on a C/S framework, deploying a customer service end on a production field, deploying a service end remotely, and performing data exchange based on a TCP/IP protocol in the middle;
step 3.2: the sensor is deployed on the production line site and is responsible for measuring pre-tightening force and stress-strain in the actual installation process, the sensor is connected with a customer service end through a Modbus control bus, and the sensor data collection and installation process control are realized at the customer service end and are transmitted to a remote service end;
step 3.3: and (3) packaging the thin-wall part stress-strain real-time prediction agent model established in the step (2) into a server, performing prediction and decision according to data transmitted by a field customer service terminal, realizing visualization based on unity, and establishing a stress-strain digital twin model in the thin-wall part installation process.
Step 4, after actual deployment, optimizing the existing agent model according to data collected by the sensor in the actual installation process: and (3) based on the digital twin system built in the step (3), carrying out optimization training on the thin-wall part stress-strain real-time prediction agent model through data collected in the actual installation process, so that the thin-wall part stress-strain real-time prediction agent model continuously approaches the actual working condition.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising a," "8230," "8230," or "comprising" does not exclude the presence of other like elements in a process, method, article, or apparatus comprising the element.
The above examples are to be construed as merely illustrative and not limitative of the remainder of the disclosure in any way whatsoever. After reading the description of the present invention, the skilled person can make various changes or modifications to the invention, and these equivalent changes and modifications also fall into the scope of the invention defined by the claims.
Claims (5)
1. A thin-wall part stress-strain real-time prediction method based on digital twinning is characterized by comprising the following steps:
step 1, establishing a finite element model for carrying out finite element simulation in the thin-wall part installation process to obtain finite element simulation data;
step 2, establishing a BP neural network, rearranging finite element data according to coordinates, dividing the finite element data into a training set, a verification set and a test set, and transmitting the finite element data into the BP neural network for network training to obtain a proxy model for stress-strain prediction of the thin-wall part;
step 3, deploying a customer service end on the production line site, realizing sensor data collection and installation process control, deploying a thin-wall part stress-strain real-time prediction agent model at the service end, realizing visualization, and establishing a stress-strain digital twin model in the thin-wall part installation process;
and 4, after actual deployment, optimizing the existing agent model according to data collected by the sensor in the actual installation process, so that the existing agent model continuously approaches the actual working condition.
2. The method for predicting stress-strain of a thin-wall part based on digital twinning in real time as claimed in claim 1, wherein in the step 1, a convenient, efficient and high-precision finite element technology is used in an early stage to replace an actual experiment which needs to consume a large amount of manpower, material resources and financial resources, and the method comprises the steps of establishing a finite element model (defining the geometric structure, the contact fit relationship and the material properties of the thin-wall part, carrying out grid division, and setting boundary conditions (pre-tightening force and gravity)), and then carrying out finite element calculation to obtain simulation data of the thin-wall part as a subsequent data set.
3. The digital twin-based thin-walled part stress-strain real-time prediction method according to claim 2, characterized in that in the step 2:
step 2.1: establishing a BP neural network to realize the input-to-output prediction of an agent model, wherein the BP neural network is based on a TensorFlow framework, an input layer is a one-dimensional vector [ F1, F2, F3, F4, E ], and an output layer is a one-dimensional vector [ epsilon 1, epsilon 2, \ 8230 ]; the middle of the model comprises 7 hidden layers, so that a large number of node outputs can be predicted by using a small number of input features;
and 2.2, selecting the pretightening force of the screw as input and the stress-strain on the surface of the thin-wall part as output by the proxy model, training the neural network by using the data set in the step 1 until the error requirement is met, and obtaining the real-time prediction proxy model which can take the pretightening force of the screw as input and the stress-strain on the surface of the thin-wall part as output.
4. The digital twin-based thin-walled part stress-strain real-time prediction method according to claim 3, wherein in the step 3:
step 3.1: based on a C/S framework, deploying a customer service end on a production field, deploying a service end remotely, and performing data exchange based on a TCP/IP protocol in the middle;
step 3.2: the sensor is deployed on the production line site and is responsible for measuring pre-tightening force and stress-strain in the actual installation process, the sensor is connected with a customer service end through a Modbus control bus, and the sensor data collection and installation process control are realized at the customer service end and are transmitted to a remote service end;
step 3.3: and (3) packaging the thin-wall part stress-strain real-time prediction agent model established in the step (2) into a service end, performing prediction and decision according to data transmitted by a field customer service end, realizing visualization based on unity, and establishing a stress-strain digital twin model in the thin-wall part installation process.
5. The thin-wall part stress-strain real-time prediction method based on the digital twin according to claim 4, wherein in the step 4, based on the digital twin system built in the step 3, the thin-wall part stress-strain real-time prediction agent model is optimized and trained through data collected in the actual installation process, so that the thin-wall part stress-strain real-time prediction agent model continuously approaches the actual working condition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210884474.1A CN115186558A (en) | 2022-07-25 | 2022-07-25 | Thin-wall part stress-strain real-time prediction method based on digital twinning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210884474.1A CN115186558A (en) | 2022-07-25 | 2022-07-25 | Thin-wall part stress-strain real-time prediction method based on digital twinning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115186558A true CN115186558A (en) | 2022-10-14 |
Family
ID=83521014
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210884474.1A Pending CN115186558A (en) | 2022-07-25 | 2022-07-25 | Thin-wall part stress-strain real-time prediction method based on digital twinning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115186558A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117910361A (en) * | 2024-03-14 | 2024-04-19 | 南京理工大学 | Quick reconstruction method for array antenna assembly performance digital twin model based on simulation analysis |
-
2022
- 2022-07-25 CN CN202210884474.1A patent/CN115186558A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117910361A (en) * | 2024-03-14 | 2024-04-19 | 南京理工大学 | Quick reconstruction method for array antenna assembly performance digital twin model based on simulation analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106197970B (en) | A kind of bridge rope monitoring method and system based on optimization tensioning string model | |
| CN100568249C (en) | System and method with ultimate principle emulation semiconductor-assisted manufacture process | |
| CN110977614B (en) | Health diagnosis method for numerical control machine tool | |
| CN101655456B (en) | Insulator equivalent salt density optical fiber test method based on particle swarm support vector machine | |
| CN113776794B (en) | Fault diagnosis method, device and system for embedded edge calculation | |
| CN115186558A (en) | Thin-wall part stress-strain real-time prediction method based on digital twinning | |
| CN111008641B (en) | Power transmission line tower external force damage detection method based on convolutional neural network | |
| CN111878322B (en) | Wind power generator device | |
| CN112270429A (en) | Cloud edge cooperation-based power battery pole piece manufacturing equipment maintenance method and system | |
| CN110991776A (en) | Method and system for realizing water level prediction based on GRU network | |
| CN105117538A (en) | Method for warning waving of power transmission channels | |
| Asanza et al. | Fpga based meteorological monitoring station | |
| Liyakat et al. | IoT Based Arduino-Powered Weather Monitoring System | |
| CN112541569A (en) | Sensor online training system and method based on machine learning | |
| Murugan et al. | AI based Weather Monitoring System | |
| CN115577856A (en) | Method and system for predicting construction cost and controlling balance of power transformation project | |
| CN109872520B (en) | Bridge health monitoring system based on wireless sensor network and working method thereof | |
| CN113570111B (en) | Bridge health state on-chip monitoring method based on lightweight network | |
| Huang et al. | The system of temperature rise monitoring and temperature prediction for power equipment | |
| CN115566997B (en) | Photovoltaic module attenuation test system and method for air compression station | |
| CN116738893B (en) | Spacecraft simulation prediction system based on analysis of synchronous transmitter | |
| CN112531679A (en) | Load measuring characteristic big data monitoring equipment and monitoring method | |
| CN111857015A (en) | Power transmission and transformation cloud intelligent controller | |
| CN113570478B (en) | Bridge health state intelligent evaluation method based on edge calculation | |
| CN117309092A (en) | Real-time water level prediction method, device and system |
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 |