CN111400962B - High-speed dynamic balance mechanics resolving method based on machine learning - Google Patents
High-speed dynamic balance mechanics resolving method based on machine learning Download PDFInfo
- Publication number
- CN111400962B CN111400962B CN202010132666.8A CN202010132666A CN111400962B CN 111400962 B CN111400962 B CN 111400962B CN 202010132666 A CN202010132666 A CN 202010132666A CN 111400962 B CN111400962 B CN 111400962B
- Authority
- CN
- China
- Prior art keywords
- rotor
- balanced
- dynamic balance
- multiplied
- speed
- 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.)
- Active
Links
Images
Landscapes
- Testing Of Balance (AREA)
Abstract
The invention discloses a high-speed dynamic balance mechanics resolving method based on machine learning, which comprises the following steps: step 1, determining positions of a supporting structure and a measuring point according to the type of a rotor to be balanced; step 2, establishing a three-dimensional solid model of the rotor, and calling the existing rotor test data of the same type; step 3, performing weight increasing and decreasing dynamic simulation by using the three-dimensional solid model to obtain a speed increasing-decreasing response amplitude value of the marked part corresponding to the actual test point; step 4, matching the generated data with the rotor test data of the same type, and performing hybrid training on the neural network, wherein the number of nodes of an input layer is the number of test points multiplied by the number of rotating speeds to be balanced, and the number of nodes of an output layer is the number of balanced surfaces multiplied by 2; step 5, using formula V0+ n (p) ═ 0 calculates the weight p.
Description
Technical Field
The invention belongs to the field of mechanical vibration calculation and vibration reduction methods, and particularly relates to a high-speed dynamic balance mechanical calculation method based on machine learning, which is suitable for high-speed dynamic balance of turbomachinery rotors.
Background
Dynamic balance is an important link for testing and correcting before delivery of turbine machinery, and fundamental frequency vibration caused by unbalance due to factors such as manufacturing tolerance is reduced by performing dynamic balance (weight adding and reducing) correction on a rotor. Dynamic balance before leaving a factory is usually developed based on a dynamic balance testing machine, and a rotor can be continuously in service after being remanufactured and also through dynamic balance correction. The traditional dynamic balance theory is quite mature, the ABC method is generally used for low-speed dynamic balance (rigid dynamic balance), and the influence coefficient method is generally used for high-speed dynamic balance, and the unbalance characteristic of the system is obtained through least square calculation to correct.
At present, with the technical transition of turbomachinery rotors from slow turbines to fast turbines, the traditional dynamic balance method has certain defects, and the ABC method is based on the rigid assumption of rotors, namely the working rotating speed is far from reaching the critical rotating speed, and the rigid assumption cannot be realized necessarily with the increase of the working rotating speed; although the influence coefficient method is applicable to the high-speed dynamic balance working condition, in the process of obtaining the influence coefficient, the influence coefficient method requires that the forced response of the system meets the superposition principle based on the linear relation of the input and the output of the system, the nonlinear performance of the system is not obvious at low speed, the error caused by approximate equivalent linearization is in an acceptable range, the nonlinear factor is strongly expressed at high-speed dynamic balance, the forced response of the system does not meet the superposition principle any more, and the result obtained by the linearization assumption can bring great error.
Therefore, a high-speed dynamic balance technology needs to be developed, which can effectively perform high-speed dynamic balance correction on a system and can meet special requirements of different types of rotors. The high-speed dynamic balance detection process technology developed by the method can break through the blockade of the technology and equipment in western developed countries, expand the application products of the existing dynamic balance technology from sliding bearing series to rolling bearing series electromechanical equipment including aero-engines, and expand the service field and range of the electromechanical equipment.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, provides a dynamic balance mechanical calculation method suitable for turbomachinery rotors with obvious nonlinear vibration characteristics, and gives consideration to the precision and operability of high-speed dynamic balance.
The purpose of the invention is realized by the following technical scheme:
a high-speed dynamic balance mechanics resolving method based on machine learning comprises the following steps:
step 4, matching the generated data with the rotor test data of the same type, and performing hybrid training on the neural network, wherein the number of nodes of an input layer is the number of test points multiplied by the number of rotating speeds to be balanced, and the number of nodes of an output layer is the number of balanced surfaces multiplied by 2;
step 5, using a formula
V0+N(p)＝0
The calculated weight p, p being a K-order column vector, V ═ V0+ N (p) is the residual unbalance of the rotor, where V0The method is characterized in that the method is an M multiplied by N order column vector for measuring the initial vibration (without trial weight), wherein M is the number of measuring points, N is the number of rotating speeds to be balanced, N () is a trained network overall mapping function and is a mapping relation from a K order to an (M multiplied by N) order, and K is the number of balanced surfaces.
Programming according to the above calculation method and steps can realize low-cost operability of high-speed dynamic balance calculation. An operator only needs to operate according to the steps 1-5 when balancing is carried out for the first time, and only needs to call the trained network to implement the step 5 when the same type of rotor is balanced again, so that the advantages of large-scale production and large-data integration are brought into play.
Compared with the prior art, the technical scheme of the invention has the following beneficial effects:
1. the method provided by the invention overcomes the defects of the existing dynamic balance calculation method, the existing unbalance method (for example, the ABC method is based on the rigid assumption of a rotor, namely the working speed is far from reaching the critical speed, and the rigid assumption cannot be realized along with the increase of the working speed, although the influence coefficient method can be applied to the high-speed dynamic balance working condition, the influence coefficient method requires the forced response of the system to meet the superposition principle based on the linear relation of the input and the output of the system in the process of obtaining the influence coefficient, the system nonlinear performance is not obvious at low speed, the error caused by the approximate equivalent linearization is in an acceptable range, and the nonlinear factor is strongly expressed at high-speed dynamic balance, the system forced response no longer meets the superposition principle, and the result obtained by the linearization assumption will bring great error.) the calculation error introduced by the nonlinear factor cannot be solved, the method provided by the invention utilizes a neural network nonlinear mapping method to reduce the calculation error, can effectively reduce the residual unbalance and obtain a good balance effect.
2. The method provided by the invention can realize the non-trial-and-repeat dynamic balance technology in the process of carrying out dynamic balance on the rotors of the same type after the neural network is complete through integrating test data before and after the method is implemented and through training and updating the network.
3. The method provided by the invention is easy to update and maintain, has certain self-adaptive characteristic, has low technical admission threshold, and can be realized by only importing data and starting updating by non-full-time technicians.
Drawings
FIG. 1 is a schematic diagram of a neural network (fully connected);
FIG. 2 is a kinetic computational model constructed for generating training data;
fig. 3 is a structural diagram of the constructed neural network.
Detailed Description
The invention is described in further detail below with reference to the figures and specific examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention provides a constrained high-speed dynamic balance mechanics resolving method, which comprises the following steps:
Step 4, matching the generated data with the same type of rotor test data, and performing hybrid training on the neural network in the shape of the graph 1, wherein the number of nodes of an input layer is the number of test points multiplied by the number of rotating speeds to be balanced, and the number of nodes of an output layer is the number of balanced surfaces multiplied by 2; the process is used for training the network, the network type can be adjusted according to the rotor type, a full-connection type is generally adopted, the number of hidden layers is generally set to 3, a sigmoid function can be adopted as an activation function, an Adam optimizer is used, parameters are updated in a back propagation mode, and 500-inch and 800-inch are generally trained.
Step 5 using the formula
V0+N(p)＝0
The calculated weight p, p being a K-order column vector, V ═ V0+ N (p) is the residual unbalance of the rotor, where V0The method is characterized in that the method is an M multiplied by N order column vector for measuring the initial vibration (without trial weight), wherein M is the number of measuring points, N is the number of rotating speeds to be balanced, N () is a trained network overall mapping function and is a mapping relation from a K order to an (M multiplied by N) order, and K is the number of balanced surfaces.
Programming according to the above calculation method and steps can realize low-cost operability of high-speed dynamic balance calculation. An operator only needs to operate according to the steps 1-5 when balancing is carried out for the first time, and only needs to call the trained network to implement the step 5 when the same type of rotor is balanced again, so that the advantages of large-scale production and large-data integration are brought into play.
In particular, the method comprises the following steps of,
taking an axial flow compressor rotor of model AV50 as an example, which is an object of an embodiment of the present invention, parameters required for dynamic balance include:
the number of correction surfaces: when the rotor is produced and processed, a full-circle balancing groove is reserved on each of two sides of a shaft section, a screw hole is reserved in the middle of the shaft end, and balancing quality can be installed, so that the number of usable correcting surfaces is three;
number of measurement points: the method comprises the following steps of performing dynamic balance by using a 50-ton high-speed balancing machine generated by a Shenke company, wherein at present, the number of measuring points is two because two supporting swing frames are arranged at the measuring points;
correcting the rotating speed: the dynamic balance process adopts speed increase and reduction, the working rotating speed is about 5200 revolutions, the measured value is recorded once every 30 revolutions, and about 150 rotating speed measured values are recorded (the former 300 revolutions reserve low-speed dynamic balance and are not used);
the working rotating speed state before the dynamic balance correction is obtained through experiments, namely the measured value of the first swing frame exceeds 1mm/s (exceeds the standard).
The dynamic balance correction is carried out by the method, the dynamic model is shown in figure 2, and the structure of the trained network is shown in figure 3.
After balancing, the intensity value is greatly reduced and is only about 0.42 mm/s. The method has good effect.
The present invention is not limited to the above-described embodiments. The foregoing description of the specific embodiments is intended to describe and illustrate the technical solutions of the present invention, and the above specific embodiments are merely illustrative and not restrictive. Those skilled in the art can make many changes and modifications to the invention without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (1)
1. A high-speed dynamic balance mechanics resolving method based on machine learning is characterized by comprising the following steps:
step 1, determining positions of a supporting structure and a measuring point according to the type of a rotor to be balanced;
step 2, establishing a three-dimensional solid model of the rotor, and calling the existing rotor test data of the same type;
step 3, performing weight increasing and decreasing dynamic simulation by using the three-dimensional solid model to obtain a speed increasing-decreasing response amplitude value of the marked part corresponding to the actual test point;
step 4, matching the generated data with the rotor test data of the same type, and performing hybrid training on the neural network, wherein the number of nodes of an input layer is the number of test points multiplied by the number of rotating speeds to be balanced, and the number of nodes of an output layer is the number of balanced surfaces multiplied by 2;
step 5, using a formula
V0+N(p)＝0
The calculated weight p, p being a K-order column vector, V ═ V0+ N (p) is the residual unbalance of the rotor, where V0The method is characterized in that the method is an M multiplied by N order column vector for measuring the initial vibration (without trial weight), wherein M is the number of measuring points, N is the number of rotating speeds to be balanced, N () is a trained network overall mapping function and is a mapping relation from a K order to an (M multiplied by N) order, and K is the number of balanced surfaces.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010132666.8A CN111400962B (en) | 2020-02-29 | 2020-02-29 | High-speed dynamic balance mechanics resolving method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010132666.8A CN111400962B (en) | 2020-02-29 | 2020-02-29 | High-speed dynamic balance mechanics resolving method based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111400962A CN111400962A (en) | 2020-07-10 |
| CN111400962B true CN111400962B (en) | 2022-04-19 |
Family
ID=71430478
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010132666.8A Active CN111400962B (en) | 2020-02-29 | 2020-02-29 | High-speed dynamic balance mechanics resolving method based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111400962B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201716154U (en) * | 2010-05-13 | 2011-01-19 | 杭州杭发发电设备有限公司 | Turbogenerator rotor dynamic balance testing system |
| CN202332064U (en) * | 2011-11-27 | 2012-07-11 | 韶关市曲江明和设备修造有限公司 | Dynamic balance training device for motor rotor |
| CN104535262A (en) * | 2015-01-20 | 2015-04-22 | 湖南科技大学 | Complete machine trial-mass-free virtual dynamic balance method for turbine machinery N+1 supporting shafting |
-
2020
- 2020-02-29 CN CN202010132666.8A patent/CN111400962B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201716154U (en) * | 2010-05-13 | 2011-01-19 | 杭州杭发发电设备有限公司 | Turbogenerator rotor dynamic balance testing system |
| CN202332064U (en) * | 2011-11-27 | 2012-07-11 | 韶关市曲江明和设备修造有限公司 | Dynamic balance training device for motor rotor |
| CN104535262A (en) * | 2015-01-20 | 2015-04-22 | 湖南科技大学 | Complete machine trial-mass-free virtual dynamic balance method for turbine machinery N+1 supporting shafting |
Non-Patent Citations (1)
| Title |
|---|
| 考虑初始弯曲的无试重动平衡测试系统研究;罗义林;《罗义林》;20170315;C030-334 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111400962A (en) | 2020-07-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110532625A (en) | Aero-engine turbine disk-twin the modeling method of rotor-bearing system number | |
| CN104535262B (en) | Complete machine trial-mass-free virtual dynamic balance method for turbine machinery N+1 supporting shafting | |
| CN105893695A (en) | Through-flow clearance error compensation method for small-size turbine in cylinder combination state | |
| CN106354920B (en) | Base bearing axle sleeve strength analysis method and interference analysis method in Wind turbines | |
| WO2021237910A1 (en) | Active magnetic bearing controller construction method based on table lookup method | |
| CN113779697B (en) | Turbogenerator unit torsional vibration joint simulation method and device and storage medium | |
| CN111141518B (en) | Model-based asymmetric rotor bearing system unbalance identification method | |
| CN111400962B (en) | High-speed dynamic balance mechanics resolving method based on machine learning | |
| CN111898238B (en) | Constraint high-speed dynamic balance mechanics resolving method | |
| Huang et al. | Modeling method of the modal analysis for turbomolecular pump rotor blades | |
| CN109323853B (en) | Fan blade fan vibration detection device and fault detection prediction method | |
| CN106682292B (en) | A kind of leaf and root structure optimization method of dimensionality reduction simulated annealing | |
| Xu et al. | Dynamic model for hydro-turbine generator units based on a database method for guide bearings | |
| CN110334440A (en) | Rotor finite element model optimum design method neural network based | |
| CN112926698B (en) | Vibration prediction and assembly evaluation method for large-scale rotating equipment | |
| CN114136573B (en) | Method for calculating early warning amplitude of key component of hydroelectric generating set and related equipment | |
| CN115081330A (en) | Virtual and real data mirror image updating method for digital twin model of gear transmission system | |
| CN111740449A (en) | Method for analyzing damping characteristics of wind turbine group to synchronous generator set based on additional excitation signal injection method | |
| CN112149261B (en) | Rotating mechanical structure strength simulation system and method based on direct analog calculation | |
| Sun et al. | Intelligent Detection Method for Offline Dynamic Balance of Flexible Rotor | |
| Ren et al. | An investigation into the behavior of intelligent fault diagnostic models under imbalanced data | |
| Niu et al. | Research on Modal Analysis of Rotor Shaft System of Hydro-generator | |
| CN112765771B (en) | Real-time monitoring algorithm for vibration data of aircraft engine in transition state | |
| CN116183231B (en) | Bearing fault diagnosis method based on digital twin | |
| CN117709190A (en) | Rolling bearing digital twin model construction and model parameter optimization method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |