CN110431395A - Fatigue crack growth prediction - Google Patents
Fatigue crack growth prediction Download PDFInfo
- Publication number
- CN110431395A CN110431395A CN201880018589.4A CN201880018589A CN110431395A CN 110431395 A CN110431395 A CN 110431395A CN 201880018589 A CN201880018589 A CN 201880018589A CN 110431395 A CN110431395 A CN 110431395A
- Authority
- CN
- China
- Prior art keywords
- data
- crack growth
- fatigue crack
- machine
- machine learning
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M5/00—Investigating the elasticity of structures, e.g. deflection of bridges or air-craft wings
- G01M5/0033—Investigating the elasticity of structures, e.g. deflection of bridges or air-craft wings by determining damage, crack or wear
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M15/00—Testing of engines
- G01M15/14—Testing gas-turbine engines or jet-propulsion engines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F15/00—Digital computers in general; Data processing equipment in general
- G06F15/76—Architectures of general purpose stored program computers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
-
- 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
Abstract
Provide the system and method for predicting fatigue crack growth.In an example embodiment, a kind of method may include obtaining historical operating data associated with one or more one or more rotary structures of machine, the data for indicating the fatigue crack size of one or more rotary structures are obtained, and are constructed using machine learning techniques by fatigue crack growth machine learning model associated with operation data.
Description
Priority claim
This application claims the equity of following priority: entitled " the fatigue crack growth prediction " submitted on October 24th, 2017
U.S. Provisional Patent Application No.62/576,234；The beauty of entitled " the fatigue crack growth prediction " submitted on March 13rd, 2017
State temporary patent application No.62/470,539, the two is both incorporated herein by reference for all purposes.
Technical field
The present subject matter relates generally to for predicting the digital display circuit of the fatigue crack growth in machinery, which includes rotatable
Structure, such as the rotor for gas-turbine unit.
Background technique
Fatigue of materials is the common phenomenon that structure fails when being subjected to cyclic loading.If load is more than some threshold value,
Microfissure is initially formed at the point that stress is concentrated.Finally, crackle will propagate to critical dimension, and structure will rupture.Therefore, quasi-
Really for tracking crack growth for ensuring the availabilities of the various industrial circles including aviation, reliability and operational safety are non-
It is often important.
Fatigue crack growth can be influenced by many factors, such as temperature, load, surface appearance, size, the microcosmic knot of metallurgy
Structure (metallurgicalmicrostructure), oxidation or the presence of inert chemi-cal product, residual stress, corrosion, fine motion etc..
In addition, crack growth can be the nonlinearity process with different advance stages.In view of these challenges, most of determinations are tired
The existing method of labor crack growth uses the method based on physics, such as linear elastic fracture mechanics (LEFM), is that calculating is close
Collection type, and may not be near real-time or the ideal chose applied in real time.
Summary of the invention
The aspect and advantage of embodiment of the disclosure will illustrate partly in the following description, or can be from description middle school
It practises, or can be learnt by the practice of embodiment.
One exemplary aspect of the disclosure is related to a kind of computing system, including one or more processors and one or
Multiple memory devices.One or more memory device for storing computer-readable instructions, computer-readable instruction is by one
Or to execute one or more processors when executing related to operation data by fatigue crack growth for constructing for multiple processors
The operation of the machine learning model of connection.Operation includes that acquisition is related to one or more one or more rotary structures of machine
The historical operating data of connection obtains the data for indicating the fatigue crack size of one or more rotary structures, and uses machine
Device learning art is constructed fatigue crack growth machine learning model associated with operation data.
Another exemplary aspect of the disclosure is related to a kind of for predicting the computer implemented method of fatigue crack growth.It should
Method includes obtaining operation data associated with one or more rotatable parts of machine by one or more processors.It should
Method includes by one or more processors access that operation data is associated with fatigue crack growth based on non-physical mould
Type.Based on historical operating data, constructed using machine learning techniques based on non-physical model.This method includes at least partly
Based on model and operation data, pass through the determining fatigue associated with one or more rotatable parts of one or more processors
Crack growth.
Another exemplary aspect of the disclosure is related to storing the tangible non-transitory computer readable medium of computer-readable instruction
Matter, computer-readable instruction make one or more processors execute operation when executed by one or more processors.Operation
Including obtaining historical operating data associated with one or more rotary structures of each machine in multiple machines, obtain
It indicates the data of the fatigue crack size of one or more rotary structures of each machine in multiple machines, and uses machine
Device learning art is constructed fatigue crack growth machine learning model associated with operation data.
These example embodiments of the disclosure can be changed and be modified.It is wanted with reference to being described below with appended right
It asks, is better understood with these and other features of various embodiments, aspect and advantage.Comprising in the present specification and constituting it
The attached drawing of a part shows embodiment of the disclosure, and is used to explain relative theory together with specification.
Detailed description of the invention
Being discussed in detail for the embodiment for being directed to those of ordinary skill in the art, specification reference are elaborated in the description
Attached drawing, in which:
Fig. 1 depicts the flow chart of exemplary method according to an example embodiment of the present disclosure；
Fig. 2 depicts the flow chart of exemplary method according to an example embodiment of the present disclosure；
Fig. 3 depicts the fatigue crack growth for being divided into four different growth regions according to an example embodiment of the present disclosure；
Fig. 4 depicts the exemplary operations circulation of the feature input that may be used as model according to an example embodiment of the present disclosure
Graphical representation；
Fig. 5 depicts the flow chart of exemplary method according to an example embodiment of the present disclosure；With
Fig. 6 depicts exemplary computing system according to an example embodiment of the present disclosure.
Specific embodiment
Now with detailed reference to embodiment of the disclosure, one or more example is shown in the accompanying drawings.By explaining institute
Disclosed technology provides each example, rather than limits disclosed technology.In fact, showing to those skilled in the art
And be clear to, in the case where not departing from the scope of the claims or spirit, can carry out various modifications in the disclosure and
Variation.For example, the feature that a part as one embodiment shows or describes can be used together with another embodiment, with
Generate another embodiment.Therefore, the disclosure is intended to cover these come within the scope of the appended claims and their
Modifications and variations.
As used in specification and appended, singular " one ", "an" and "the" refer to including plural number
Show object, unless the context is clearly stated.Term " about " and numerical value are used in combination within refer to the amount 25%.
The exemplary aspect of the disclosure is related to the system and method predicted for fatigue crack growth.It can be with reference to for aviation
Fatigue crack growth in the armature spindle of the gas-turbine unit of (for example, providing propulsive force for aircraft) discusses the disclosure
Various aspects.However, using disclosure provided herein, it will be appreciated by the skilled addressee that all aspects of this disclosure
It can be used for predicting the fatigue crack growth of any kind of rotary structure in various applications, various application such as wind-force whirlpools
Turbine, jet engine, turboprop, navigate change-based gas turbine, amateurish gas turbine, auxiliary power unit, hair
Electric gas turbine, turboaxle motor, radial gas turbine, ratio jet engine, Microturbine, internal combustion engine are electronic
Engine, drilling machine and other tool/equipment, speed changer or other application.
According to example embodiment, it provides by the data of one or more monitoring system registers, which is configured
For the parameter for monitoring the machine including one or more rotatable parts during operation.For example, flying during flight can be collected
The parameter (for example, " flying quality ") of the gas-turbine unit of row device, the ginseng of the turbine during steam, water or wind-power electricity generation
The parameter (for example, " driving data ") of number (for example, " dynamic date ") or driving period internal combustion engine or speed changer.Operand
According to may include such as core speed relevant to gas-turbine unit or other machines, temperature, torque, the ginseng of acceleration etc.
Number.In one example, operation data is the flying quality for including the high frequency sensing data collected by onboard flight logger.It is dynamic
Force data and driving data can also be collected by airborne operation recorder.Machine learning techniques, which can be used for constructing, reflects operation data
It is mapped to one or more models of fatigue crack growth.One or more models may each be based on non-physical model.Citing
For, model can be used for come Cycle by Cycle predicting the crack growth of each rotor of gas-turbine unit based on actual use.
Similarly, model predicts each gear in speed changer, each axis in turbine, engine or speed changer in which can be used for Cycle by Cycle
Or the crack growth of rotor etc..It should be appreciated that any operation data associated with the rotatable part of machine can be used
Generate model.
The exemplary aspect of the disclosure can provide many technical effects and benefit.For example, carrying out structure using machine learning techniques
Build by operation data be mapped to fatigue crack growth model can around by the model of physics calculate used in based on complexity
It calculates, such as calculating stress strength factor and other complicated LEFM parameters.This can permit processing and storage resource for other
Function.In addition, can be analysis model according to the model that the exemplary aspect of the disclosure constructs, can permit based on actual use
Almost instantaneously predict fatigue crack growth.Analysis model can provide accurate near real-time or real-time engineering prediction on fatigue crack growth.Make
Benefit with the model prediction fatigue crack growth constructed according to an example embodiment of the present disclosure may include: that (1) setting is used for
The safe and suitable interface that component is removed and repaired；(2) extend the running time of assets；(3) optimize asset operation and its with
The correlation of problem on the spot.
The exemplary aspect of the disclosure can provide the improvement of computing technique.For example, using machine learning rather than being based on object
The calculating of reason can provide the exploitation that the model assessed is easier relative to the model based on physics, for predicting that fatigue crack increases
It is long.This can save the processing and storage resource of computing system.Model can also provide the processing of faster fatigue crack growth
And prediction.
In the exemplary embodiment, model can be used to select such as component inspection, repair and/or the maintenance of replacement is grasped
Make.For example, system can receive the operation data of component or machine associated with component.Data and model can be used in system
Determine the crack growth of prediction.Then, system can be arranged and/or be executed attended operation based on the crack growth of prediction.In
In some examples, system can the instruction generation automatic maintenance associated with component based on the fatigue crack growth of prediction disappear
Breath.These technologies can be with the optimization component service life, while minimizing downtime relevant to attended operation.For example, passing through prediction
When crackle reaches the size of worth attended operation, can be to avoid unnecessary or premature attended operation.
Fig. 1 depicts the example side for being used to construct fatigue crack growth prediction model according to an example embodiment of the present disclosure
The flow chart of method (100).This method can be realized by any suitable computing system, such as the computing system described in Fig. 6.Separately
Outside, although Fig. 1 depicts the step of executing for explanation and the purpose discussed with particular order.Using in disclosure provided herein
Hold, it will be appreciated by the skilled addressee that any method disclosed herein or each step of processing can be without departing from these
It is adjusted in the case where the range of invention, extends, be performed simultaneously, omit and/or rearrange.
In the exemplary embodiment, method 100 can be by the one or more first of such as one or more first processors
Equipment is calculated to execute.The behaviour that a sensor more than first monitors more than first a machines can be used in one or more first processors
Make.For example, one or more processors associated with health and use monitoring system (HUMS) can be from for aircraft machine
Multiple sensor collection data of multiple aircraft in group.
At (102), this method includes the historical sensor data for obtaining the parameter that can influence fatigue crack growth.History
Sensing data can be such as flying quality, the operation data of dynamic date or driving data, and may include with such as
Rotor, axis, the relevant parameter of operation of the rotatable parts such as gear.In some embodiments, sensing data can be health
It may include related to the physical unit operation of such as actual rotor operation with the data for using monitoring system (HUMS) to collect
Parameter comprehensive and continuous record.Parameter may include such as core speed, temperature, torque, acceleration etc..Historical sensor
Data can be the rotatable knot of one or more with one or more machines (such as the first aircraft or first group of aircraft)
The associated historical operating data of structure.In some instances, (102) may include using more than first a sensor monitorings more than first
The operation of a machine is to determine historical operating data.In the exemplary embodiment, (102) may include using full flying quality, it is all
It such as include engine parameter, the operation data of environmental parameter and other vehicle parameters.
At (104), this method may include obtaining historical environmental condition data.Historical environmental condition data may include
With the data of the environmental correclation connection of machine operation.For example, the historical environmental condition data of gas-turbine unit may include environment
Temperature, operating condition and other data relevant to the operating environment of gas-turbine unit.
At (106), this method may include obtaining the data for indicating practical fatigue crack size.This can be used for determining instruction
Practice the brass tacks of model.Indicate that the data of practical fatigue crack size can obtain in various ways.For example, can be by straight
It connects measurement and obtains data.Data can be obtained by LEFM.Operation data whether or not using can be based on physics by other
Method obtain data.
At (108), the data of instruction fatigue crack growth can be determined based on the data of instruction crack size.One
In a little embodiments, sensing data and/or environmental data can be considered when determining fatigue crack growth.Indicate the data of fatigue
It may be used as the dependent variable of training pattern (such as machine learning model or other based on non-physical model).Indicate fatigue crack
The data of growth can be, for example, instruction Fatigue Crack Growth Rate or absolute crack growth.
In (110), machine learning techniques can be used for training based on the data of instruction fatigue crack growth and flying quality
Model.In some embodiments, environmental data can be considered in training pattern.It can structure according to an example embodiment of the present disclosure
Make the model of any suitable type.For example, Random Forest model (" RF model ") and/or neural network model (" NN can be constructed
Model ").In some embodiments, the nonlinear regression with or without regularization can be used.In some embodiments,
Grad enhancement machine, artificial neural network, one or more of self organization map and/or deep learning can be used.
In some embodiments, two kinds of RF model can be constructed.For example, RF disaggregated model can be constructed to identify
Crack growth region.Furthermore, it is possible to construct the RF regression model of the Fatigue Crack Growth Rate in each crack growth region.
For RF model, indicate that the data of Fatigue Crack Growth Rate may be used as the dependent variable of model.Crack size can
To be defined as the crack length of intended size or area along crackle.For crackle area model, can training pattern it
The preceding logarithmic transformation using crack growth rate.Crack length is modeled, may not be needed crackle increasing before training pattern
The logarithmic transformation of long rate.
For RF disaggregated model, the sum in region can change according to the observing pattern from training data.Fig. 3 is shown
An example when crack growth is divided into four different growth regions.For RF disaggregated model, can intentionally make
Error propagation is fought with the training data of uneven class (non-equilibrium class), to use RF model to carry out n step advanced prediction.It lifts
For example, some classes can have more multi-site data, such as no slow crack growth compared with having the faster class of crack growth
Class.Model may tend to be partial to include more multi-site data class.In general, can be by selecting equal numbers from each class
The sample of amount is forced using balanced class.However, it is not ideal that this balance method, which may model crack growth,.Cause
The training data of uneven class can be used for the disaggregated model in example embodiment in this.Due to Fatigue Crack Growth Rate
Difference, system can be randomly selected or using pre-defined rule.For RF disaggregated model, current crackle size may or may not
As fallout predictor.Regulation mechanism is guarded, that is, will be pre- when implementing the model for n step advanced prediction for RF regression model
The crack growth of survey may or may not be used multiplied by the coefficient less than 100%.
In some embodiments, NN disaggregated model can be constructed.Current crack size may or may not be used as fallout predictor.
It may or may not implement the starting point of training data.
For NN model, crack growth rate, absolute crack growth and/or crack size can be the dependent variable of model.
To model for both crackle area and length, the logarithmic transformation of dependent variable may or may not be executed.
In some embodiments, crack growth rate can be defined as follows:
Crack growth rate is defined as percentage growth as described above can provide more meaningful output for model.
Fig. 2 depicts the exemplary method being used for using machine learning training pattern according to an example embodiment of the present disclosure
101 flow chart.The model can be machine learning model.As set forth above, it is possible to be obtained by obtaining operation data (102)
Indicate the data (106) of fatigue crack size and/or by obtaining environmental aspect data (104) come training pattern.
At (112), this method may include preprocessed data.For example, can handle the original behaviour of such as flying quality
Make data, with identification by sensor fault, the intake of imperfect or repeated data passes through the incorrect data type transmitted or stored
Quality problems caused by conversion etc..
In (114), this method may include executing activity classification.For example, can be based on pretreated operation data to behaviour
Classify.It can identify the operation for being suitable for machine learning model exploitation.
In some embodiments, growth region sort operation relevant to fatigue crack growth can be based on.With reference to Fig. 3,
For example, fatigue crack growth can be considered as tool, there are four different growth regions.These regions include that crackle forms and increases
Multiple and different stages, since crack initiation to crackle reaches critical dimension (for example, it may be possible to leading to component failure).For example, Fig. 3
The first chart 120 is depicted, it illustrates the superpositions for the line for indicating the crack growth in each different crack growths region.Fig. 3
The second chart 122 is depicted, it illustrates the crack growths in the first crack growth stage, wherein crack growth is relative to following
Number of rings is slightly linear.Fig. 3 depicts third chart 124, and it illustrates the crack growths in the second crack growth stage, wherein
Crack growth increases with very small circulation, is followed by bigger and increased crack growth rate.Fig. 3 depicts the 4th chart
126, it illustrates the crack growths in the third crack growth stage, wherein crack growth slightly linearly carries out, subsequent crackle
Rate of rise quicklys increase.Fig. 3 depicts the 5th chart 128, it illustrates the crack growth in the 4th crack growth stage,
In, crack growth is seldom, and subsequent crack growth rate quicklys increase.
At (114), it can be operated according to corresponding crack growth region or Stage Classification.It in some instances, can be with
Model is created for each different fatigue crack growth region.Therefore, operation can be divided into different regions and for instructing
Practice the operation data of the model of corresponding region.It has been provided by way of example only the mould using four crack growth regions and respective numbers
Type carrys out simulating crack growth.Any amount of region and model can be used.
At (116), this method may include Feature Engineering (feature engineering) based on for training mould
The operation data of type determines feature appropriate.Exemplary characteristics discussed further below.
In some embodiments, residence time characteristic is determined.Residence time characteristic may include flight, and power generation process is driven
Dynamic or any other moving event duration, and selected engine parameter is maintained at the spy specified by upper and lower bound
Determine in range.For example, selected engine parameter may include the temperature of for example each position for engine, core is started
Machine speed, acceleration etc..It can be respectively each engine or determine upper and lower bound jointly for various engines.Similar ginseng
Several and boundary can be used for other machines, such as speed changer, tool etc..
In some embodiments, value temporal characteristics (time-at-value feature) and the temporal characteristics that overflow are determined
(time-above-value feature).For example, value temporal characteristics and the temporal characteristics that overflow may include duration flight,
Selected engine parameter is maintained at selected lower limit or is higher than selected lower limit simultaneously.Selected operating parameter can wrap
Include the temperature of each position, core-engine speed, torque, acceleration etc..Lower limit can individually be extracted from each machine or from
Various machines extract together.
In some embodiments, rolling window feature can be determined.Rolling window feature may include for example in selected length
The statistics and convergence value or their combination of selected machine parameter during the rolling window of degree.Statistics and convergence function may include
The average value of preset value, intermediate value, maximum value, minimum value, standard deviation, quartile range, summation, product, counting, it is all aforementioned
The accumulated value of function, logarithmic transformation of all aforementioned functions etc..Combination may include the product of another feature, division, subtraction,
With exponential depth etc..For certain features of combination, can be used or without using non-uniform rolling window length.Selected hair
Motivation or other machines parameter include but is not limited to the temperature and core-engine speed of various positions, torque, acceleration etc..Root
According to the sampling interval, rolling window length changes in known flight or other operations from 1 sampling interval to maximum length.
In some embodiments, the counting of known operation circulation relevant to fatigue can be determined.As shown in Figure 4 is certain
Operation circulation is defined as being moved to another speed from an engine speed band (being indicated by upper and lower bound threshold value)
Band then returnes to the complete cycle of original engine speed band, is the known factor for influencing crack growth.These operations follow
The counting of ring may be used as input feature vector.Similar speed band can be used for speed changer and other machines with rotary structure
Device.
In some embodiments, the accumulation feature in the different flights executed by same engine is determined.All of above spy
Sign can be extracted from solo hop.But a specific engine can execute thousands of flight in its lifetime.Cause
This, the cumulative effect of all features described above also is used as input feature vector in different task.Similarly, it can determine across machine not
The accumulation feature of same driver, power generation process or other action panes.
In some embodiments, characteristic optimization can be executed.Processing historical operating data can include determining that for using
Machine learning techniques training machine learning model or other one or more input feature vectors based on non-physical model.It can be with base
Come identification feature group in similitude.It, can be based on specific machine (for example, engine, speed changer, work during model training
Tool etc.) position/crack growth part identifies important feature.Then these important features are used as the excellent of machine learning model
Change feature.It is also based on each machine recognition important feature.
At (118), this method may include training, adjustment and cross validation one or more model.In some implementations
In example, input feature vector can be mapped to the crack growth rate or other dependent variables of each circulation by one or more models.
Fig. 5 depict using the model constructed according to the exemplary aspect of the disclosure come based on real-time or near real-time flight or
Other operation datas predict the flow chart of the exemplary method (200) of fatigue crack growth.In the exemplary embodiment, model can be with
It is machine learning model.Method (200) can be realized by any suitable computing system, such as the computing system described in Fig. 6.
In addition, although Fig. 1 depicts the step of executing for explanation and the purpose discussed with particular order.Use disclosure provided herein
Content, it will be appreciated by the skilled addressee that any method disclosed herein or each step of processing can without departing from
It is adjusted in the case where the scope of the present invention, extends, be performed simultaneously, omit and/or rearrange.
In the exemplary embodiment, method 200 can be by the one or more second of such as one or more second processors
Equipment is calculated to execute, and method 100 is executed by one or more first processors.One or more second processors can be matched
It is set to using previously trained model and predicts the crack growth of a machine more than second.For example, one or more second processors
The operation data of more than second a aircraft can be supplied to machine learning model, and receive with more than second a aircraft can
The instruction of the fatigue crack growth of the associated prediction of rotary part is as output.
At (202), this method may include Access Model.Machine learning techniques as described above can be used to instruct in advance
Practice model.The model can be associated with fatigue crack growth by operation data.This method may include obtaining sensing data
(for example, flying quality) (204) and/or environmental aspect data (206).In some instances, (204) may include using second
Multiple sensors monitor the operation of more than second a machines, to determine operation data associated with more than second a machines.Based on this
Data can obtain the crack growth (210) of prediction using the model (208).The crack growth (210) of prediction can be fed back
Into model, for predicting the crack growth in subsequent cycle.
According to the example embodiment of disclosed technology, one or more rotary structure phases at least the first machine are used
Associated historical operating data carrys out training machine learning model.In some embodiments, system can will be with one or more
The associated operation data of additional machine is input to model.For example, model can be configurable to include one or more inputs,
It is configured as receiving additional operations data associated with having the additional machine of rotary structure.The model may include one
Or multiple outputs, it is configured to supply the finger of the fatigue crack growth of prediction associated with the rotary structure of additional machine
Show.As the one or more output of machine learning model, system can be generated associated with the rotary structure of additional machine
Prediction fatigue crack growth instruction.
According to some aspects of disclosed technology, system can instruction generation based on the fatigue crack growth of prediction and machine
Or the associated automatic maintenance message of rotary structure of machine.It can be in response to safeguarding that message executes one or more dimensions automatically
Shield operation.For example, can be replaced automatically in response to safeguarding message automatically or inspection part.
Fig. 6 depicts the exemplary computing system that can be used to implement system and method according to an example embodiment of the present disclosure
Block diagram.As shown, the system may include one or more calculating equipment 802.One or more calculates equipment 802 can be with
Including one or more processors 804 and one or more memory devices 806.One or more processors 804 may include
Any suitable processing equipment, such as microprocessor, microcontroller, integrated circuit, logical device or other suitable processing are set
It is standby.One or more memory devices 806 may include one or more computer-readable mediums, including but not limited to nonvolatile
Property computer-readable medium, RAM, ROM, hard disk drive, flash drive or other memory devices.
One or more memory devices 806 can store the information that can be accessed by one or more processors 804, including
The computer-readable instruction 808 that can be executed by one or more processors 804.Instruction 808 can be any instruction set, work as
When being executed by one or more processors 804, so that one or more processors 804 execute operation.Instruction 808 can be to appoint
The software what suitable programming language is write, or can be implemented in hardware.In some embodiments, instruction 806 can be by one
A or multiple processors 804 execute so that one or more processors 804 execute operation.Memory devices 806 can also store
The data 810 that can be accessed by processor 804.For example, data 810 may include operation data (for example, flying quality), crackle
Increase data, environmental aspect data associated with model etc..
One or more calculate equipment 802 can also include communication interface 812, for for example with the other component of system
And/or other computing device communications.Communication interface 812 may include for any suitable with one or more network interfaces
Component, including such as transmitter, receiver, port, controller, antenna or other suitable components.
Technical Reference computer based system discussed here and the movement taken by computer based system and
The information for being sent to computer based system and being sent from computer based system.Those of ordinary skill in the art will recognize that
Arrive, the intrinsic flexibility of computer based system allows the various possible configurations between component and in component, combination and
The division of task and function.For example, the single multiple calculating for calculating equipment or work in combination can be used in processing discussed here
Equipment is realized.Database, memory, instruction and application program can realize on a single, can also be across multiple systems
Distribution.Distributed elements can sequence or parallel work-flow.
Although the specific features of various embodiments may show in some drawings and be not shown in the other drawings, this
It is merely for convenience.According to the principle of the disclosure, it can refer to and/or require to protect in conjunction with any feature of any other attached drawing
Protect any feature of attached drawing.
This written description uses examples to disclose the disclosure, including optimal mode, and also enables those skilled in the art
The disclosure is enough practiced, the method including manufacturing and using any device or system and executing any combination.The disclosure can be special
Sharp range is defined by the claims, and may include other examples that those skilled in the art expect.If these other show
Example includes the structural detail not different from the literal language of claim, or if they include literal with claim
Equivalent structural elements of the language without essential difference, then these other examples intention is fallen within the scope of the appended claims.
Claims (20)
1. a kind of computing system characterized by comprising
One or more processors；With
One or more memory devices, one or more of memory device for storing computer-readable instructions, the calculating
Machine readable instruction executes one or more of processors when being executed by one or more of processors will for building
The operation of fatigue crack growth machine learning model associated with operation data, the operation include:
Obtain historical operating data relevant to one or more one or more rotary structures of machine；
Obtain the data for indicating the fatigue crack size of one or more of rotary structures；With
It is constructed using machine learning techniques by fatigue crack growth machine learning model associated with operation data.
2. computing system according to claim 1, which is characterized in that wherein one or more of machines are more than first
Machine, the machine learning model include one or more inputs and one or more outputs, one or more of input quilts
Be configured to receive operation data associated with more than second a machines, it is one or more of export be configured to supply with it is described
The finger of the fatigue crack growth of the associated prediction of one or more rotary structures of each machine more than second in a machine
Show, the operation further comprises:
Operation data associated with more than described second first machines of a machine is input to the machine learning model；
The first instruction for generating the fatigue crack growth of prediction associated with the first rotary structure of first machine, makees
For one or more of outputs of the machine learning model；With
First instruction of fatigue crack growth based on prediction, generates associated with first rotary structure automatic
Safeguard message.
3. computing system according to claim 2, which is characterized in that wherein:
The operation further comprises monitoring the operation of a machine more than described first described in determination using more than first a sensors
Historical operating data, and the operation of a machine more than described second is monitored using more than second a sensors to determine and described second
Multiple associated operation datas of machine；
The machine learning model is constructed to be executed by at least first processor in one or more of processors；And
First instruction of the fatigue crack growth of prediction is generated by least second in one or more of processors
Device is managed to execute.
4. computing system according to claim 2, which is characterized in that the wherein operation further comprises:
One or more attended operations associated with first rotary structure are executed based on the automatic maintenance message.
5. computing system according to claim 1, which is characterized in that wherein:
The historical operating data includes flying quality relevant to multiple aircraft；And
The historical operating data is by one or more associated with the health of the multiple aircraft and use monitoring system
Sensor collection.
6. computing system according to claim 1, which is characterized in that wherein constructing the machine learning model includes:
Determine Fatigue Crack Growth Rate associated with for constructing multiple circulations of the machine learning model.
7. computing system according to claim 6, which is characterized in that wherein:
The operation further comprises obtaining environmental aspect data；
Determine that the Fatigue Crack Growth Rate is based at least partially on the environmental aspect data；And
It constructs the machine learning model and is based at least partially on the Fatigue Crack Growth Rate.
8. computing system according to claim 1, which is characterized in that wherein:
The operation data includes indicating the data of at least one of temperature, core speed, torque or acceleration.
9. computing system according to claim 1, which is characterized in that wherein:
The operation further comprises obtaining environmental aspect data；And
It constructs the machine learning model and is based at least partially on the environmental aspect data.
10. computing system according to claim 1, which is characterized in that the wherein operation further comprises:
The historical operating data is handled to determine one using the machine learning techniques training machine learning model
Or multiple input feature vectors；
Wherein, one or more of input feature vectors include residence time characteristic, are worth temporal characteristics, the temporal characteristics that overflow, roll
At least one of the counting of window feature or known operation circulation.
11. computing system according to claim 1, which is characterized in that wherein indicate the data of fatigue crack size
It is obtained from the model based on physics.
12. computing system according to claim 1, which is characterized in that wherein:
The machine learning model includes Random Forest model；And
The Random Forest model includes disaggregated model and regression model.
13. computing system according to claim 1, which is characterized in that wherein the machine learning model is neural network
Model.
14. a kind of for predicting the computer implemented method of fatigue crack growth characterized by comprising
Operation data associated with one or more rotatable parts of machine is obtained by one or more processors；
It is accessed by one or more of processors operation data is associated with fatigue crack growth based on non-physical
Model is wherein at least based in part on historical operating data, is constructed using machine learning techniques described based on non-physical mould
Type；With
Based on non-physical model and the operation data described in being based at least partially on, pass through one or more of processors
Determine fatigue crack growth associated with one or more of rotatable parts.
15. computer implemented method according to claim 14, which is characterized in that further comprise:
It is based at least partially on the fatigue crack growth, one is executed to one or more of rotatable parts of the machine
A or multiple attended operations.
16. computer implemented method according to claim 14, which is characterized in that further comprise:
Obtain environmental aspect data；
Determine Fatigue Crack Growth Rate associated with for constructing multiple circulations based on non-physical model, it is described
Fatigue Crack Growth Rate is based at least partially on the environmental aspect data；With
It is described based on non-physical model to be based at least partially on the Fatigue Crack Growth Rate building.
17. computer implemented method according to claim 14, which is characterized in that further comprise:
Obtain historical operating data associated with one or more rotary structures；With
It is described based on non-physical model using machine learning techniques training to determine to handle the historical operating data
One or more input feature vectors；
Wherein, one or more of input feature vectors include residence time characteristic, are worth temporal characteristics, the temporal characteristics that overflow, roll
At least one of the counting of window feature or known operation circulation.
18. a kind of tangible non-transitory computer-readable medium for storing computer-readable instruction, which is characterized in that the meter
Calculation machine readable instruction makes one or more of processors execute operation, the operation when executed by one or more processors
Include:
Obtain historical operating data associated with one or more rotary structures of each machine in multiple machines；
Obtain the fatigue crack size for indicating one or more of rotary structures of each machine in the multiple machine
Data；With
It is constructed using machine learning techniques by fatigue crack growth machine learning model associated with operation data.
19. non-transitory computer-readable medium according to claim 18, which is characterized in that wherein described to operate into one
Step includes:
Additional operations data are input to the machine learning model, the additional operations data with include first add it is rotatable
First additional machine of structure is associated；
Generate fatigue crack growth prediction, the output as the machine learning model；With
Automatic maintenance message is generated based on fatigue crack growth prediction.
20. non-transitory computer-readable medium according to claim 19, which is characterized in that wherein described to operate into one
Step includes:
One or more attended operations are executed to the described first additional rotary structure based on the automatic maintenance message.
Applications Claiming Priority (5)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US201762470539P | 2017-03-13 | 2017-03-13 | |
| US62/470,539 | 2017-03-13 | ||
| US15/910,412 US20180260720A1 (en) | 2017-03-13 | 2018-03-02 | Fatigue Crack Growth Prediction |
| US15/910,412 | 2018-03-02 | ||
| PCT/US2018/021011 WO2018169722A1 (en) | 2017-03-13 | 2018-03-06 | Fatigue crack growth prediction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110431395A true CN110431395A (en) | 2019-11-08 |
Family
ID=63523228
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201880018589.4A Pending CN110431395A (en) | 2017-03-13 | 2018-03-06 | Fatigue crack growth prediction |
Country Status (3)
| Country | Link |
|---|---|
| EP (1) | EP3596439A4 (en) |
| CN (1) | CN110431395A (en) |
| WO (1) | WO2018169722A1 (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113065224A (en) * | 2021-03-05 | 2021-07-02 | 天津大学 | Deep sea pipeline crack propagation monitoring and reliability evaluation method based on image recognition |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11397134B2 (en) | 2018-12-03 | 2022-07-26 | Raytheon Technologies Corporation | Intelligent learning device for part state detection and identification |
| CN111914353B (en) * | 2020-07-13 | 2024-01-19 | 中广核核电运营有限公司 | Rotor low-cycle fatigue loss detection method, device and computer equipment |
| CN113343530B (en) * | 2021-06-11 | 2022-05-06 | 清华大学 | Design method and device for controlling fatigue damage fracture of space station shell structure |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060009951A1 (en) * | 2001-01-08 | 2006-01-12 | Vextec Corporation | Method and apparatus for predicting failure in a system |
| JP2010025715A (en) * | 2008-07-18 | 2010-02-04 | Nippon Oil Corp | Pipe crack diagnosing apparatus and pipe crack diagnosing method |
| CN102128880A (en) * | 2010-01-12 | 2011-07-20 | 上海工程技术大学 | Crack shape inversion method |
| JP2011242365A (en) * | 2010-05-21 | 2011-12-01 | Nippon Telegr & Teleph Corp <Ntt> | Crack detection device and crack detection method for concrete structure image, and program therefor |
| US20130124166A1 (en) * | 2011-11-15 | 2013-05-16 | Jack Gammill Clemens | Modeling operation of a tool in a wellbore |
| CN103733322A (en) * | 2011-06-17 | 2014-04-16 | 弗劳恩霍弗实用研究促进协会 | Method for providing a prediction model for crack detection and method for crack detection on a semiconductor structure |
| CN104515786A (en) * | 2015-01-08 | 2015-04-15 | 北京科技大学 | Method for detecting and analyzing internal defect evolution of metal casting in fatigue process |
| CN105051527A (en) * | 2013-01-04 | 2015-11-11 | 西门子公司 | Probabilistic modeling and sizing of embedded flaws in nondestructive inspections for fatigue damage prognostics and structural integrity assessment |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1143242B1 (en) * | 1994-08-31 | 2003-07-16 | Honeywell Inc. | Remote self-powered structure monitor method |
| US7027953B2 (en) * | 2002-12-30 | 2006-04-11 | Rsl Electronics Ltd. | Method and system for diagnostics and prognostics of a mechanical system |
| US8768657B2 (en) * | 2006-01-12 | 2014-07-01 | Jentek Sensors, Inc. | Remaining life prediction for individual components from sparse data |
| WO2013040315A1 (en) * | 2011-09-16 | 2013-03-21 | Sentient Corporation | Method and system for predicting surface contact fatigue life |
| EP2637010B1 (en) * | 2012-03-05 | 2015-06-24 | EADS Construcciones Aeronauticas, S.A. | Method and system for monitoring a structure |
-
2018
- 2018-03-06 EP EP18766900.7A patent/EP3596439A4/en active Pending
- 2018-03-06 CN CN201880018589.4A patent/CN110431395A/en active Pending
- 2018-03-06 WO PCT/US2018/021011 patent/WO2018169722A1/en unknown
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060009951A1 (en) * | 2001-01-08 | 2006-01-12 | Vextec Corporation | Method and apparatus for predicting failure in a system |
| JP2010025715A (en) * | 2008-07-18 | 2010-02-04 | Nippon Oil Corp | Pipe crack diagnosing apparatus and pipe crack diagnosing method |
| CN102128880A (en) * | 2010-01-12 | 2011-07-20 | 上海工程技术大学 | Crack shape inversion method |
| JP2011242365A (en) * | 2010-05-21 | 2011-12-01 | Nippon Telegr & Teleph Corp <Ntt> | Crack detection device and crack detection method for concrete structure image, and program therefor |
| CN103733322A (en) * | 2011-06-17 | 2014-04-16 | 弗劳恩霍弗实用研究促进协会 | Method for providing a prediction model for crack detection and method for crack detection on a semiconductor structure |
| US20130124166A1 (en) * | 2011-11-15 | 2013-05-16 | Jack Gammill Clemens | Modeling operation of a tool in a wellbore |
| CN105051527A (en) * | 2013-01-04 | 2015-11-11 | 西门子公司 | Probabilistic modeling and sizing of embedded flaws in nondestructive inspections for fatigue damage prognostics and structural integrity assessment |
| CN104515786A (en) * | 2015-01-08 | 2015-04-15 | 北京科技大学 | Method for detecting and analyzing internal defect evolution of metal casting in fatigue process |
Non-Patent Citations (3)
| Title |
|---|
| 向家伟 等: "基于神经网络的短粗转轴裂纹诊断研究", 《振动与冲击》 * |
| 周绮凤 等: "基于随机森林的建筑结构损伤识别方法", 《振动、测试与诊断》 * |
| 邓韧 等: "基于神经网络的疲劳短裂纹扩展速率时序预测方法", 《机械强度》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113065224A (en) * | 2021-03-05 | 2021-07-02 | 天津大学 | Deep sea pipeline crack propagation monitoring and reliability evaluation method based on image recognition |
Also Published As
| Publication number | Publication date |
|---|---|
| EP3596439A1 (en) | 2020-01-22 |
| WO2018169722A1 (en) | 2018-09-20 |
| EP3596439A4 (en) | 2021-03-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20230214691A1 (en) | Fatigue Crack Growth Prediction | |
| Filz et al. | Data-driven failure mode and effect analysis (FMEA) to enhance maintenance planning | |
| CN110431395A (en) | Fatigue crack growth prediction | |
| Elattar et al. | Prognostics: a literature review | |
| Lim et al. | Multimodal degradation prognostics based on switching Kalman filter ensemble | |
| US7395188B1 (en) | System and method for equipment life estimation | |
| Giantomassi et al. | Hidden Markov model for health estimation and prognosis of turbofan engines | |
| CN111259515A (en) | Aircraft health management method and system | |
| US20150234951A1 (en) | Determining life consumption of a mechanical part | |
| CN106951695A (en) | Plant equipment remaining life computational methods and system under multi-state | |
| CN105512483B (en) | Aero-engine predicting residual useful life based on damage propagation model and data analysis | |
| US10401847B2 (en) | System and method for issue detection of industrial processes | |
| CN107111309A (en) | Utilize the combustion gas turbine failure predication of supervised learning method | |
| CN103649858A (en) | Method and device for predicting the condition of a component or system, computer program product | |
| Lv et al. | Prognostics health management of condition-based maintenance for aircraft engine systems | |
| Balakrishnan et al. | Aero-engine health monitoring with real flight data using whale optimization algorithm based artificial neural network technique | |
| Knezevic et al. | Imperfect maintenance model for estimating aircraft fleet availability | |
| Alozie et al. | An adaptive model-based framework for prognostics of gas path faults in aircraft gas turbine engines | |
| Sambrekar et al. | Maintenance strategies for realizing Industry 4.0: An overview | |
| Baptista et al. | Improved time-based maintenance in aeronautics with regressive support vector machines | |
| Heier et al. | The use of PHM for a dynamic reliability assessment | |
| Daouayry et al. | Data-centric helicopter failure anticipation: The mgb oil pressure virtual sensor case | |
| Ayyalasomayajula et al. | A Tool-kit for Rotorcraft Regime Recognition Codes Validation | |
| Mehringskötter et al. | Consideration of Variable Operating States in a Data-Based Prognostic Algorithm | |
| Azevedo et al. | Online simulation of methods to predict the remaining useful lifetime of aircraft components |
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 |