CN115081321B - Corrosion fatigue life prediction method, system and equipment for marine welding structure - Google Patents
Corrosion fatigue life prediction method, system and equipment for marine welding structure Download PDFInfo
- Publication number
- CN115081321B CN115081321B CN202210675768.3A CN202210675768A CN115081321B CN 115081321 B CN115081321 B CN 115081321B CN 202210675768 A CN202210675768 A CN 202210675768A CN 115081321 B CN115081321 B CN 115081321B
- Authority
- CN
- China
- Prior art keywords
- fatigue life
- fatigue
- data set
- welding structure
- performance data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—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
- 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
-
- 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
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Investigating Strength Of Materials By Application Of Mechanical Stress (AREA)
Abstract
The invention relates to a corrosion fatigue life prediction method and a corrosion fatigue life prediction system for an ocean welding structure, which relate to the field of fatigue life prediction of engineering parts, and the method comprises the following steps: each sample data in the fatigue performance data set comprises a plurality of fatigue life influencing factors of the marine welding structure and the fatigue life of the marine welding structure; acquiring a category unbalanced data area of each fatigue life influencing factor in the fatigue performance data set, and performing data enhancement on the category unbalanced data area of each fatigue life influencing factor; determining the weight of each fatigue life influence factor by adopting an extreme gradient lifting algorithm; training a deep convolutional neural network model based on the fatigue performance data set after data enhancement; the deep convolutional neural network model is a deep convolutional neural network model in which the weight of each fatigue life influencing factor is embedded through an attention mechanism. The method improves the generalization capability of the service life prediction of the marine welding structure based on the data-driven algorithm.
Description
Technical Field
The invention relates to the field of fatigue life prediction of engineering components, in particular to a corrosion fatigue life prediction method, a corrosion fatigue life prediction system and corrosion fatigue life prediction equipment for an ocean welding structure.
Background
In the past, due to the reasons of large size, complex structure, complex service condition and the like, the welded structure of ships and ocean platforms has large discreteness of fatigue behaviors, and system analysis is difficult to perform. Particularly, the marine welding structure is influenced by seawater corrosion, compared with the fatigue behavior under the non-corrosive condition, the difference is large, and the corrosion fatigue damage is often larger than the sum of the damages caused by corrosion and fatigue respectively, so that the prediction of the fatigue life of the marine welding structure under the seawater corrosion condition has very important guiding effect on the fatigue performance research and the fatigue design of the marine welding structure.
Generally, the fatigue test of the marine welding structure has large consumption of manpower and material resources, and the fatigue performance research of large-size structural parts has great difficulty. At present, the fatigue life prediction result developed based on the artificial intelligence technology shows that the hybrid intelligent algorithm is an effective means for accurately predicting the fatigue life. However, at present, the fatigue life is predicted directly under a certain stress level condition based on a machine learning method, but due to the discreteness of the fatigue performance and the complexity under the corrosion condition, the error of the prediction result is very large, so that the fatigue performance of the marine welding structure is difficult to describe stably.
Particularly, at present, the fatigue life prediction based on a data-driven algorithm is often limited by more conditions, and a fatigue database is not rich enough, so that the stability of a prediction model, particularly the generalization capability, is poor, and the method is difficult to be widely applied to fatigue life prediction under different conditions.
Disclosure of Invention
The invention aims to provide a method, a system and equipment for predicting the corrosion fatigue life of an ocean welding structure, and the generalization capability of the life prediction of the ocean welding structure is improved.
In order to achieve the purpose, the invention provides the following scheme:
a method for predicting corrosion fatigue life of a marine welded structure, comprising:
constructing a fatigue performance data set of the marine welding structure, wherein each sample data in the fatigue performance data set comprises a plurality of fatigue life influencing factors of the marine welding structure and the fatigue life of the marine welding structure;
traversing each fatigue life influencing factor, acquiring a category unbalanced data area of each fatigue life influencing factor in the fatigue performance data set, and performing data enhancement on the category unbalanced data area of each fatigue life influencing factor to obtain a fatigue performance data set after data enhancement;
determining the weight of each fatigue life influence factor by adopting an extreme gradient lifting algorithm based on the fatigue performance data set after data enhancement;
training a deep convolutional neural network model based on the fatigue performance data set after the data enhancement to obtain a corrosion fatigue life prediction model of the marine welding structure; the deep convolutional neural network model is a deep convolutional neural network model which is embedded with the weight of each fatigue life influence factor through an attention mechanism;
and predicting the service life of the marine welding structure to be predicted by adopting the corrosion fatigue life prediction model of the marine welding structure.
Optionally, the fatigue life influencing factors include material type, stress ratio, loading frequency, load type, surface state and temperature.
Optionally, the traversing each fatigue life influencing factor, obtaining a category unbalanced data region of each fatigue life influencing factor in the fatigue performance data set, and performing data enhancement on the category unbalanced data region of each fatigue life influencing factor to obtain the fatigue performance data set after data enhancement specifically includes:
and determining the type unbalanced data area of each fatigue life influencing factor in the obtained fatigue performance data set by using an edge oversampling method, and interpolating the type unbalanced data area of each fatigue life influencing factor to obtain the fatigue performance data set after data enhancement.
Optionally, the training of the deep convolutional neural network model based on the fatigue performance data set after data enhancement to obtain a corrosion fatigue life prediction model of the marine welding structure specifically includes:
dividing the fatigue performance data set after the data enhancement into a training set and a test set;
and training the deep convolutional neural network model by adopting the training set and taking a plurality of fatigue life influencing factors as input and the fatigue life of the marine welding structure as output, and testing and optimizing the trained deep convolutional neural network model by adopting the testing set until the error of the predicted value of the output of the deep convolutional neural network model reaches a preset condition to obtain a corrosion fatigue life prediction model of the marine welding structure.
The invention discloses a corrosion fatigue life prediction system of an ocean welding structure, which comprises:
the system comprises a fatigue performance data set construction module, a data acquisition module and a data processing module, wherein the fatigue performance data set construction module is used for constructing a fatigue performance data set of the marine welding structure, and each sample data in the fatigue performance data set comprises a plurality of fatigue life influencing factors of the marine welding structure and the fatigue life of the marine welding structure;
the fatigue performance data set data enhancement module is used for traversing each fatigue life influence factor, acquiring the category unbalanced data area of each fatigue life influence factor in the fatigue performance data set, and performing data enhancement on the category unbalanced data area of each fatigue life influence factor to obtain a fatigue performance data set after data enhancement;
the weight determining module of each fatigue life influencing factor is used for determining the weight of each fatigue life influencing factor by adopting an extreme gradient lifting algorithm based on the fatigue performance data set after the data enhancement;
the deep convolution neural network model training module is used for training a deep convolution neural network model based on the fatigue performance data set after the data enhancement to obtain a corrosion fatigue life prediction model of the marine welding structure; the deep convolutional neural network model is a deep convolutional neural network model which is embedded with the weight of each fatigue life influence factor through an attention mechanism;
and the corrosion fatigue life prediction module of the marine welding structure is used for predicting the life of the marine welding structure to be predicted by adopting the corrosion fatigue life prediction model of the marine welding structure.
Optionally, the fatigue life influencing factors include material type, stress ratio, loading frequency, load type, surface state and temperature.
Optionally, the fatigue performance data set data enhancement module specifically includes:
and the fatigue performance data set data enhancement unit is used for determining and acquiring the type unbalanced data area of each fatigue life influence factor in the fatigue performance data set by using an edge oversampling method, and interpolating the type unbalanced data area of each fatigue life influence factor to obtain the fatigue performance data set after data enhancement.
Optionally, the deep convolutional neural network model training module specifically includes:
the data set dividing unit is used for dividing the fatigue performance data set after the data enhancement into a training set and a test set;
and the deep convolutional neural network model training unit is used for training the deep convolutional neural network model by using the training set and taking a plurality of fatigue life influence factors as input and the fatigue life of the marine welding structure as output, and testing and optimizing the trained deep convolutional neural network model by using the test set until the error of the predicted value of the output of the deep convolutional neural network model reaches a preset condition to obtain a corrosion fatigue life prediction model of the marine welding structure.
The invention also discloses corrosion fatigue life prediction equipment of the marine welding structure, which comprises at least one processor, at least one memory and computer program instructions stored in the memory, wherein the computer program instructions realize the corrosion fatigue life prediction method of the marine welding structure when being executed by the processor.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention discloses a corrosion fatigue life prediction method, a corrosion fatigue life prediction system and corrosion fatigue life prediction equipment for an ocean welding structure.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
FIG. 1 is a schematic flow chart of a corrosion fatigue life prediction method for an ocean welding structure according to the present invention;
FIG. 2 is a schematic flow chart of a corrosion fatigue life prediction method for an ocean welding structure according to the present invention;
FIG. 3 is a diagram illustrating the F value and TP rate results after data enhancement according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of the weight analysis result of the fatigue performance influence factors of the marine welding structure based on the XGboost algorithm in the embodiment of the invention;
FIG. 5 is a diagram illustrating the necessity analysis results of the LIME and SHAP methods for influencing factors according to the embodiment of the present invention;
FIG. 6 is a diagram illustrating the results of analysis of the sufficiency of influence factors based on the LIME and SHAP methods in the embodiment of the present invention;
FIG. 7 is a diagram illustrating the result of an important index analysis of influencing factors based on feature attribution and counterfactual interpretation according to an embodiment of the present invention;
FIG. 8 is a diagram illustrating the correlation analysis of influence factors based on feature attribution and counterfactual interpretation according to an embodiment of the present invention;
FIG. 9 is a precision curve of a marine welded structure corrosion fatigue life prediction model based on a DCNN method according to an embodiment of the present invention;
FIG. 10 is a loss curve of a marine welding structure corrosion fatigue life prediction model based on the DCNN method according to an embodiment of the present invention;
FIG. 11 is a schematic diagram showing the comparison of the predicted results of the corrosion fatigue life prediction model of the marine welded structure of the present invention with the predicted results of other prediction methods;
FIG. 12 is a schematic diagram showing comparison of the predicted results of the corrosion fatigue life prediction model of the marine welded structure of the present invention with the predicted errors of other prediction methods;
FIG. 13 is a graphical representation of the results of a return capability analysis of a corrosion fatigue life prediction model of a marine welded structure in accordance with the present invention;
FIG. 14 is a first comparison diagram of the predicted results under the condition of different number of influencing factors according to the present invention;
FIG. 15 is a comparison diagram of the predicted results under the condition of different number of influencing factors according to the present invention;
FIG. 16 is a schematic diagram of a corrosion fatigue life prediction system for a marine welded structure according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a method, a system and equipment for predicting the corrosion fatigue life of an ocean welding structure, and the generalization capability of the life prediction of the ocean welding structure is improved.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Fig. 1 is a schematic flow chart of a corrosion fatigue life prediction method of an ocean welding structure of the invention, fig. 2 is a schematic flow chart of a corrosion fatigue life prediction method of an ocean welding structure of the invention, and as shown in fig. 1-2, a corrosion fatigue life prediction method of an ocean welding structure comprises:
step 101: and constructing a fatigue performance data set of the marine welding structure, wherein each sample data in the fatigue performance data set comprises a plurality of fatigue life influencing factors of the marine welding structure and the fatigue life of the marine welding structure.
Step 102: and traversing each fatigue life influencing factor, acquiring the category unbalanced data area of each fatigue life influencing factor in the fatigue performance data set, and performing data enhancement on the category unbalanced data area of each fatigue life influencing factor to obtain the fatigue performance data set after data enhancement.
Fatigue life influencing factors include the material type, stress ratio, loading frequency, load type, surface state, temperature, corrosive environment and joint type of the marine welded structure. The surface state is whether the corresponding work piece of the ocean welding structure is subjected to heat treatment, cold working and other treatment modes. The loading frequency is the loading frequency of the fatigue test.
Wherein, step 102 specifically comprises:
and determining a category unbalanced data area of each fatigue life influencing factor in the obtained fatigue performance data set by using an edge oversampling method, and interpolating the category unbalanced data area of each fatigue life influencing factor to obtain the fatigue performance data set with enhanced data.
Compared with a conventional welding joint, the marine welding structure is influenced by more complex factors, and different influencing factors influence the fatigue performance of the marine welding structure to different degrees. At present, most corrosion fatigue life prediction methods are provided based on databases under specific conditions, the generalization capability is poor, and the further application of most artificial intelligence prediction methods is limited. Many variables in the fatigue process interfere with fatigue life, resulting in database class imbalance problems. Therefore, in order to overcome the data category imbalance and avoid data overlap, a data category imbalance area is found by using a Borderline-SMOTE (Borderline-SMOTE) method and interpolation is carried out, so that the effectiveness of the database is enhanced.
Determining and acquiring a category unbalanced data area of each fatigue life influence factor in the fatigue performance data set by using an edge oversampling method, and interpolating the category unbalanced data area of each fatigue life influence factor to obtain a data-enhanced fatigue performance data set, specifically comprising:
the Borderline-SMOTE sampling process divides a few classes of samples into 3 classes, safe, danger and Noise, respectively. Only a few class samples whose table is Danger are oversampled.
Safe: more than half of the samples are of a few classes.
Danger: more than half of the samples around the sample are most samples of the class, and are considered as samples on the boundary.
Noise: the samples are all surrounded by a plurality of types of samples, and are considered as noise.
Step1: assume that the entire training set is S t The minority is P and the majority is N ma For each p s (s =1,2, …, p), calculate p s Set of distances S t M nearest neighbors, the number of majority class samples in the m nearest neighbors is represented by m '(0 ≦ m' ≦ m).
P＝{p 1 ，p 1 ，...，p p }，N ma ＝{n 1 ，n 1 ，...，n n }
Wherein n is the number of majority classes and p is the number of minority classes
Step2: if m' = m, then p s Considered as Noise (Noise), the following steps are no longer performed if m/2 ≦ m'<m, represents p s Is easily misclassified and is therefore classified as a hazard set D a (Danger), if 0 is not more than m'<m/2, then p s Considered Safe (Safe), the following steps are not performed.
Step3: collection
Where the samples are boundary data, for D a Respectively, calculating its distance P to the i neighbors.
D a ＝{p' 1 ，p' 2 ，...，p' d }，0≤d≤p
Step4: in this step, from D a Generation of e × d synthetic samples (1)<e<l), calculate p' s Difference (dif) from its e neighbors c ) And will dif c Multiplied by a random number r between 0 and 1 c And then e new synthesized few samples are generated.
synthetic c ＝p' s +r c ×dif j ，c＝1，2，...，e。
The F value represents the classification accuracy of the database (fatigue performance data set), and the TP rate is the percentage of correctly classified samples in the fatigue performance database, as shown in fig. 3, the higher F value and TP rate indicate that the classification accuracy and quality of the processed fatigue performance database are improved, which is helpful for realizing reliable corrosion fatigue life prediction.
Step 103: and determining the weight of each fatigue life influence factor by adopting an extreme gradient lifting algorithm based on the fatigue performance data set after data enhancement.
Wherein, step 103 specifically comprises: analyzing the weight of each fatigue life influence factor by using an extreme gradient boost (XGboost) algorithm based on the fatigue performance data set after data enhancement, and verifying the weight analysis result by using a cross verification method; and opening a black box of the analysis result of the weight analysis method of the influence factors by using a characteristic attribution and counterfactual interpretation method, and interpreting the weight analysis result. The explanation process comprises the following steps: the influence factors are scored or sequenced through a characteristic cause and counterfactual interpretation method, an XGboost weight analysis model (a model for determining the weight of each fatigue life influence factor by adopting an extreme gradient lifting algorithm) is interpreted, a local interpretable model unknown interpretation (LIME) model generates new samples through disturbance near related samples so as to obtain a predicted value of a black box model (the XGboost weight analysis model), and the interpretable model is trained so as to obtain a good local approximation with the black box model. The Shapley Supply interpretation (SHAP) method uses Shapley values to assign weights to the different influencing factors, which are calculated by averaging the marginal contributions of the features in different sets of all influencing factors. Based on the evaluation of the importance and the necessity of the interpretation, the XGboost weight analysis black box model is moderately opened and interpreted.
Related formula of XGboost algorithm:
the additive model is represented as:
wherein, b k (x) Is a base learner, M is the number of base learners, b k (x) The minimum risk value of (d) is expressed as:
where D represents a fatigue performance data set.
The XGBoost method is to use the difference of the second-order taylor expansion of the loss function as a learning target, which is equivalent to performing optimization by using a newton method to approach the minimum value of the loss function, that is, to make the loss function be 0. The above problems are understood on this basis. f. of k-1 (x) A first term corresponding to a second order Taylor expansion;
calculating the partial derivative of f (x); l (f) k-1 (x) And y) represents a second order taylor expansion.
Thus, the formula
Updated to formula->
The XGboost weight analysis model is expressed as
Where j and T are the current and total leaf node numbers, w, respectively j Is R j Is the index set (fatigue)Performance data set), R j Is a value range, the weight (w) of the leaf node can be represented by a formula
And (4) calculating. The leaf node is a hyper-parameter related to the influencing factor in the XGboost algorithm.
LIME is represented as
SHAP is expressed as
The significance of the LIME and SHAP based interpretation is expressed as
The sufficiency based on LIME and SHAP interpretation is expressed as:
wherein L is a Measurement of g L And f L At x and π x Approximate values of the neighborhood, x being the sample, f L Is a machine learning model, g L Is a local linear model, and Ω (g) is g L C is the counterfactual sample, y is the desired output, λ 1 And λ 2 Is a constant, N is the sum of the test cases that each time a counterfactual sample is generated, f (.) is the fraction of the number of times a unique counterfactual sample is generated, G represents the training data sample space, nCF represents the number of counterfactual samples. c. C i Refers to the ith counterfactual sample, CF i Also represents the ith counterfactual sample, f N (.) is the sum of all counterfactual sample test cases, a is a hyperparameter. x is the number of j Is the interference value of a.
As a specific embodiment, based on an extreme gradient boost (XGBoost) algorithm, the weights of the fatigue life influencing factors are respectively quantitatively analyzed, and the weight analysis results are verified by using 10-fold cross verification, and the weight analysis and verification results are shown in fig. 4, which shows that the stress ratio and the frequency are the two factors with the highest influence weight on the corrosion fatigue behavior. The analysis results based on the necessity and sufficiency of the LIME and SHAP methods on the influence factors are shown in FIGS. 5-6, and the analysis results based on the importance indexes and the relevance of the characteristic attribution and counterfactual explanation on the influence factors are shown in FIGS. 7-8, which indicate that the necessity and the weight of the influence factors are positively correlated, and the relationship between the two is weakened as the number of the influence factors is increased; further, even the highly important factors are often neither necessary nor sufficient, and the necessity and the sufficiency of the highly important factors become weaker as the number of influencing factors increases.
Step 104: training a deep convolutional neural network model based on the fatigue performance data set after data enhancement to obtain a corrosion fatigue life prediction model of the marine welding structure; the deep convolutional neural network model is a deep convolutional neural network model in which the weight of each fatigue life influencing factor is embedded through an attention mechanism.
Wherein, step 104 specifically includes:
dividing the fatigue performance data set after data enhancement into a training set and a testing set according to a certain proportion, specifically training set data: the test set data is 8:2.
Establishing a Deep Convolutional Neural Network (DCNN) model, embedding the weight of each fatigue life influence factor into the DCNN model through an attention mechanism, and initializing the parameters of the DCNN model.
The parameterized soft attention mechanism is a technique that enables a model to concentrate on important information and to adequately learn and absorb the information, wherein a kernel vector O in the attention mechanism is represented as:
wherein, N f Is the number of influencing factors considered (number of fatigue life influencing factors), w i The weight of the influence factor is obtained based on the XGboost algorithm, and g (.) is the recessive state of the influence factor. The hidden state is a hidden vector of the input features (influence factors), and is a vector formed by unobservable random variables, the hidden vector is an intermediate state of the input features in the calculation process, and the hidden vector can reduce the influence of insufficient learning of model cross terms, so that the prediction capability of the model is improved.
The method comprises the steps of adopting a training set, using a plurality of fatigue life influence factors as input through error propagation of a DCNN model, using fatigue life of the marine welding structure as output training deep convolution neural network models, and adopting a testing set to test and optimize the trained deep convolution neural network models until predicted value errors of the output of the deep convolution neural network models reach preset conditions, so as to obtain corrosion fatigue life prediction models of the marine welding structure, and realize accurate and stable prediction of the fatigue life of the marine welding structure.
Error (E) of prediction result through deep convolution neural network model, average error (E) a ) Standard deviation of error (S) e ) The accuracy, the stability and the generalization capability of the established corrosion fatigue life prediction model are verified by using a one-step generalization ratio (OSGR) and a gradient signal-to-noise ratio (GSNR).
E＝(N p -N a )/N a ×100％；
Wherein N is p And N a Respectively predicting fatigue life and actual fatigue life, n is the number of test samples, m train Is a sample in the training set, M train Is the training set of the model, t is the number of iteration steps, L is the training loss, L' is the test loss, g factor And ρ factor Respectively the mean and variance of the gradient calculated over all samples.
As a specific embodiment, the fatigue life is directly predicted and output by respectively inputting six fatigue life influencing factors of a sample, wherein the accuracy and the error curve of the training and verifying process are respectively shown in FIGS. 9-10.
Five fatigue life prediction models of SWT, IG, chu, performance base and ANN (artificial neural network) are selected to compare the built model (the corrosion fatigue life prediction model of the marine welding structure of the invention), according to the prediction results of different models shown in figures 11-12 and comparative analysis of prediction errors, the maximum error of the built model is not more than 15 percent and is far superior to that of other models, and the average prediction error of the built model is minimum (5.2 percent), 89.8 percent less than that of the SWT model, 91.4 percent less than that of the IG model, 91.3 percent less than that of the Chu model, 93.3 percent less than that of the performance attribute-based model and 92.9 percent less than that of the ANN model; in addition, the standard deviation of the prediction error of this model (4.1%) was also minimal, 91.3% less than the SWT model, 87.0% less than the IG model, 89.6% less than the Chu model, 90.7% less than the performance basis model, and 92.1% less than the ANN model. The above results indicate the significance of the model created in terms of accuracy as well as stability. According to the results of the model generalization ability analysis as shown in fig. 13, the GSNR of the built model significantly increased at the beginning of training and then continuously decreased, but was always superior to the linear regression model, which emphasizes the excellent generalization ability of the proposed fatigue life prediction model.
Among these, the SWT (Smith-Watson-Topper) model, reference R.N.Smith, P.Watson, T.H.Topper, stress strain function for the fault of metals, J Mater,5 (4) (1970), 767-778.
IG (Ince-Glinka) model, reference A.Ince, G.Glinka, A generated robust facial parameter for multi-axial facial life prediction unit delivery and non-delivery loads, int J Fativue, 62 (2014), 34-41,10.1016/j.ijfacial.2013.10.007.
Chu model, ref C.C.Chu, fatigue damage computers using the critical plane approach, JEngMater Technol,1995,10.1115/1.2804370.
Performance-based model, reference: zhang, C.N.Sun, X.Zhang, P.C.Goh, J.Wei, D.Hardacre, H.Li, high cycle failure life prediction of laser additive manufacturing stainlessheel A machine learning approach, int J failure 128 (2019), 105194,10.1016/j.ijfailure.2019.105194.
Two factors of stress ratio and frequency are ignored in a database, two factors of corrosion environment and joint type are added, the built model is tested by the same method, the prediction error result is shown in figures 14-15, the result shows that more influence factors are considered to be beneficial in the fatigue life prediction method based on deep learning, the built model has excellent generalization capability, the mixed prediction of pure mechanical fatigue and corrosion fatigue can be realized, good precision and stability are ensured, and the reliability and the advancement of the built corrosion fatigue life intelligent prediction and analysis method are also proved.
Step 105: and predicting the service life of the marine welding structure to be predicted by adopting a corrosion fatigue life prediction model of the marine welding structure.
Wherein, step 105 specifically comprises:
and inputting each fatigue life influence factor of the marine welding structure to be predicted into a corrosion fatigue life prediction model of the marine welding structure, and outputting a life prediction result.
The corrosion fatigue life prediction method of the marine welding structure is a fatigue life prediction and analysis method with enhanced generalization capability, and data amplification is carried out by using a Borderline-SMOTE method on the basis of collected fatigue performance data, so that the fatigue behavior characteristics of the marine welding structure can be more effectively reflected; the XGboost method is used for analyzing the weight of the fatigue life influencing factors, and the characteristic attribution and counterfactual interpretation method is used for interpreting the algorithm black box of the weight analysis model, so that the important degrees of different fatigue life influencing factors in fatigue behaviors can be better reflected; the fatigue life influence factor weighted values obtained through analysis are embedded into the DCNN model by using an attention model, so that the DCNN model can better understand and absorb the action rules of different influence factors, and more reliable fatigue life prediction is realized. The fatigue life prediction and analysis method can improve the limitations of poor interpretability, small application range, poor generalization capability and the like of the conventional fatigue life prediction method, provides a reliable method for predicting the fatigue life of the marine welded structure under the corrosion condition, provides an effective idea for researching the fatigue behavior of a data driving method, and further reduces a large amount of manpower and material resources required by the conventional fatigue performance test.
Fig. 16 is a schematic structural diagram of a corrosion fatigue life prediction system of an ocean welding structure, and as shown in fig. 16, the corrosion fatigue life prediction system of the ocean welding structure comprises:
the fatigue performance data set building module 201 is used for building a fatigue performance data set of the marine welding structure, and each sample data in the fatigue performance data set comprises a plurality of fatigue life influencing factors of the marine welding structure and the fatigue life of the marine welding structure.
The fatigue performance data set data enhancement module 202 is configured to traverse each fatigue life influencing factor, obtain a category unbalanced data region of each fatigue life influencing factor in the fatigue performance data set, and perform data enhancement on the category unbalanced data region of each fatigue life influencing factor to obtain a fatigue performance data set after data enhancement.
And the weight determining module 203 for each fatigue life influencing factor is used for determining the weight of each fatigue life influencing factor by adopting an extreme gradient lifting algorithm based on the fatigue performance data set after data enhancement.
The deep convolutional neural network model training module 204 is used for training a deep convolutional neural network model based on the fatigue performance data set after data enhancement to obtain a corrosion fatigue life prediction model of the marine welding structure; the deep convolutional neural network model is a deep convolutional neural network model in which the weight of each fatigue life influencing factor is embedded through an attention mechanism.
And the corrosion fatigue life prediction module 205 of the marine welding structure is used for predicting the life of the marine welding structure to be predicted by adopting the corrosion fatigue life prediction model of the marine welding structure.
Fatigue life influencing factors include material type, stress ratio, loading frequency, load type, surface state and temperature.
The fatigue performance data set data enhancement module 202 specifically includes:
and the fatigue performance data set data enhancement unit is used for determining and acquiring the type unbalanced data area of each fatigue life influence factor in the fatigue performance data set by using an edge oversampling method, and interpolating the type unbalanced data area of each fatigue life influence factor to obtain the fatigue performance data set after data enhancement.
The deep convolutional neural network model training module 204 specifically includes:
the data set dividing unit is used for dividing the fatigue performance data set after data enhancement into a training set and a test set;
and the deep convolutional neural network model training unit is used for adopting a training set, taking a plurality of fatigue life influencing factors as input, taking the fatigue life of the marine welding structure as output to train the deep convolutional neural network model, and testing and optimizing the trained deep convolutional neural network model by adopting a test set until the error of the predicted value output by the deep convolutional neural network model reaches a preset condition to obtain a corrosion fatigue life prediction model of the marine welding structure.
The invention also discloses corrosion fatigue life prediction equipment of the marine welding structure, which comprises at least one processor, at least one memory and computer program instructions stored in the memory, wherein when the computer program instructions are executed by the processor, the corrosion fatigue life prediction method of the marine welding structure is realized.
In the present specification, the embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principle and the embodiment of the present invention are explained by applying specific examples, and the above description of the embodiments is only used to help understanding the method and the core idea of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the foregoing, the description is not to be taken in a limiting sense.
Claims (7)
1. A corrosion fatigue life prediction method for an ocean welded structure is characterized by comprising the following steps:
constructing a fatigue performance data set of the marine welding structure, wherein each sample data in the fatigue performance data set comprises a plurality of fatigue life influencing factors of the marine welding structure and the fatigue life of the marine welding structure;
traversing each fatigue life influencing factor, acquiring a category unbalanced data area of each fatigue life influencing factor in the fatigue performance data set, and performing data enhancement on the category unbalanced data area of each fatigue life influencing factor to obtain a fatigue performance data set after data enhancement;
determining the weight of each fatigue life influence factor by adopting an extreme gradient lifting algorithm based on the fatigue performance data set after data enhancement;
training a deep convolution neural network model based on the fatigue performance data set after the data enhancement to obtain a corrosion fatigue life prediction model of the marine welding structure; the deep convolutional neural network model is a deep convolutional neural network model which is embedded with the weight of each fatigue life influence factor through an attention mechanism;
predicting the service life of the marine welding structure to be predicted by adopting the corrosion fatigue life prediction model of the marine welding structure;
the training of the deep convolutional neural network model based on the fatigue performance data set after data enhancement to obtain the corrosion fatigue life prediction model of the marine welding structure specifically comprises the following steps:
dividing the fatigue performance data set after the data enhancement into a training set and a test set;
and training the deep convolutional neural network model by adopting the training set and taking a plurality of fatigue life influencing factors as input and the fatigue life of the marine welding structure as output, and testing and optimizing the trained deep convolutional neural network model by adopting the testing set until the error of the predicted value of the output of the deep convolutional neural network model reaches a preset condition to obtain a corrosion fatigue life prediction model of the marine welding structure.
2. The method of claim 1, wherein the fatigue life influencing factors comprise material type, stress ratio, loading frequency, load type, surface condition, and temperature.
3. The method for predicting corrosion fatigue life of a marine welded structure according to claim 1, wherein the traversing each fatigue life affecting factor, obtaining a category imbalance data region of each fatigue life affecting factor in the fatigue performance data set, and performing data enhancement on the category imbalance data region of each fatigue life affecting factor to obtain a data-enhanced fatigue performance data set specifically comprises:
and determining a category unbalanced data area of each fatigue life influencing factor in the obtained fatigue performance data set by using an edge oversampling method, and interpolating the category unbalanced data area of each fatigue life influencing factor to obtain the fatigue performance data set with enhanced data.
4. A corrosion fatigue life prediction system for a marine welded structure, comprising:
the system comprises a fatigue performance data set construction module, a data acquisition module and a data processing module, wherein the fatigue performance data set construction module is used for constructing a fatigue performance data set of the marine welding structure, and each sample data in the fatigue performance data set comprises a plurality of fatigue life influencing factors of the marine welding structure and the fatigue life of the marine welding structure;
the fatigue performance data set data enhancement module is used for traversing each fatigue life influence factor, acquiring a category unbalanced data area of each fatigue life influence factor in the fatigue performance data set, and performing data enhancement on the category unbalanced data area of each fatigue life influence factor to obtain a fatigue performance data set after data enhancement;
the weight determining module of each fatigue life influencing factor is used for determining the weight of each fatigue life influencing factor by adopting an extreme gradient lifting algorithm based on the fatigue performance data set after the data enhancement;
the deep convolutional neural network model training module is used for training a deep convolutional neural network model based on the fatigue performance data set after the data enhancement to obtain a corrosion fatigue life prediction model of the marine welding structure; the deep convolutional neural network model is a deep convolutional neural network model which is embedded with the weight of each fatigue life influence factor through an attention mechanism;
the corrosion fatigue life prediction module of the marine welding structure is used for predicting the life of the marine welding structure to be predicted by adopting the corrosion fatigue life prediction model of the marine welding structure;
the deep convolutional neural network model training module specifically comprises:
the data set dividing unit is used for dividing the fatigue performance data set after the data enhancement into a training set and a test set;
and the deep convolutional neural network model training unit is used for training the deep convolutional neural network model by using the training set and taking a plurality of fatigue life influence factors as input and the fatigue life of the marine welding structure as output, and testing and optimizing the trained deep convolutional neural network model by using the test set until the error of the predicted value of the output of the deep convolutional neural network model reaches a preset condition to obtain a corrosion fatigue life prediction model of the marine welding structure.
5. The marine welded structure corrosion fatigue life prediction system of claim 4, wherein the fatigue life affecting factors comprise material type, stress ratio, loading frequency, load type, surface condition, and temperature.
6. The corrosion fatigue life prediction system of a marine welded structure of claim 4, wherein the fatigue performance data set data enhancement module specifically comprises:
and the fatigue performance data set data enhancement unit is used for determining and acquiring the type unbalanced data area of each fatigue life influence factor in the fatigue performance data set by using an edge oversampling method, and interpolating the type unbalanced data area of each fatigue life influence factor to obtain the fatigue performance data set after data enhancement.
7. An apparatus for corrosion fatigue life prediction of a marine welded structure, comprising at least one processor, at least one memory, and computer program instructions stored in the memory, which when executed by the processor, implement the method of any one of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210675768.3A CN115081321B (en) | 2022-06-15 | 2022-06-15 | Corrosion fatigue life prediction method, system and equipment for marine welding structure |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210675768.3A CN115081321B (en) | 2022-06-15 | 2022-06-15 | Corrosion fatigue life prediction method, system and equipment for marine welding structure |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115081321A CN115081321A (en) | 2022-09-20 |
| CN115081321B true CN115081321B (en) | 2023-04-07 |
Family
ID=83254232
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210675768.3A Active CN115081321B (en) | 2022-06-15 | 2022-06-15 | Corrosion fatigue life prediction method, system and equipment for marine welding structure |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115081321B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115983135B (en) * | 2023-01-28 | 2023-12-26 | 扬州浩海蓝生海洋装备有限公司 | Method and system for evaluating service life of deep sea Christmas tree oil pipe hanger |
| CN116738731A (en) * | 2023-06-16 | 2023-09-12 | 天津大学 | Universal prediction method and system suitable for fatigue performance analysis of various materials |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106442291A (en) * | 2016-09-30 | 2017-02-22 | 中国石油大学(华东) | Corrosion fatigue life prediction method based on BP neural network and application |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP4291480B2 (en) * | 1999-12-03 | 2009-07-08 | 新日本製鐵株式会社 | Structural steel with excellent corrosion resistance and corrosion fatigue resistance |
| CN111581830B (en) * | 2020-05-11 | 2021-03-30 | 湖北特种设备检验检测研究院 | Metal structure fatigue life calculation method |
| CN113281151A (en) * | 2021-05-13 | 2021-08-20 | 中国海洋石油集团有限公司 | Fatigue crack propagation strain characterization correlation method for welded structural part |
| CN216743221U (en) * | 2021-12-23 | 2022-06-14 | 江苏三耐特钢有限公司 | Seawater corrosion resistant welded steel pipe for low-pressure fluid transportation |
| CN114595625A (en) * | 2022-01-25 | 2022-06-07 | 天津大学 | Welding structure fatigue performance analysis method based on data driving method |
-
2022
- 2022-06-15 CN CN202210675768.3A patent/CN115081321B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106442291A (en) * | 2016-09-30 | 2017-02-22 | 中国石油大学(华东) | Corrosion fatigue life prediction method based on BP neural network and application |
Non-Patent Citations (1)
| Title |
|---|
| 基于人工神经网络的混凝土疲劳寿命估算;吕培印等;《海洋工程》(第03期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115081321A (en) | 2022-09-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115081321B (en) | Corrosion fatigue life prediction method, system and equipment for marine welding structure | |
| Ma et al. | Discriminative deep belief networks with ant colony optimization for health status assessment of machine | |
| CN109164343B (en) | Transformer fault diagnosis method based on characteristic information quantization and weighted KNN | |
| CN111062508B (en) | Method for evaluating real-time running state of wind turbine generator based on big data technology | |
| CN109035242B (en) | HF-ERW welding state detection method and device based on deep learning | |
| CN105843743A (en) | Method for verifying correctness of actual output result of special automatic test case | |
| CN110147323A (en) | A kind of change intelligence inspection method and device based on generation confrontation network | |
| Chen et al. | Intelligent model-based integrity assessment of nonstationary mechanical system | |
| CN112116198A (en) | Data-driven process industrial state perception network key node screening method | |
| Nguyen et al. | New methodology for improving the inspection policies for degradation model selection according to prognostic measures | |
| CN116049719A (en) | Fault detection and estimation method, system and medium based on migration type generation model | |
| CN114397577A (en) | New energy automobile lithium battery health state assessment method based on ASTUKF-GRA-LSTM model | |
| Guachamin-Acero et al. | Prediction of dynamic responses for execution of marine operations using partitioning of multimodal directional wave spectra and machine learning regression models | |
| CN110244690B (en) | Multivariable industrial process fault identification method and system | |
| CN111612036A (en) | Oil-immersed transformer fault diagnosis method based on particle swarm optimization XGboost | |
| Aslam et al. | Decision procedure for the weibull distribution based on run lengths of conforming items | |
| CN116644358A (en) | Power system transient stability evaluation method based on Bayesian convolutional neural network | |
| CN115272000A (en) | Transient stability evaluation method and system for power system | |
| CN115292820A (en) | Method for predicting residual service life of urban rail train bearing | |
| Ribeiro Machado da Silva et al. | Convolutional Neural Networks Applied to Flexible Pipes for Fatigue Calculations | |
| Yeter et al. | Structural Health Monitoring Data Analysis for Ageing Fixed Offshore Wind Turbine Structures | |
| CN114088027A (en) | Transformer winding deformation identification method based on LSTM neural network | |
| CN113609456A (en) | Reliability analysis method, system, medium and terminal for photovoltaic charging station of electric vehicle | |
| Lakra et al. | Development of Efficient and Optimal Models for Software Maintainability Prediction using Feature Selection Techniques | |
| Qidwai | Autonomous corrosion detection in gas pipelines: a hybrid-fuzzy classifier approach using ultrasonic nondestructive evaluation protocols |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |