CN116539459A - Fatigue crack growth rate prediction method and system based on acoustic emission monitoring and machine learning - Google Patents
Fatigue crack growth rate prediction method and system based on acoustic emission monitoring and machine learning Download PDFInfo
- Publication number
- CN116539459A CN116539459A CN202310496808.2A CN202310496808A CN116539459A CN 116539459 A CN116539459 A CN 116539459A CN 202310496808 A CN202310496808 A CN 202310496808A CN 116539459 A CN116539459 A CN 116539459A
- Authority
- CN
- China
- Prior art keywords
- acoustic emission
- rate
- crack growth
- fatigue crack
- fatigue
- 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
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000012544 monitoring process Methods 0.000 title claims abstract description 40
- 238000010801 machine learning Methods 0.000 title claims abstract description 27
- 239000011159 matrix material Substances 0.000 claims abstract description 39
- 230000002068 genetic effect Effects 0.000 claims abstract description 28
- 238000012549 training Methods 0.000 claims description 65
- 210000000349 chromosome Anatomy 0.000 claims description 58
- 238000012360 testing method Methods 0.000 claims description 19
- 210000002569 neuron Anatomy 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 6
- 230000035772 mutation Effects 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000012163 sequencing technique Methods 0.000 claims description 4
- 238000007430 reference method Methods 0.000 claims 1
- 238000002474 experimental method Methods 0.000 abstract description 8
- 238000010276 construction Methods 0.000 abstract description 4
- 238000013507 mapping Methods 0.000 abstract description 4
- 230000001186 cumulative effect Effects 0.000 description 10
- 239000000463 material Substances 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 5
- 238000009825 accumulation Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000009661 fatigue test Methods 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 3
- 238000007781 pre-processing Methods 0.000 description 3
- 238000010998 test method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000036541 health Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000001687 destabilization Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000000977 initiatory effect Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000013001 point bending Methods 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 229910001220 stainless steel Inorganic materials 0.000 description 1
- 239000010935 stainless steel Substances 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N3/00—Investigating strength properties of solid materials by application of mechanical stress
- G01N3/32—Investigating strength properties of solid materials by application of mechanical stress by applying repeated or pulsating forces
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/14—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object using acoustic emission techniques
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2203/00—Investigating strength properties of solid materials by application of mechanical stress
- G01N2203/0058—Kind of property studied
- G01N2203/0069—Fatigue, creep, strain-stress relations or elastic constants
- G01N2203/0073—Fatigue
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Acoustics & Sound (AREA)
- Investigating Strength Of Materials By Application Of Mechanical Stress (AREA)
Abstract
The invention discloses a fatigue crack growth rate prediction method and a fatigue crack growth rate prediction system based on acoustic emission monitoring and machine learning, which are characterized in that a genetic algorithm is utilized to optimize learning rate, weight matrix and bias vector in a BPNN model, and acoustic emission counting rate, amplitude rate, energy rate and information entropy rate obtained in an online fatigue crack growth monitoring process are used as input data of a GA-BPNN model together with fatigue experiment working condition parameter average stress and stress ratio so as to predict corresponding fatigue crack growth rate. Compared with the traditional method for predicting the linear relation between the single acoustic emission parameter and the crack expansion rate, the method provided by the invention considers the influence of the multiple acoustic emission characteristic parameter and the fatigue experimental working condition parameter, realizes the construction of the mapping relation among the complex fatigue load factor, acoustic emission monitoring data and the fatigue crack expansion rate, reduces the prediction error and improves the prediction precision.
Description
Technical Field
The invention belongs to the field of structural health monitoring and life prediction, and relates to a fatigue crack growth rate prediction method and system based on acoustic emission monitoring and machine learning.
Background
Fatigue fracture is one of the most dominant failure modes faced by materials and components during engineering service. Under the action of disturbance stress, high stress or high strain concentrated parts in materials or components, such as connecting parts of pressure container connecting pipes or cylinder circumferential welds, are easy to form fatigue damage, and fatigue fracture accidents finally occur through three stages of fatigue crack initiation, crack steady-state expansion, destabilization rapid expansion and the like. The propagation behavior of fatigue cracks brings great hidden trouble to the safety and reliability of engineering structures. Therefore, how to monitor the fatigue crack growth behavior of materials and structures on line and accurately evaluate and predict the fatigue crack growth rate according to the monitored data is an important research subject in the current academia and engineering world.
As a structural health monitoring technology, the acoustic emission technology has the unique advantages of high sensitivity, wide application range, long-term online monitoring and the like, and has been applied to online fatigue crack propagation monitoring and life evaluation of different materials and structures. Research shows that by extracting different sound emission parameters such as count, amplitude, information entropy and the like, not only can qualitative identification and characterization of fatigue crack growth in three stages be realized, but also quantitative prediction of fatigue crack growth rate and service life can be further realized by establishing a quantitative relation between the sound emission parameters and the fatigue crack growth rate. However, the existing fatigue crack growth rate prediction model based on acoustic emission monitoring data is mostly based on a linear empirical relationship between a single acoustic emission parameter and a crack growth rate, lacks strict theoretical basis, has prediction accuracy to be improved, and does not fully consider the influence of fatigue load factors such as stress ratio. These factors severely limit the further development and application of acoustic emission techniques in quantitative prediction of fatigue crack growth.
In recent years, with the rapid development of technologies such as artificial intelligence and data mining, a machine learning method has been successfully applied to various fields such as material life/performance prediction and new material design, and has significant advantages in predicting performance and time cost. Compared with the traditional crack growth rate or life prediction method, the machine learning method has the greatest advantage that the nonlinear mapping relation between the high-dimensional input variable and the output variable can be excellently established, so that the crack growth rate or life can be predicted efficiently and accurately.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a fatigue crack growth rate prediction method and a fatigue crack growth rate prediction system based on acoustic emission monitoring and machine learning, so as to solve the problems that the acoustic emission technology in the prior art has low prediction accuracy in the aspect of quantitative prediction of fatigue crack growth and the influence of fatigue load factors such as stress ratio is not fully considered, thereby realizing efficient and accurate prediction of fatigue crack growth behavior.
In order to achieve the purpose, the invention is realized by adopting the following technical scheme:
a fatigue crack growth rate prediction method based on acoustic emission monitoring and machine learning comprises the following steps:
step 1, acquiring acoustic emission parameters and fatigue working condition parameters;
step 2, predicting a fatigue crack growth rate value through a fatigue crack growth rate prediction model based on the acoustic emission parameter and the fatigue working condition parameter; the fatigue crack propagation rate model is a GA-BPNN model, and the learning rate, the weight matrix and the bias vector of the GA-BPNN model are all obtained through genetic algorithm training; the optimal learning rate, the weight matrix and the bias vector corresponding to the chromosome with the maximum fitness value in the genetic algorithm training process are the learning rate, the weight matrix and the bias vector of the GA-BPNN model;
and 3, outputting a fatigue crack growth rate value.
The invention further improves that:
preferably, the acoustic emission parameters include count rate, amplitude rate, energy rate and information entropy rate; the acoustic emission parameters are all obtained through calculation by a line cutting method.
Preferably, the fatigue condition parameters include average stress to stress ratio.
Preferably, the number of neurons of the input layer in the GA-BPNN model is the sum of the number of acoustic emission parameters and the number of fatigue working condition parameters.
Preferably, in step 2, determining the number of hidden layers and the number of neurons in each layer in the GA-BPNN model by a parameter test method; the activation function of the GA-BPNN model is the LeakyReLU function.
Preferably, in step 2, the step of training the GP-BPNN model by the genetic algorithm is:
step 2.1, setting parameters of a genetic algorithm; taking acoustic emission parameters and fatigue working condition parameters in the training set as input data of the GA-BPNN model, and taking fatigue crack growth rate values in the training set as output data of the GA-BPNN model;
step 2.2, training the GA-BPNN model through input data and output data, calculating the fitness value of each chromosome in the initial population according to the training result, and sequencing the chromosomes according to the fitness from high to low;
step 2.3, repeating the step 2.2, recoding the chromosome, and performing selection, crossing and mutation operations on the chromosome according to the fitness to obtain a new generation population;
step 2.4, decoding chromosomes in the new generation population, training a GA-BPNN model by using the learning rate, the weight matrix and the bias vector obtained by decoding, and calculating the fitness value of all the chromosomes in the new generation population according to the training result;
and 2.5, repeating the steps 2.3-2.4 until the iteration times meet the preset maximum iteration times, and selecting the chromosome with the maximum fitness as the optimal chromosome.
Preferably, the genetic algorithm parameters set before step 2.1 include maximum iteration number, initial population size, crossover probability and mutation probability.
Preferably, the calculation formula of the fitness value is:
where n is the number of samples in the training set, y i Is the true value of the logarithmic rate of fatigue crack growth in the training set, f (x i ) Is a predicted value of the logarithmic rate of fatigue crack growth.
Preferably, in step 2, after training the GA-BPNN model by the parameters of the training set, testing by the parameters of the testing set, and calculating the Root Mean Square Error (RMSE) and the determination coefficient (R) by the parameters of the testing set 2 Judging whether the model meets the requirements.
A fatigue crack growth rate prediction system based on acoustic emission monitoring and machine learning, comprising:
inputting a model, and acquiring acoustic emission parameters and fatigue working condition parameters;
the prediction model predicts the fatigue crack growth rate value through the fatigue crack growth rate prediction model based on the acoustic emission parameter and the fatigue working condition parameter; the fatigue crack propagation rate model is a GA-BPNN model, and the learning rate, the weight matrix and the bias vector of the GA-BPNN model are all obtained through genetic algorithm training; the optimal learning rate, the weight matrix and the bias vector corresponding to the chromosome with the maximum fitness value in the genetic algorithm training process are the learning rate, the weight matrix and the bias vector of the GA-BPNN model;
and outputting the model and outputting the fatigue crack growth rate value.
Compared with the prior art, the invention has the following beneficial effects:
the invention discloses a fatigue crack growth rate prediction method and a fatigue crack growth rate prediction system based on acoustic emission monitoring and machine learning, which are characterized in that a genetic algorithm is utilized to optimize learning rate, weight matrix and bias vector in a BPNN model, and acoustic emission counting rate, amplitude rate, energy rate and information entropy rate obtained in an online fatigue crack growth monitoring process are used as input data of a GA-BPNN model together with fatigue experiment working condition parameter average stress and stress ratio so as to predict corresponding fatigue crack growth rate. Compared with the traditional method for predicting the linear relation between the single acoustic emission parameter and the crack expansion rate, the method provided by the invention considers the influence of the multiple acoustic emission characteristic parameter and the fatigue experimental working condition parameter, realizes the construction of the mapping relation among the complex fatigue load factor, acoustic emission monitoring data and the fatigue crack expansion rate, reduces the prediction error and improves the prediction precision. In addition, by increasing the iteration times of the genetic algorithm, the chromosome can obtain the optimal solution as far as possible through repeated selection, crossing and mutation processes, so that the possibility of the algorithm to search for the optimal solution is improved. Therefore, the method is suitable for acoustic emission monitoring of fatigue crack growth under complex load factors, and has important engineering application value for realizing accurate and efficient prediction of fatigue crack growth rate.
Drawings
FIG. 1 is a flow chart of a fatigue crack growth rate prediction method based on acoustic emission monitoring and machine learning of the present invention;
FIG. 2 is a flow chart of the construction of a fatigue crack growth rate prediction model;
FIG. 3 is a schematic flow chart of GA-BPNN model training obtained by the machine learning-based fatigue crack growth rate prediction method of the present invention;
FIG. 4 is a graph comparing predicted values of fatigue crack growth log rates obtained using a test set with corresponding true values according to the obtained fatigue crack growth rate prediction method based on acoustic emission monitoring and machine learning in the present invention.
Detailed Description
The invention is described in further detail below with reference to the attached drawing figures:
the invention discloses a fatigue crack growth rate prediction method based on acoustic emission monitoring and machine learning. Firstly, through carrying out acoustic emission on-line monitoring of fatigue crack growth, the invention obtains the characteristic growth rate and the fatigue crack growth rate of each acoustic emission under different fatigue cycle times; next, constructing a fatigue crack growth data set to be predicted, wherein each group of data comprises 4 acoustic emission parameters, 2 fatigue working condition parameters and corresponding fatigue crack growth speed values, and randomly dividing the data set into a training set and a prediction set; thirdly, establishing a fatigue crack growth rate prediction model, and obtaining an optimal learning rate, an optimal weight matrix and an optimal bias vector of the BP neural network through a genetic algorithm to further obtain the fatigue crack growth rate prediction model in the form of a GA-BPNN model; and finally, verifying the accuracy of the established fatigue crack growth rate prediction model according to the input data in the test set. Compared with the traditional method, the method realizes the construction of the mapping relation among the complex fatigue load factors, the acoustic emission monitoring data and the fatigue crack expansion rate, and improves the prediction precision. The advantages of the learning rate, the weight matrix and the bias vector of the BPNN model are optimized by fully utilizing the genetic algorithm, and the method has the advantages of small error and high efficiency.
Referring to fig. 1, one of the embodiments of the present invention is to provide a fatigue crack growth rate prediction method based on acoustic emission monitoring and machine learning, which inputs a count rate, an amplitude rate, an energy rate, and an information entropy rate, as well as an average stress-stress ratio, at the time of actual prediction; and predicting the fatigue crack growth rate value through the trained fatigue crack growth rate prediction model, and finally outputting the predicted value of the fatigue crack growth rate.
Referring to fig. 2, one of the embodiments of the present invention provides a method for obtaining a fatigue crack growth rate prediction model, which specifically includes the following steps:
s1, carrying out acoustic emission on-line monitoring of fatigue crack growth, extracting multi-element acoustic emission characteristics from acoustic emission waveforms, and obtaining the growth rate of each acoustic emission characteristic and the fatigue crack growth rate under different fatigue cycle times;
the multielement acoustic emission characteristics comprise count, amplitude, energy and information entropy; the acoustic emission characteristic increase rate under different fatigue cycle cycles comprises a count rateAmplitude ratio->Energy Rate->Information entropy Rate->Fatigue crack growth rate->All are obtained by calculation through a thread cutting method:
wherein N is the number of fatigue cycles, C i+1 Is that the fatigue cycle number is N i+1 Acoustic emission count cumulative value at time C i Is that the fatigue cycle number is N i Acoustic emission count accumulation value at that time.
Wherein A is i+1 Is that the fatigue cycle number is N i+1 Cumulative value of acoustic emission amplitude at time, A i Is that the fatigue cycle number is N i The cumulative value of acoustic emission amplitude at that time.
Wherein E is i+1 Is that the fatigue cycle number is N i+1 Acoustic emission energy cumulative value at the time, E i Is that the fatigue cycle number is N i Acoustic emission energy cumulative value at that time.
Wherein SE is i+1 Is that the fatigue cycle number is N i+1 Entropy accumulation value, SE of acoustic emission information at time i Is fatigue cycle number ofN i And (5) accumulating the entropy of the acoustic emission information.
Wherein a is i+1 Is that the fatigue cycle number is N i+1 Crack size at time, a i Is that the fatigue cycle number is N i Crack size at time.
S2, constructing a fatigue crack growth data set to be predicted, wherein each group of data comprises 4 acoustic emission parameters, 2 fatigue working condition parameters and corresponding fatigue crack growth speed values, randomly dividing the data set into a training set and a prediction set, and preprocessing the data;
the acoustic emission parameters in S2 comprise acoustic emission count rate, amplitude rate, energy rate and information entropy rate under different fatigue cycle times; fatigue operating parameters include average stress and stress ratio; the randomly divided training set and prediction set account for 70% and 30% of the total data set, respectively; all data in the dataset are subjected to a logarithmic transformation preprocessing operation.
S3: establishing a fatigue crack growth rate prediction model, and obtaining an optimal learning rate, an optimal weight matrix and an optimal bias vector of the BP neural network through a genetic algorithm to further obtain the fatigue crack growth rate prediction model in the form of a GA-BPNN model; the specific step S3 includes:
s31: the method for determining the structure and the activation function of the BP neural network model specifically comprises the following steps: setting the number of neurons of an input layer of the GA-BPNN model to be the same as the total category number of acoustic emission characteristic parameters and fatigue experiment working condition parameters, and setting the number of neurons of an output layer to be 1; determining the number of hidden layers and the number of neurons in each layer of the GA-BPNN model by a parameter test method; the LeakyReLU function is selected as the activation function for the BPNN model.
S32: setting parameters of a genetic algorithm, which specifically comprises: setting the maximum iteration times, initial population scale, cross probability and variation probability; the chromosomes are encoded according to the learning rate, the weight matrix of the neural network and the order of the bias vector.
S33: and taking the acoustic emission parameters and the fatigue test working condition parameters in the training set as input data together, and taking the fatigue crack growth rate corresponding to the training set as output data.
S34: and training the GA-BPNN model by using input data and output data in the training set, calculating the fitness value of the chromosome according to the training result, selecting the chromosome with the largest fitness value as an optimal chromosome, and taking the position corresponding to the optimal chromosome as the optimal learning rate, the optimal weight matrix and the optimal bias vector.
The fitness value F has the following calculation formula:
where n is the number of samples in the training set, y i Is the true value of the logarithmic rate of fatigue crack growth in the training set, f (x i ) Is a predicted value of the logarithmic rate of fatigue crack growth.
The step S34 includes:
s341: decoding each chromosome in the initial population, inputting the learning rate, the weight matrix and the bias vector obtained by decoding the chromosomes into a GA-BPNN model, training the GA-BPNN model by using input data and output data in a training set, calculating the fitness value of each chromosome in the initial population according to a training result, and sequencing the chromosomes according to the fitness from high to low.
S342: and adding one to the iteration number, recoding the chromosome, and selecting, crossing and mutating the chromosome according to the fitness to obtain a new generation population.
S343: and decoding the chromosomes in the new generation population again, training the GA-BPNN model by using the learning rate, the weight matrix and the bias vector obtained by decoding, calculating the fitness values of all the chromosomes in the new generation population according to the training result, and judging whether the iteration times meet the preset maximum iteration times.
S344: and repeating the step S342 and the step S343 until the iteration times meet the preset maximum iteration times, selecting the chromosome with the largest fitness as the optimal chromosome, and obtaining the position corresponding to the optimal chromosome as the optimal learning rate, the optimal weight matrix and the optimal bias vector.
S35: and setting the optimal learning rate, the optimal weight matrix and the optimal bias vector as the learning rate, the weight matrix and the bias vector of the GA-BPNN model, so as to obtain the fatigue crack propagation rate prediction model.
S4: and verifying the accuracy of the established fatigue crack growth rate prediction model according to the input data in the test set.
The step S4 includes: the acoustic emission characteristic and the fatigue test working condition parameter in the test set are taken as input data together, the input data is predicted by utilizing the obtained GA-BPNN fatigue crack growth rate prediction model, the predicted value of the fatigue crack growth rate is obtained, the predicted value is compared with the corresponding true value, and the Root Mean Square Error (RMSE) and the determination coefficient (R) of the fatigue crack growth rate prediction model in the test set are calculated 2 The accuracy of the model was checked.
Preferably, the root mean square error RMSE is:
preferably, said decision coefficient R 2 The method comprises the following steps:
where n' is the number of data samples in the test set, y j Is the true value of the logarithmic rate of fatigue crack growth in the test set, y is the average of the logarithmic rates of fatigue crack growth for all samples in the test set, f (x) j ) Is a predicted value of the logarithmic rate of fatigue crack growth.
One of the embodiments of the present invention is to disclose a fatigue crack growth rate prediction model based on acoustic emission monitoring and machine learning, the model comprising:
inputting a model, and acquiring acoustic emission parameters and fatigue working condition parameters;
a prediction model for predicting a fatigue crack growth rate value through the fatigue crack growth rate prediction model; the fatigue crack propagation rate model is a GA-BPNN model, and the learning rate, the weight matrix and the bias vector of the GA-BPNN model are all obtained through genetic algorithm training; the optimal learning rate, the weight matrix and the bias vector corresponding to the chromosome with the maximum fitness value in the genetic algorithm training process are the learning rate, the weight matrix and the bias vector of the GA-BPNN model;
and outputting the model and outputting the fatigue crack growth rate value.
Examples
In the embodiment, the adopted experimental material is 316LN stainless steel, the sample form is a three-point bending sample, the fatigue crack growth experiment is completed in an MTS universal tester, the stress ratio of the fatigue experiment is set to be 0.1, 0.3 and 0.5, and the acoustic emission signal generated by fatigue crack growth is acquired through a PAC acoustic emission monitoring system; the data set used for model training and prediction was from a total of 6 samples, and experiments were performed using 2 samples for each stress ratio, giving 185 sets of data.
Referring to fig. 1, the fatigue crack growth rate prediction method based on acoustic emission monitoring and machine learning according to the invention comprises the following steps:
step S1: developing acoustic emission on-line monitoring of fatigue crack growth, extracting multi-element acoustic emission characteristics from acoustic emission waveforms, and obtaining the growth rate of each acoustic emission characteristic and the fatigue crack growth rate under different fatigue cycle times;
wherein the extracted acoustic emission characteristic parameters comprise count, amplitude, energy and information entropy; acoustic emission characteristic growth rates at different cycles of fatigue including count rateAmplitude ratio->Energy Rate->Information entropy Rate->Fatigue crack growth rate->All are obtained by calculation through a thread cutting method:
wherein N is the number of fatigue cycles, C i+1 Is that the fatigue cycle number is N i+1 Acoustic emission count cumulative value at time C i Is that the fatigue cycle number is N i Acoustic emission count accumulation value at that time.
Wherein A is i+1 Is that the fatigue cycle number is N i+1 Cumulative value of acoustic emission amplitude at time, A i Is that the fatigue cycle number is N i The cumulative value of acoustic emission amplitude at that time.
Wherein E is i+1 Is that the fatigue cycle number is N i+1 Acoustic emission energy cumulative value at the time, E i Is that the fatigue cycle number is N i Acoustic emission energy cumulative value at that time.
Wherein SE is i+1 Is that the fatigue cycle number is N i+1 Entropy accumulation value, SE of acoustic emission information at time i Is that the fatigue cycle number is N i And (5) accumulating the entropy of the acoustic emission information.
Wherein a is i+1 Is that the fatigue cycle number is N i+1 Crack size at time, a i Is that the fatigue cycle number is N i Crack size at time.
Step S2: constructing a fatigue crack growth data set to be predicted, wherein each group of data comprises 4 acoustic emission parameters, 2 fatigue working condition parameters and corresponding fatigue crack growth speed values, randomly dividing the data set into a training set and a prediction set, and preprocessing the data;
the acoustic emission parameters comprise acoustic emission counting rate, amplitude rate, energy rate and information entropy rate under different fatigue cycle times, and are obtained through calculation of acoustic emission monitoring data; the fatigue working condition parameters comprise average stress and stress ratio, which are set manually before the experiment; the dataset containing 185 samples was randomly partitioned into a 70% training set and a 30% prediction set, and all acoustic emission features and fatigue crack growth rates in the dataset were logarithmically transformed.
Step S3: establishing a fatigue crack growth rate prediction model, and obtaining an optimal learning rate, an optimal weight matrix and an optimal bias vector of the BP neural network through a GA genetic algorithm to further obtain the fatigue crack growth rate prediction model in the form of a GA-BPNN model;
the learning rate, the weight matrix and the bias vector code of the BPNN model form chromosomes in the GA genetic algorithm, and the fitness value of each chromosome in the population is calculated according to the predicted value (i.e. the predicted value of the fatigue crack growth rate logarithmic value) and the true value (i.e. the logarithmic value of the fatigue crack growth rate obtained in step S1) of the BPNN model, so that the optimal learning rate, the weight matrix and the bias vector of the BPNN model are obtained through the genetic algorithm. In this embodiment, the chromosome is subjected to operations such as selection, crossover, mutation, and the like, so that the likelihood that the genetic algorithm seeks an optimal solution can be improved through a certain number of iterative evolutions.
Referring to fig. 2, the step S3 specifically includes the following steps:
step S31: the method for determining the structure and the activation function of the BP neural network model specifically comprises the following steps: setting the number of neurons of an input layer of the GA-BPNN model to be the same as the total category number of acoustic emission characteristic parameters and fatigue experiment working condition parameters (namely, the number of neurons is 6), and setting the number of neurons of an output layer to be 1 (corresponding to the logarithmic value of fatigue crack growth rate); the number of hidden layers and the number of neurons of each hidden layer of the GA-BPNN model are determined through a parameter test method, and the result shows that when the number of hidden layers is 3, and the number of neurons of each layer is 3, 5 and 5, the root mean square error of the model is minimum, so that the structure of the BP neural network model is 6-3-5-5-1 in the embodiment; the LeakyReLU function is selected as the activation function for the BPNN model.
Step S32: the parameter initializing genetic algorithm specifically comprises the following steps: setting the maximum iteration number to 10, setting the initial population scale to 100, and setting the crossover probability and the variation probability to 0.3 and 0.02 respectively; the chromosomes are encoded according to the learning rate, the weight matrix of the neural network and the order of the bias vector.
S33: the acoustic emission data (namely the acoustic emission count rate, the amplitude rate, the energy rate and the information entropy rate) and the fatigue test working condition parameters (namely the average stress and stress ratio) in the training set are taken as input data together, and the corresponding fatigue crack growth rate logarithmic value in the training set is taken as output data.
S34: and training the GA-BPNN model by using input data and output data in the training set, calculating the fitness value of the chromosome according to the training result, selecting the chromosome with the largest fitness value as an optimal chromosome, and taking the position corresponding to the optimal chromosome as the optimal learning rate, the optimal weight matrix and the optimal bias vector.
In this embodiment, the training result is a predicted value of the fatigue crack propagation rate logarithmic value, and the fitness value F is calculated according to the formula:
where n is the number of samples in the training set, y i Is the true fatigue crack growth rate logarithmic value in the training set, f (x i ) Is the predicted fatigue crack growth rate logarithmic value, i.e., the model training result described above.
The step S34 specifically includes:
step S341: decoding each chromosome in the initial population, inputting the learning rate, the weight matrix and the bias vector obtained by decoding the chromosomes into a GA-BPNN model, training the GA-BPNN model by using input data and output data in a training set, calculating the fitness value of each chromosome in the initial population according to a training result, and sequencing the chromosomes according to the fitness from high to low.
Step S342: and adding one to the iteration number, recoding the chromosome, and selecting, crossing and mutating the chromosome according to the fitness to obtain a new generation population.
Step S343: and decoding the chromosomes in the new generation population again, training the GA-BPNN model by using the learning rate, the weight matrix and the bias vector obtained by decoding, calculating the fitness values of all the chromosomes in the new generation population according to the training result, and judging whether the iteration times meet the preset maximum iteration times.
Step S344: and repeating the step S342 and the step S343 until the iteration times meet the preset maximum iteration times, selecting the chromosome with the optimal fitness (namely, the maximum fitness) as the optimal chromosome, and obtaining the optimal chromosome with the optimal learning rate, the optimal weight matrix and the optimal bias vector at the position corresponding to the obtained optimal chromosome.
Step S35: and setting the optimal learning rate, the optimal weight matrix and the optimal bias vector as the learning rate, the weight matrix and the bias vector of the GA-BPNN model, so as to obtain the fatigue crack propagation rate prediction model.
Step S4: and verifying the accuracy of the established fatigue crack growth rate prediction model according to the input data in the test set.
The step S4 specifically comprises the following steps: the acoustic emission characteristic and the fatigue test working condition parameter in the test set are taken as input data together, the input data is predicted by utilizing the obtained GA-BPNN fatigue crack growth rate prediction model, the predicted value of the fatigue crack growth rate is obtained, the predicted value is compared with the corresponding true value, and the Root Mean Square Error (RMSE) and the determination coefficient (R) of the fatigue crack growth rate prediction model in the test set are calculated 2 The accuracy of the model was checked.
The root mean square error RMSE calculation formula is as follows:
said decision coefficient R 2 The formula is calculated as follows:
where n' is the number of data samples in the test set, y j Is a true value of the logarithmic rate of fatigue crack growth in the test set,is the average of the logarithmic rate of fatigue crack growth, f (x) j ) Is a predicted value of the logarithmic rate of fatigue crack growth.
With this embodiment, the following predicted results can be obtained:
FIG. 3 is a graph comparing fatigue crack growth rate versus value predicted based on the GA-BPNN model with a true value. Wherein the dashed line indicates that the predicted value is consistent with the true value, and the closer the data point is to the dashed line, the higher the model prediction accuracy is. Quantification of prediction accuracy by means of root mean square error RMSE and decision coefficient R 2 The embodiment is that the root mean square error RMSE and the determination coefficient R are calculated 2 The volumes are 0.617 and 0.308, respectively, and basically meet the expectations, which indicates that the prediction result is good. As can be seen in FIG. 3, most of the data points fall near the dashed lineHowever, some data points are far away from the dotted line, so that the data points show larger dispersion, and the prediction accuracy can be further improved by increasing the sample size of the data set.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.
Claims (10)
1. The fatigue crack growth rate prediction method based on acoustic emission monitoring and machine learning is characterized by comprising the following steps of:
step 1, acquiring acoustic emission parameters and fatigue working condition parameters;
step 2, predicting a fatigue crack growth rate value through a fatigue crack growth rate prediction model based on the acoustic emission parameter and the fatigue working condition parameter; the fatigue crack propagation rate model is a GA-BPNN model, and the learning rate, the weight matrix and the bias vector of the GA-BPNN model are all obtained through genetic algorithm training; the optimal learning rate, the weight matrix and the bias vector corresponding to the chromosome with the maximum fitness value in the genetic algorithm training process are the learning rate, the weight matrix and the bias vector of the GA-BPNN model;
and 3, outputting a fatigue crack growth rate value.
2. The method for predicting the fatigue crack growth rate based on acoustic emission monitoring and machine learning as claimed in claim 1, wherein the acoustic emission parameters include a count rate, an amplitude rate, an energy rate and an information entropy rate; the acoustic emission parameters are all obtained through calculation by a line cutting method.
3. The method for predicting the fatigue crack growth rate based on acoustic emission monitoring and machine learning as set forth in claim 1, wherein the fatigue condition parameters include average stress and stress ratio.
4. The fatigue crack growth rate prediction method based on acoustic emission monitoring and machine learning according to claim 1, wherein the number of neurons of the input layer in the GA-BPNN model is the sum of acoustic emission parameters and the number of fatigue condition parameters.
5. The fatigue crack growth rate prediction method based on acoustic emission monitoring and machine learning according to claim 1, wherein in step 2, the number of hidden layers and the number of neurons in each layer in the GA-BPNN model are determined by a trial-and-reference method; the activation function of the GA-BPNN model is the LeakyReLU function.
6. The fatigue crack growth rate prediction method based on acoustic emission monitoring and machine learning according to claim 1, wherein in step 2, the step of training the GP-BPNN model by the genetic algorithm is:
step 2.1, setting parameters of a genetic algorithm; taking acoustic emission parameters and fatigue working condition parameters in the training set as input data of the GA-BPNN model, and taking fatigue crack growth rate values in the training set as output data of the GA-BPNN model;
step 2.2, training the GA-BPNN model through input data and output data, calculating the fitness value of each chromosome in the initial population according to the training result, and sequencing the chromosomes according to the fitness from high to low;
step 2.3, repeating the step 2.2, recoding the chromosome, and performing selection, crossing and mutation operations on the chromosome according to the fitness to obtain a new generation population;
step 2.4, decoding chromosomes in the new generation population, training a GA-BPNN model by using the learning rate, the weight matrix and the bias vector obtained by decoding, and calculating the fitness value of all the chromosomes in the new generation population according to the training result;
and 2.5, repeating the steps 2.3-2.4 until the iteration times meet the preset maximum iteration times, and selecting the chromosome with the maximum fitness as the optimal chromosome.
7. The method for predicting fatigue crack growth rate based on acoustic emission monitoring and machine learning as set forth in claim 6, wherein the genetic algorithm parameters set before step 2.1 include maximum iteration number, initial population size, crossover probability and mutation probability.
8. The fatigue crack growth rate prediction method based on acoustic emission monitoring and machine learning according to claim 1, wherein the fitness value is calculated by the formula:
where n is the number of samples in the training set, y i Is the true value of the logarithmic rate of fatigue crack growth in the training set, f (x i ) Is a predicted value of the logarithmic rate of fatigue crack growth.
9. The method for predicting fatigue crack growth rate based on acoustic emission monitoring and machine learning as set forth in claim 1, wherein in step 2, after training the GA-BPNN model by the parameters of the training set, testing by the parameters of the testing set, and calculating the root mean square error RMSE and the determination coefficient R by the parameters of the testing set 2 Judging whether the model meets the requirements.
10. A fatigue crack growth rate prediction system based on acoustic emission monitoring and machine learning, comprising:
inputting a model, and acquiring acoustic emission parameters and fatigue working condition parameters;
the prediction model predicts the fatigue crack growth rate value through the fatigue crack growth rate prediction model based on the acoustic emission parameter and the fatigue working condition parameter; the fatigue crack propagation rate model is a GA-BPNN model, and the learning rate, the weight matrix and the bias vector of the GA-BPNN model are all obtained through genetic algorithm training;
the optimal learning rate, the weight matrix and the bias vector corresponding to the chromosome with the maximum fitness value in the genetic algorithm training process are the learning rate, the weight matrix and the bias vector of the GA-BPNN model;
and outputting the model and outputting the fatigue crack growth rate value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310496808.2A CN116539459A (en) | 2023-05-05 | 2023-05-05 | Fatigue crack growth rate prediction method and system based on acoustic emission monitoring and machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310496808.2A CN116539459A (en) | 2023-05-05 | 2023-05-05 | Fatigue crack growth rate prediction method and system based on acoustic emission monitoring and machine learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116539459A true CN116539459A (en) | 2023-08-04 |
Family
ID=87442943
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310496808.2A Pending CN116539459A (en) | 2023-05-05 | 2023-05-05 | Fatigue crack growth rate prediction method and system based on acoustic emission monitoring and machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116539459A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117669390A (en) * | 2024-02-01 | 2024-03-08 | 中国石油大学(华东) | Metal full-stage fatigue crack growth prediction method and system based on neural network |
-
2023
- 2023-05-05 CN CN202310496808.2A patent/CN116539459A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117669390A (en) * | 2024-02-01 | 2024-03-08 | 中国石油大学(华东) | Metal full-stage fatigue crack growth prediction method and system based on neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ma et al. | Discriminative deep belief networks with ant colony optimization for health status assessment of machine | |
| CN116539459A (en) | Fatigue crack growth rate prediction method and system based on acoustic emission monitoring and machine learning | |
| CN111950099A (en) | Method, system, medium and computer equipment for testing mechanical property of equipment material | |
| CN112364560B (en) | Intelligent prediction method for working hours of mine rock drilling equipment | |
| CN110555247A (en) | structure damage early warning method based on multipoint sensor data and BilSTM | |
| Nasiri et al. | Online damage monitoring of SiC f-SiC m composite materials using acoustic emission and deep learning | |
| CN106896219A (en) | The identification of transformer sub-health state and average remaining lifetime method of estimation based on Gases Dissolved in Transformer Oil data | |
| CN110633504A (en) | Prediction method for coal bed gas permeability | |
| CN115063051B (en) | Steel pipe performance evaluation method and system based on metal element analysis | |
| CN113344261A (en) | PM2.5 concentration prediction method and system based on random forest and ISCA (interference Signal code analysis) optimization RELM (residual stress modeling) | |
| CN115982141A (en) | Characteristic optimization method for time series data prediction | |
| CN114638164B (en) | Method for predicting high-temperature creep life of power station pressure pipeline | |
| CN112084709A (en) | Large-scale generator insulation state evaluation method based on genetic algorithm and radial basis function neural network | |
| CN117233347B (en) | Carbon steel spheroidization grade measuring method, system and equipment | |
| CN107977672A (en) | SF6 equipment secondary failure diagnostic methods based on mass data concurrent operation | |
| CN112613105B (en) | Reinforced concrete beam damage state identification method based on acoustic emission parameter and genetic algorithm optimization BP neural network | |
| CN111400964B (en) | Fault occurrence time prediction method and device | |
| CN111291518B (en) | BP neural network-based breaker cutter wear degree judging method | |
| CN114265764A (en) | Deep learning operator test data generation method based on weighted sampling | |
| CN111861041A (en) | Method for predicting dynamic recrystallization type flowing stress of Nb microalloyed steel | |
| CN112465347A (en) | Method for cooperatively predicting roof stability based on cluster analysis and improved neural network | |
| CN113533967B (en) | Binary optimized battery anomaly detection method for panoramic theory analysis | |
| CN117149551B (en) | Test method of vehicle-mounted wireless communication chip | |
| CN115936166B (en) | Electric energy meter verification error analysis and prediction method | |
| CN116106430A (en) | Acoustic emission technology-based refractory material cracking diagnosis method for casting |
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 |