CN109492345A - A kind of turbine blade high-cycle fatigue life method based on SENet - Google Patents
A kind of turbine blade high-cycle fatigue life method based on SENet Download PDFInfo
- Publication number
- CN109492345A CN109492345A CN201910024307.8A CN201910024307A CN109492345A CN 109492345 A CN109492345 A CN 109492345A CN 201910024307 A CN201910024307 A CN 201910024307A CN 109492345 A CN109492345 A CN 109492345A
- Authority
- CN
- China
- Prior art keywords
- turbine blade
- cycle fatigue
- senet
- fatigue life
- blade
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- 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
- G06F2119/04—Ageing analysis or optimisation against ageing
Landscapes
- Physics & Mathematics (AREA)
- Geometry (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computational Mathematics (AREA)
- Control Of Turbines (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The turbine blade high-cycle fatigue life method based on SENet that the invention discloses a kind of, comprising steps of the turbine blade fault-signal of high cycle fatigue failure occurs for 1, acquisition；2, the forced vibration of turbine blade and self-excited vibration clock signal when calculating operation using SFI and FEM method are converted to frequency-region signal by FFT, calculate the first six rank vibratory response when operation；3, Calculation of Vibration Response is carried out to all sampling blades, by data normalization, divides training set and verifying collection；4, SE block is integrated into ResNet neural network, forms the SENet network that SE-ResNet is framework and training；5, high-cycle fatigue life is carried out for the turbine blade of actual motion；6, increase new data, carry out new network training for different materials.The present invention can be realized accurately high-cycle fatigue life, avoid complicated life prediction Analysis on Mechanism, and the artificial of life prediction and experiment testing cost is greatly reduced, has important engineering significance and wide application prospect.
Description
Technical field
The invention belongs to turbomachinery blade technology fields, and in particular to a kind of high week of the turbine blade based on SENet is tired
Labor life-span prediction method.
Background technique
Turbine blade is the core component of steam turbine and gas turbine.Its working environment is severe, and structure is complicated, while by
Centrifugal load, aerodynamic loading, thermal force, vibration stress etc. are one of steam turbine or the multiple part of gas turbine fracture defect.High week
The stress level of fatigue is lower than elastic limit, and without apparent macroscopic view plastic deformation, linear relationship is presented in stress and strain.By gas
The fracture of turbine blade high cycle fatigue caused by stream motivates will seriously affect the reliability and working efficiency of steam turbine or gas turbine.
Therefore, establishing high-cycle fatigue life system of the blade under air-flow excitation has important engineering significance.
The high-cycle fatigue life for steam turbine and gas turbine blades is generallyd use at present is answered based on local stress
The improved method of political reform, concrete analysis are that the static stress and vibration stress of blade are calculated using solid finite element model, together
When consider the various operating conditions of blade operation and influence the principal element of leaf longevity, the fatigue life of quantitative blade.This method
The ess-strain of blade part is directly simply established corresponding relationship between standard smooth specimen Fatigue Property Curve, it is believed that
As long as maximum A LOCAL STRESS-STRAIN is identical, fatigue life is identical, is predicted according to the life curve that fatigue experiment obtains corresponding
High-Cycle Fatigue Life Prediction, prediction result error are big, it is difficult to meet engineering precision.
In recent years, the fast development of deep learning algorithm is that many engineering problems lay a solid foundation.SENet is logical
It crosses this minor structure by SE block to be inserted into other convolutional neural networks models, makes network according to loss learning characteristic weight, thus
So that validity feature weight is bigger, reduce invalid feature, facilitates the essence for reacting physical phenomenon.Directly adopt data mining
Method predicted fatigue life does not need engineers and technicians and grasps esoteric life prediction mechanism and relevant knowledge abundant；Together
When based on the Life Prediction Model that deep learning algorithm obtains have the characteristics that precision is high, speed is fast, transportable property is strong, especially fit
It closes and is applied to engineer application.
Summary of the invention
In view of the above-mentioned deficiencies in the prior art, it is an object of the present invention to provide a kind of high week of the turbine blade based on SENet
Prediction method for fatigue life is mainly used in high-cycle fatigue life of the turbine blade under air-flow excitation, can be realized
Accurately high-cycle fatigue life, and complicated life prediction Analysis on Mechanism is avoided, while life prediction is greatly reduced
Cost of labor and experiment testing cost, have important engineering significance and wide application prospect.
The present invention adopts the following technical scheme that realize:
A kind of turbine blade high-cycle fatigue life method based on SENet, comprising the following steps:
The fault-signal of the same material turbine blade of high cycle fatigue failure occurs for the first step, acquisition, and record corresponds at this time
Turbine blade service life [Yi], wherein i indicates i-th of turbine blade for high cycle fatigue failure occur；
Second step, the forced vibration generated when analyzing turbine blade by periodical unsteady flow exciting force, to turbine
The fluid domain and solid domain partition structure grid of blade load revolving speed and inlet and outlet state boundaries condition, solid using one-way flow
The periodical unsteady flow that the turbine blade that the method for coupling calculates generation high cycle fatigue failure is subject under job state swashs
Vibration power, obtains the clock signal for the periodical unsteady flow exciting force that turbine blade is subject to, the clock signal conversion that will acquire
For the frequency-region signal [A of Airflow Exciting-Vibration Forcei]m×n, wherein unidirectional fluid structurecoupling hereinafter referred to as SFI；
Third step analyzes the vibratory response of turbine blade self-excited vibration, using generation high cycle fatigue failure in second step
The solid domain structured grid of turbine blade, and turbine blade Temperature Distribution obtained in SFI calculating is extracted, using finite element meter
Calculation method loads revolving speed and temperature boundary condition, calculates the self-excited vibration clock signal of turbine blade and be translated into frequency domain
Signal [Si]m×n, [V is obtained by the linear superposition to forced vibration amplitude and self-excited vibration amplitudei]m×n=[Ai]m×n+
[Si]m×n, by [Vi]m×nIt is applied to each node, calculates its vibration displacement data, that is, can extract the turbine blade under job state
The first six rank vibratory response [Di]m×n×6, wherein finite element method hereinafter referred to as FEM；
4th step carries out the calculating of second step and third step to i blade for extracting lifetime data, and to [Di]m×n×6
Data be normalized, obtainIt is according to training set/verifying collection=4.0 ratio cut partition training setVerifying collectsAnd upset training set data at random, as neural network
Input；
5th step constructs SENet, each training data signalPass through the main body knot of SENet
Structure eventually passes through a full articulamentum and obtains the regression forecasting value of turbine blade high-Cycle Fatigue Life Prediction, in conjunction with turbine blade high week
Variable learning rate is arranged by synchronous SGD optimizer to train network in this practical problem of Prediction method for fatigue life；
6th step exports the reality by the SENet trained in the high-cycle fatigue life of practical turbine blade
The high-cycle fatigue life value of turbine blade under operating condition.
A further improvement of the present invention lies in that further including the following contents in the 6th step:
When needing to increase new training set data, using the SENet trained as pre-training model, same setting is variable
Learning rate trains new network；For the turbine blade of other materials, new using the method training of the first step to the 5th step
SENet network, to adapt to turbine blade material not of the same race.
A further improvement of the present invention lies in that the turbine blade failure letter of high cycle fatigue failure occurs for acquisition in the first step
Number, this signal derives from large-scale axial-flow turbine group movable vane piece or high-temperature fuel gas wheel movable blade, or derives from small-sized space division
The radial-inward-flow turbine blade of equipment and ORC working medium；For the turbine blade of same material, record occurs corresponding when high cycle fatigue failure
Service life [Yi]。
A further improvement of the present invention lies in that occur the turbine blade of high cycle fatigue failure for i-th in second step,
Unsteady flow exciting force caused by stator wake stream is considered, to analyze the frequency domain distribution situation of its forced vibration amplitude；Tool
Body are as follows: its fluid domain and solid domain are divided into hexahedron structure grid first, wherein blade grid is along blade profile in-plane
Number of nodes is m, is n along the high direction number of nodes of leaf；Secondly load revolving speed and inlet and outlet state boundaries condition, in ANSYS CFX
Primary unsteady FSI is carried out to calculate, it is a cycle that movable vane, which turns over a stator blade channel, and each period selects 20 time steps,
Calculate the unsteady flow exciting force in 6 periods, i.e. 120 time steps；It again passes by unsteady FSI to calculate, obtains the turbine
The Airflow Exciting-Vibration Force clock signal of m × n node of blade；Finally, by the method for Fast Fourier Transform (FFT), the timing that will acquire
Signal is converted into frequency-region signal [Ai]m×n。
A further improvement of the present invention lies in that the turbine blade of high cycle fatigue failure occurs to i-th, adopts in third step
With the solid domain structured grid in second step, the Temperature Distribution of turbine blade in FSI calculated result is extracted；Turbine blade is existed
Revolving speed and temperature boundary condition under job state load in ANSYS APDL, equally turn over a stator blade channel with movable vane and are
A cycle calculates the self-excited vibration clock signal of turbine blade in 6 periods, finally by FFT method by turbine blade m ×
The vibration amplitude clock signal of n node is converted into frequency-region signal [Si]m×n；To forced vibration amplitude and self-excited vibration amplitude
Linear superposition obtains [Vi]m×n=[Ai]m×n+[Si]m×n, by [Vi]m×nM × n node being applied on turbine blade,
Harmonic responding analysis is carried out in ANSYS APDL, calculates its vibration displacement data, before extracting the turbine blade under job state
Six rank vibratory responses, i.e., the matrix of 6 m × n sizes are denoted as [Di]m×n×6。
A further improvement of the present invention lies in that in the 4th step, to i blade for extracting lifetime data carry out second step with
The calculating of the third step, [D that will be extractedi]m×n×6Data normalization is carried out, wherein method for normalizing is as follows:
Wherein, [Di]m×n×6Indicate original vibratory response data, Max, Min, Mean are respectively indicated to [Di]m×n×6Take maximum
Value, minimum value and average value；Max_value indicates the upper limit of range after normalization, is herein the pixel upper limit value of grayscale image
255；Min_value indicates the lower limit of range after normalization, is herein the pixel lower limit value 0 of grayscale image；By normalization operation,
The normalization numerical value in input six channels of neural network can be obtained
A further improvement of the present invention lies in that SE block is integrated into ResNet neural network in the 5th step, SE- is formed
ResNet is the SENet of framework, this network includes the structure of SE block and the main structure of ResNet；Each training data letter
NumberBy the main structure of SENet, ReLU activation primitive and sigmoid activation primitive are used in SE block,
And two layers of door machine system connected and composed entirely is used；In conjunction with this practical problem of turbine blade high-cycle fatigue life method,
Small lot sampling is carried out based on data balancing strategy, training process is optimized using synchronous SGD gradient descent algorithm, initial to learn
Habit rate is set as 0.5, and every 25 step learning rate is decayed 10 times thereafter.
A further improvement of the present invention lies in that in the high-cycle fatigue life of practical turbine blade, being adopted in the 6th step
With the method for the first step to third step, carry out FSI and FEM and calculate, input the blade forced vibration after normalization and from
The overall response of excited vibrationThe high cycle fatigue of turbine blade under the actual operating mode is exported by the SENet trained
Life prediction value.
Compared with prior art, the present invention has following beneficial technical effect:
The present invention passes through the advantage for successfully integrating a variety of prior arts, for the high-cycle fatigue life side of turbine blade
Method has carried out reformed AHP, proposes a kind of turbine blade high-cycle fatigue life method based on SENet.Traditional height week
Prediction method for fatigue life simplifiedly thinks that, as long as maximum A LOCAL STRESS-STRAIN is identical, fatigue life is identical, according to tired real
The life curve tested predicts corresponding high-Cycle Fatigue Life Prediction, therefore has biggish prediction error, is unable to satisfy industrial life
Production demand.The present invention is based on SENet, using blade material, the operating condition, vibration for generating high cycle fatigue failure in actual industrial
Response data is trained neural network, to realize to the high-cycle fatigue life of turbine blade, substantially increases pre-
Precision is surveyed,
Specifically, SE block is integrated into ResNet neural network by the present invention, and forming SE-ResNet is framework
SENet, the main structure of structure and ResNet comprising SE block.The depth residual error network ResNet of use solves convolutional Neural
This minor structure of SE block is inserted into ResNet convolutional neural networks model by the problem of network is degenerated when network depth is promoted
In, so that network is gone learning characteristic weight according to loss, so that validity feature weight is bigger, reduces invalid feature, in image
There is significant advantage in processing.By SENet this Application of Neural Network in turbine blade high-cycle fatigue life, work is not needed
Cheng personnel grasp complicated life prediction mechanism and relevant knowledge abundant, and the method for directlying adopt data mining realizes that high week is tired
The prediction in labor service life.In general, have that precision is high, speed is fast, can based on the Life Prediction Model that deep learning algorithm obtains
The strong feature of migration, is extremely applicable to the turbine blade of industry park plan.
In conclusion the present invention is directed to the deficiency of existing life-span prediction method, a kind of turbine leaf based on SENet is established
Piece high-cycle fatigue life method, can be realized accurately high-cycle fatigue life, and avoid complicated life prediction
Analysis on Mechanism, while the artificial of life prediction and experiment testing cost is greatly reduced, with important engineering significance and wide
Application prospect.
Detailed description of the invention
Fig. 1 is a kind of overview flow chart of the turbine blade high-cycle fatigue life method based on SENet of the present invention.
Fig. 2 is m × n grid node matrix schematic diagram of certain example turbine blade.
Fig. 3 is the unsteady flow exciting force clock signal figure of certain node on certain example turbine blade grid
Fig. 4 is that the unsteady flow exciting force clock signal of certain node on certain example turbine blade grid passes through FFT transform
Obtained frequency domain signal diagrams.
Fig. 5 is certain grayscale image of example turbine blade single order vibratory response after normalized.
Specific embodiment
The present invention will be further explained below with reference to the attached drawings.
Referring to Fig. 1, a kind of turbine blade high-cycle fatigue life method based on SENet provided by the invention, including
Following steps:
1) fault-signal of the same material turbine blade of high cycle fatigue failure occurs for acquisition, records corresponding turbine at this time
Leaf longevity [Yi], wherein i indicates i-th of turbine blade for high cycle fatigue failure occur.
2) forced vibration generated when analyzing turbine blade by periodical unsteady flow exciting force.To turbine blade
Fluid domain and solid domain partition structure grid.Revolving speed and inlet and outlet state boundaries condition are loaded, using unidirectional fluid structurecoupling
(SFI) the periodical unsteady flow that the turbine blade that method calculates generation high cycle fatigue failure is subject under job state swashs
Vibration power, obtains the clock signal for the periodical unsteady flow exciting force that turbine blade is subject to, the clock signal conversion that will acquire
For the frequency-region signal [A of Airflow Exciting-Vibration Forcei]m×n。
3) vibratory response of turbine blade self-excited vibration is analyzed.Using the turbine leaf that high cycle fatigue failure occurs in second step
The solid domain structured grid of piece, and extract turbine blade Temperature Distribution obtained in SFI calculating.Using FEM calculation
(FEM) method loads revolving speed and temperature boundary condition, calculates the self-excited vibration clock signal of turbine blade and be translated into
Frequency-region signal [Si]m×n.[V is obtained by the linear superposition to forced vibration amplitude and self-excited vibration amplitudei]m×n=[Ai]m×n+
[Si]m×n, by [Vi]m×nIt is applied to each node, calculates its vibration displacement data, that is, can extract the turbine blade under job state
The first six rank vibratory response [Di]m×n×6。
4) calculating of second step and third step is carried out to i blade for extracting lifetime data, and to [Di]m×n×6Number
According to being normalized, obtainIt is according to training set/verifying collection=4.0 ratio cut partition training setVerifying collectsAnd upset training set data at random, as neural network
Input.
5) SENet is constructed.Each training data signalBy the main structure of SENet, finally
The regression forecasting value of turbine blade high-Cycle Fatigue Life Prediction is obtained by a full articulamentum.In conjunction with turbine blade high-Cycle Fatigue Life Prediction
Variable learning rate is arranged by synchronous SGD optimizer to train network in this practical problem of prediction technique.
6) in the high-cycle fatigue life of practical turbine blade, which is exported by the SENet trained
The high-cycle fatigue life value of turbine blade under operating condition.When needing to increase new training set data, can will train
SENet is same that variable learning rate is arranged to train new network as pre-training model.For the turbine blade of other materials, adopt
New SENet network is trained with the method for step 1 to five, to adapt to turbine blade material not of the same race.
Referring to fig. 2, with a certain turbine blade as an example, considering that the unsteady flow of stator wake stream under working condition swashs
Power of shaking influences, its fluid domain and solid domain is divided into hexahedron structure grid first, wherein blade solid domain grid is along leaf
Type in-plane number of nodes is m, is n along the high direction number of nodes of leaf.
Referring to Fig. 3 and Fig. 4, by taking forced vibration is analyzed as an example, revolving speed and inlet and outlet state boundaries condition are loaded, in ANSYS
It carries out primary unsteady FSI in CFX to calculate, it is a cycle that movable vane, which turns over a stator blade channel, when each period selects 20
Spacer step calculates the unsteady flow exciting force in 6 periods, i.e. 120 time steps.It is calculated by unsteady FSI, it is available to be somebody's turn to do
The Airflow Exciting-Vibration Force clock signal of m × n node of turbine blade.Finally, pass through the method for Fast Fourier Transform (FFT) (FFT), it will
The clock signal of acquisition is converted into frequency-region signal [Ai]m×n.The Airflow Exciting-Vibration Force clock signal result of a certain node is as shown in Figure 3.
By the method for Fast Fourier Transform (FFT) (FFT), the clock signal that can be will acquire is converted into the frequency-region signal of forced vibration
[Ai], as a result as shown in figure 4, the corresponding frequency of the Airflow Exciting-Vibration Force of amplitude maximum is 3750Hz at this time.
Referring to Fig. 5, the single order vibration of each grid node of example turbine blade is illustrated by taking m=40, n=18 as an example, in figure
Respond [Di]m×n×1After normalization, it is converted into grayscale imageThe input of one of six channels as neural network
The specific method of SENet.
Claims (8)
1. a kind of turbine blade high-cycle fatigue life method based on SENet, which comprises the following steps:
The first step, acquisition occur the fault-signal of the same material turbine blade of high cycle fatigue failure, record corresponding at this time
Flat blade service life [Yi], wherein i indicates i-th of turbine blade for high cycle fatigue failure occur；
Second step, the forced vibration generated when analyzing turbine blade by periodical unsteady flow exciting force, to turbine blade
Fluid domain and solid domain partition structure grid, load revolving speed and inlet and outlet state boundaries condition, using unidirectional fluid structurecoupling
Method calculate the periodical unsteady flow exciting force that is subject under job state of turbine blade that high cycle fatigue failure occurs,
The clock signal for the periodical unsteady flow exciting force that turbine blade is subject to is obtained, the clock signal that will acquire is converted into air-flow
Frequency-region signal [the A of exciting forcei]m×n, wherein unidirectional fluid structurecoupling hereinafter referred to as SFI；
Third step analyzes the vibratory response of turbine blade self-excited vibration, using the turbine that high cycle fatigue failure occurs in second step
The solid domain structured grid of blade, and turbine blade Temperature Distribution obtained in SFI calculating is extracted, using FEM calculation side
Method loads revolving speed and temperature boundary condition, calculates the self-excited vibration clock signal of turbine blade and be translated into frequency-region signal
[Si]m×n, [V is obtained by the linear superposition to forced vibration amplitude and self-excited vibration amplitudei]m×n=[Ai]m×n+[Si]m×n, will
[Vi]m×nIt is applied to each node, calculates its vibration displacement data, that is, can extract the first six rank of the turbine blade under job state
Vibratory response [Di]m×n×6, wherein finite element method hereinafter referred to as FEM；
4th step carries out the calculating of second step and third step to i blade for extracting lifetime data, and to [Di]m×n×6Number
According to being normalized, obtainIt is according to training set/verifying collection=4.0 ratio cut partition training setVerifying collectsAnd upset training set data at random, as neural network
Input；
5th step constructs SENet, each training data signalBy the main structure of SENet, most
The regression forecasting value of turbine blade high-Cycle Fatigue Life Prediction is obtained by a full articulamentum eventually, in conjunction with the turbine blade high cycle fatigue longevity
This practical problem of prediction technique is ordered, by synchronous SGD optimizer, variable learning rate is set to train network；
6th step exports the actual motion by the SENet trained in the high-cycle fatigue life of practical turbine blade
The high-cycle fatigue life value of turbine blade under operating condition.
2. a kind of turbine blade high-cycle fatigue life method based on SENet according to claim 1, feature exist
In further including the following contents in the 6th step:
When needing to increase new training set data, using the SENet trained as pre-training model, equally setting variable learning
Rate trains new network；For the turbine blade of other materials, using the new SENet net of the method training of the first step to the 5th step
Network, to adapt to turbine blade material not of the same race.
3. a kind of turbine blade high-cycle fatigue life method based on SENet according to claim 1 or 2, special
Sign is, in the first step, the turbine blade fault-signal of high cycle fatigue failure occurs for acquisition, this signal derives from big profile shaft stream vapour
Unit movable vane piece or high-temperature fuel gas wheel movable blade are taken turns, or from packaged air separation plant and the radial-inward-flow turbine leaf of ORC working medium
Piece；For the turbine blade of same material, corresponding service life [Y when high cycle fatigue failure occurs for recordi]。
4. a kind of turbine blade high-cycle fatigue life method based on SENet according to claim 3, feature exist
In for the turbine blade for occurring high cycle fatigue failure for i-th, considering unsteady gas caused by stator wake stream in second step
Exciting force is flowed, to analyze the frequency domain distribution situation of its forced vibration amplitude；Specifically: its fluid domain and solid domain are drawn first
It is divided into hexahedron structure grid, wherein blade grid is m along blade profile in-plane number of nodes, is n along the high direction number of nodes of leaf；
Secondly load revolving speed and inlet and outlet state boundaries condition carry out primary unsteady FSI in ANSYS CFX and calculate, and movable vane turns over
One stator blade channel is a cycle, and each period selects 20 time steps, calculates the unsteady flow exciting force in 6 periods,
That is 120 time steps；It again passes by unsteady FSI to calculate, obtains the Airflow Exciting-Vibration Force timing of m × n node of the turbine blade
Signal；Finally, the clock signal that will acquire is converted into frequency-region signal [A by the method for Fast Fourier Transform (FFT)i]m×n。
5. a kind of turbine blade high-cycle fatigue life method based on SENet according to claim 4, feature exist
In, in third step, i-th occurs the turbine blade of high cycle fatigue failure, using the solid domain structured grid in second step,
Extract the Temperature Distribution of turbine blade in FSI calculated result；By revolving speed and temperature boundary item of the turbine blade under job state
Part loads in ANSYS APDL, equally turns over a stator blade channel as a cycle using movable vane, calculates turbine leaf in 6 periods
The self-excited vibration clock signal of piece turns the vibration amplitude clock signal of m × n node of turbine blade finally by FFT method
Turn to frequency-region signal [Si]m×n；[V is obtained to the linear superposition of forced vibration amplitude and self-excited vibration amplitudei]m×n=[Ai]m×n+
[Si]m×n, by [Vi]m×nM × n node being applied on turbine blade carries out harmonic responding analysis in ANSYS APDL, calculates
Its vibration displacement data extracts the first six the rank vibratory response of the turbine blade under job state, i.e., the square of 6 m × n sizes
Battle array, is denoted as [Di]m×n×6。
6. a kind of turbine blade high-cycle fatigue life method based on SENet according to claim 5, feature exist
In the calculating of second step and third step being carried out to i blade for extracting lifetime data, by what is extracted in the 4th step
[Di]m×n×6Data normalization is carried out, wherein method for normalizing is as follows:
Wherein, [Di]m×n×6Indicate original vibratory response data, Max, Min, Mean are respectively indicated to [Di]m×n×6Be maximized,
Minimum value and average value；Max_value indicates the upper limit of range after normalization, is herein the pixel upper limit value 255 of grayscale image；
Min_value indicates the lower limit of range after normalization, is herein the pixel lower limit value 0 of grayscale image；By normalization operation
Obtain the normalization numerical value in input six channels of neural network
7. a kind of turbine blade high-cycle fatigue life method based on SENet according to claim 6, feature exist
In, in the 5th step, SE block is integrated into ResNet neural network, formed SE-ResNet be framework SENet, this network
The main structure of structure and ResNet comprising SE block；Each training data signalPass through SENet's
Main structure uses ReLU activation primitive and sigmoid activation primitive in SE block, and has used two layers of door machine connected and composed entirely
System；In conjunction with this practical problem of turbine blade high-cycle fatigue life method, small lot is carried out based on data balancing strategy and is adopted
Sample, training process are optimized using synchronous SGD gradient descent algorithm, and initial learning rate is set as 0.5, thereafter every 25 step study
Rate decays 10 times.
8. a kind of turbine blade high-cycle fatigue life method based on SENet according to claim 7, feature exist
In in the 6th step, in the high-cycle fatigue life of practical turbine blade, using the method for the first step to third step, progress
FSI and FEM is calculated, and inputs the overall response of the blade forced vibration and self-excited vibration after normalizationBy
Trained SENet exports the high-cycle fatigue life value of turbine blade under the actual operating mode.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910024307.8A CN109492345B (en) | 2019-01-10 | 2019-01-10 | SENET-based turbine blade high cycle fatigue life prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910024307.8A CN109492345B (en) | 2019-01-10 | 2019-01-10 | SENET-based turbine blade high cycle fatigue life prediction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109492345A true CN109492345A (en) | 2019-03-19 |
| CN109492345B CN109492345B (en) | 2020-11-10 |
Family
ID=65714389
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910024307.8A Active CN109492345B (en) | 2019-01-10 | 2019-01-10 | SENET-based turbine blade high cycle fatigue life prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109492345B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110046409A (en) * | 2019-03-29 | 2019-07-23 | 西安交通大学 | A kind of components of steam turbine health state evaluation method based on ResNet |
| CN110378870A (en) * | 2019-06-06 | 2019-10-25 | 西安交通大学 | A kind of turbine blade erosion degree method of discrimination based on ResNet-GRU network |
| CN111714111A (en) * | 2019-03-22 | 2020-09-29 | 安徽华米信息科技有限公司 | Electrocardio data processing method and device, wearable device and storage medium |
| CN112559305A (en) * | 2020-11-10 | 2021-03-26 | 深圳前海有电物联科技有限公司 | Method, device and equipment for predicting performance of key module in uninterruptible power supply system |
| CN113033105A (en) * | 2021-04-06 | 2021-06-25 | 东北大学 | Steel material fatigue performance prediction method based on mechanical theory guide transfer learning |
| CN113361048A (en) * | 2021-08-11 | 2021-09-07 | 中国联合重型燃气轮机技术有限公司 | Reliability prediction method and device for gas turbine blade and electronic equipment |
| CN113834621A (en) * | 2020-06-08 | 2021-12-24 | 中国航发商用航空发动机有限责任公司 | Blade vibration fatigue test method and system, control device and storage medium |
| CN114429060A (en) * | 2021-12-02 | 2022-05-03 | 中国兵器科学研究院宁波分院 | Method for assessing structure dislocation failure and service life prediction in fatigue vibration |
| CN114741922A (en) * | 2022-04-11 | 2022-07-12 | 西安交通大学 | Turbine blade creep-fatigue life prediction method based on Attention mechanism |
| CN118278225A (en) * | 2024-06-03 | 2024-07-02 | 西安航天动力研究所 | Material vibration S-N characteristic identification method and device based on structural vibration test |
| US20240257079A1 (en) * | 2023-01-31 | 2024-08-01 | Vestas Wind Systems A/S | System and methods for using machine learning to make intelligent recycling decisions |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030202883A1 (en) * | 2002-04-26 | 2003-10-30 | Davis Gary A. | Turbine blade assembly with stranded wire cable dampers |
| CN102588012A (en) * | 2012-02-28 | 2012-07-18 | 上海发电设备成套设计研究院 | Method and device for predicting and monitoring high-cycle fatigue life of turbine welded rotor |
| CN102589867A (en) * | 2012-02-28 | 2012-07-18 | 上海发电设备成套设计研究院 | Method and device for monitoring high-cycle fatigue life of steam turbine integral rotor |
| CN103745132A (en) * | 2014-01-23 | 2014-04-23 | 中国北方发动机研究所(天津) | Evaluation method for reliable life of blade vibration fatigue of gas compressor impeller of supercharger |
| CN106202630A (en) * | 2016-06-28 | 2016-12-07 | 四川大学 | Turbine rotor super high cycle fatigue fatigue strength and estimating method for fatigue life |
| CN106528932A (en) * | 2016-10-09 | 2017-03-22 | 西安交通大学 | Vibration stress numerical analysis method for turbomachinery blades |
| CN107844668A (en) * | 2017-11-29 | 2018-03-27 | 扬州大学 | A kind of analysis method of the axial-flow pump fatigue reliability based on pump installation |
| CN108510001A (en) * | 2018-04-04 | 2018-09-07 | 北京交通大学 | A kind of blade of wind-driven generator defect classification method and its categorizing system |
| CN108829917A (en) * | 2018-04-25 | 2018-11-16 | 内蒙古工业大学 | A kind of stalk throws impeller Prediction method for fatigue life |
| CN109117788A (en) * | 2018-08-10 | 2019-01-01 | 重庆大学 | A kind of public transport compartment crowding detection method merging ResNet and LSTM |
-
2019
- 2019-01-10 CN CN201910024307.8A patent/CN109492345B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030202883A1 (en) * | 2002-04-26 | 2003-10-30 | Davis Gary A. | Turbine blade assembly with stranded wire cable dampers |
| CN102588012A (en) * | 2012-02-28 | 2012-07-18 | 上海发电设备成套设计研究院 | Method and device for predicting and monitoring high-cycle fatigue life of turbine welded rotor |
| CN102589867A (en) * | 2012-02-28 | 2012-07-18 | 上海发电设备成套设计研究院 | Method and device for monitoring high-cycle fatigue life of steam turbine integral rotor |
| CN103745132A (en) * | 2014-01-23 | 2014-04-23 | 中国北方发动机研究所(天津) | Evaluation method for reliable life of blade vibration fatigue of gas compressor impeller of supercharger |
| CN106202630A (en) * | 2016-06-28 | 2016-12-07 | 四川大学 | Turbine rotor super high cycle fatigue fatigue strength and estimating method for fatigue life |
| CN106528932A (en) * | 2016-10-09 | 2017-03-22 | 西安交通大学 | Vibration stress numerical analysis method for turbomachinery blades |
| CN107844668A (en) * | 2017-11-29 | 2018-03-27 | 扬州大学 | A kind of analysis method of the axial-flow pump fatigue reliability based on pump installation |
| CN108510001A (en) * | 2018-04-04 | 2018-09-07 | 北京交通大学 | A kind of blade of wind-driven generator defect classification method and its categorizing system |
| CN108829917A (en) * | 2018-04-25 | 2018-11-16 | 内蒙古工业大学 | A kind of stalk throws impeller Prediction method for fatigue life |
| CN109117788A (en) * | 2018-08-10 | 2019-01-01 | 重庆大学 | A kind of public transport compartment crowding detection method merging ResNet and LSTM |
Non-Patent Citations (3)
| Title |
|---|
| 杨鹏: "基于流固耦合的轴流风机叶轮静动特性和疲劳研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 谢永慧 等: "基于热流固耦合分析的重型燃气轮机透平高压叶片寿命研究", 《热力透平》 * |
| 谢永慧 等: "汽轮机叶片疲劳寿命预测方法的研究", 《西安交通大学学报》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111714111A (en) * | 2019-03-22 | 2020-09-29 | 安徽华米信息科技有限公司 | Electrocardio data processing method and device, wearable device and storage medium |
| CN110046409A (en) * | 2019-03-29 | 2019-07-23 | 西安交通大学 | A kind of components of steam turbine health state evaluation method based on ResNet |
| CN110378870A (en) * | 2019-06-06 | 2019-10-25 | 西安交通大学 | A kind of turbine blade erosion degree method of discrimination based on ResNet-GRU network |
| CN110378870B (en) * | 2019-06-06 | 2021-04-20 | 西安交通大学 | ResNet-GRU network-based turbine blade water erosion degree judging method |
| CN113834621A (en) * | 2020-06-08 | 2021-12-24 | 中国航发商用航空发动机有限责任公司 | Blade vibration fatigue test method and system, control device and storage medium |
| CN113834621B (en) * | 2020-06-08 | 2024-03-29 | 中国航发商用航空发动机有限责任公司 | Blade vibration fatigue test method and system, control device and storage medium |
| CN112559305A (en) * | 2020-11-10 | 2021-03-26 | 深圳前海有电物联科技有限公司 | Method, device and equipment for predicting performance of key module in uninterruptible power supply system |
| CN113033105B (en) * | 2021-04-06 | 2023-09-19 | 东北大学 | Method for predicting fatigue performance of steel material based on mechanical theory guiding transfer learning |
| CN113033105A (en) * | 2021-04-06 | 2021-06-25 | 东北大学 | Steel material fatigue performance prediction method based on mechanical theory guide transfer learning |
| CN113361048A (en) * | 2021-08-11 | 2021-09-07 | 中国联合重型燃气轮机技术有限公司 | Reliability prediction method and device for gas turbine blade and electronic equipment |
| CN114429060A (en) * | 2021-12-02 | 2022-05-03 | 中国兵器科学研究院宁波分院 | Method for assessing structure dislocation failure and service life prediction in fatigue vibration |
| CN114429060B (en) * | 2021-12-02 | 2022-12-27 | 中国兵器科学研究院宁波分院 | Method for examining structure dislocation failure and service life prediction in fatigue vibration |
| CN114741922B (en) * | 2022-04-11 | 2024-02-23 | 西安交通大学 | Turbine blade creep-fatigue life prediction method based on Attention mechanism |
| CN114741922A (en) * | 2022-04-11 | 2022-07-12 | 西安交通大学 | Turbine blade creep-fatigue life prediction method based on Attention mechanism |
| US20240257079A1 (en) * | 2023-01-31 | 2024-08-01 | Vestas Wind Systems A/S | System and methods for using machine learning to make intelligent recycling decisions |
| CN118278225A (en) * | 2024-06-03 | 2024-07-02 | 西安航天动力研究所 | Material vibration S-N characteristic identification method and device based on structural vibration test |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109492345B (en) | 2020-11-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109492345A (en) | A kind of turbine blade high-cycle fatigue life method based on SENet | |
| CN112149316B (en) | Aero-engine residual life prediction method based on improved CNN model | |
| Rossi et al. | A general methodology for performance prediction of pumps-as-turbines using Artificial Neural Networks | |
| CN106404403B (en) | Method and system for analysis of a turbomachine | |
| CN107223206A (en) | Testing fatigue | |
| CN109472241A (en) | Combustion engine bearing remaining life prediction technique based on support vector regression | |
| CN106682376A (en) | Whole-process steam turbine modeling and recognizing method of actual characteristics of parameters changing with working conditions | |
| CN102135021B (en) | Method for predicting shaft power of industrial extraction condensing steam turbine | |
| CN105352586A (en) | Method used for realizing blade fatigue crack detection through resonance frequency detection | |
| CN110046409A (en) | A kind of components of steam turbine health state evaluation method based on ResNet | |
| CN109029884A (en) | A kind of method of vehicle cantilever structural member vibrating fatigue analysis | |
| CN103268388A (en) | Method for analyzing reliability of turbine blade disk system of aircraft engine | |
| CN109359668A (en) | A kind of concurrent diagnostic method of aero-engine multiple faults | |
| CN111160457A (en) | Turboshaft engine fault detection method based on soft class extreme learning machine | |
| CN114741922B (en) | Turbine blade creep-fatigue life prediction method based on Attention mechanism | |
| CN116090191A (en) | Simulation method and system of offshore wind turbine under comprehensive airflow factors | |
| CN114548539B (en) | Turbine energy consumption prediction method, device, equipment and medium for circulating water system compressor | |
| CN107545112A (en) | Complex equipment Performance Evaluation and Forecasting Methodology of the multi-source without label data machine learning | |
| CN115163424A (en) | Wind turbine generator gearbox oil temperature fault detection method and system based on neural network | |
| Gu et al. | Study on intelligent diagnosis of rotor fault causes with the PSO‐xgboost algorithm | |
| Bruni et al. | Application of Machine Learning to Forced Response Predictions of an Industrial Axial Compressor Rotor Blade | |
| Ma et al. | An interpretability research of the Xgboost algorithm in remaining useful life prediction | |
| CN116910513A (en) | Centrifugal pump rotor fault diagnosis method based on CWGAN-GP and double-flow CNN model | |
| CN116167220A (en) | Simulation method and system of offshore wind turbine under comprehensive airflow factors | |
| Zhang et al. | Study on the Intelligent Modeling of the Blade Aerodynamic Force in Compressors Based on Machine Learning. Mathematics 2021, 9, 476 |
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 |