CN119670499A - A data- and physics-driven method for predicting low-cycle fatigue life of mechanical parts - Google Patents
A data- and physics-driven method for predicting low-cycle fatigue life of mechanical parts Download PDFInfo
- Publication number
- CN119670499A CN119670499A CN202411805294.5A CN202411805294A CN119670499A CN 119670499 A CN119670499 A CN 119670499A CN 202411805294 A CN202411805294 A CN 202411805294A CN 119670499 A CN119670499 A CN 119670499A
- Authority
- CN
- China
- Prior art keywords
- stress
- physical
- data
- life
- strain
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Investigating Strength Of Materials By Application Of Mechanical Stress (AREA)
Abstract
The invention relates to a method for predicting low cycle fatigue life of a mechanical part by data and physical dual driving, which comprises the steps of S1, S2, taking stress and strain as input, designing and developing a low cycle fatigue test of a mechanical part material test piece, and obtaining test stress, strain and low cycle fatigue cycle test data, S3, obtaining physical knowledge of low cycle fatigue failure, carrying out material low cycle fatigue physical knowledge analysis by combining stock data, and obtaining physical knowledge of fatigue life and stress, S4, establishing a physical information neural network model, inputting the test data and the physical knowledge, and S5, carrying out low cycle fatigue life prediction of multiple dangerous parts of the mechanical part. The lowest life of the dangerous part is taken as the life of the mechanical part. The invention can obviously improve the life prediction precision.
Description
Technical Field
The invention belongs to the technical field of aerospace engines, and particularly relates to a data and physical dual-drive low-cycle fatigue life prediction method for mechanical parts.
Background
Fatigue failure is a critical issue in engineering applications that affects the long-term reliability of materials and structures. Traditional fatigue life prediction methods rely on empirical and simplified physical models, which often do not accurately describe complex failure mechanisms and material behavior under a variety of stress conditions. With the development of machine learning technology, neural networks have been widely used for fatigue failure analysis. These methods can theoretically cope with more complex situations than the conventional methods by learning a large number of data samples to directly predict the result. However, there are also significant disadvantages to purely data driven neural networks. First, they are extremely dependent on large amounts of high quality data. In the case of data scarcity or low data quality, the accuracy and reliability of the model can be greatly reduced. Second, the relationship between input and output is only learned from data, possibly violating some well-known physical laws or knowledge. Furthermore, for scenes outside the training data range, existing data may be overfitted, producing unreasonable results. Therefore, it is necessary to combine machine learning technology and physical law to develop fatigue failure analysis, and establish a data and physical dual-drive method for predicting the low cycle fatigue life of the mechanical part, so as to support the prediction of the low cycle fatigue life of the mechanical part.
A machine learning-based fatigue life prediction method is established at present, namely ① existing literature "A deep learning approach for low-cycle fatigue life prediction under thermal–mechanical loading based on a novel neural network model",Engineering Fracture Mechanics,, wherein a neural network is trained by using low-cycle fatigue data of four different materials, and the low-cycle fatigue life corresponding to each material is predicted. However, research shows that the prediction accuracy of the method mainly depends on the quality and quantity of training data, and the quality and usability of the data are excessively dependent. ② The prior literature is a fatigue life prediction method under data driving based on parameter influence, and mechanical engineering report, wherein the established fatigue life prediction method based on BP neural network effectively predicts the fatigue life of single-axis loading, multi-axis loading, high cycle fatigue and low cycle fatigue life, but when the distribution quality of training data is lower, the prediction precision is also reduced. In the aspect of the invention, the low cycle fatigue life prediction method (CN 115859464A) of the titanium alloy structural member is provided by the low cycle fatigue life prediction method of the titanium alloy structural member based on the GA-BP neural network, the low cycle fatigue life of the titanium alloy structural member can be predicted, but the method is completely driven by data, physical boundary conditions are not considered, the data distribution has great influence on the prediction accuracy, the prediction result sometimes has the phenomenon of violating the physical rule, the efficient and accurate prediction cannot be performed, and engineering application is difficult to realize. In summary, the existing method cannot meet the requirements of engineering high-efficiency and high-precision low-cycle fatigue life prediction.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a data and physical dual-drive method for predicting the low cycle fatigue life of a mechanical part, which supports the low cycle fatigue life assessment of the mechanical part of an aeroengine. In order to achieve the purpose, the invention adopts the following technical scheme that the method for predicting the low cycle fatigue life of the mechanical part driven by data and physics comprises the following steps:
step S1, determining stress-strain distribution of a dangerous part, which comprises the steps of carrying out finite element analysis of a mechanical part to obtain stress-strain distribution, and determining maximum stress and strain values of the dangerous part;
s2, acquiring data, namely, taking stress and strain as input, designing and developing a low cycle fatigue test of a mechanical part material test piece, and acquiring test stress, strain and low cycle fatigue cycle test data;
s3, acquiring physical knowledge of low cycle fatigue failure, wherein the physical knowledge comprises the steps of carrying out material low cycle fatigue physical knowledge analysis by combining stock data to acquire physical knowledge of fatigue life and stress;
S4, establishing a physical information neural network model, and inputting test data and physical knowledge;
And S5, predicting the low cycle fatigue life of multiple dangerous parts of the mechanical part, wherein the lowest life of the dangerous parts is used as the life of the mechanical part.
Compared with the prior art, the invention has the advantages that:
(1) The invention combines the advantages of data driving and physical driving, not only uses test data, but also combines physical knowledge of materials and mechanics. By introducing physical information, the method can capture more material behavior details and improve the prediction accuracy of fatigue life under different working conditions.
(2) According to the invention, complex stress-strain distribution is accurately simulated through finite element analysis, and complex stress states can be effectively treated by combining physical knowledge and a neural network model. The method can capture more load working condition characteristics and has stronger adaptability.
(3) According to the invention, the neural network model can be continuously trained and optimized by continuously updating the test data and the physical model, so that the neural network model is suitable for new materials and new working conditions at any time, and the prediction accuracy and the frontier are kept. This allows the model to dynamically adapt to new engineering requirements.
Drawings
FIG. 1 is a flow chart of a method for predicting low cycle fatigue life of a mechanical part driven by both data and physics in the invention;
FIG. 2 is a physical neural network architecture diagram;
Fig. 3 is a flow chart of physical neural network model training.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other. In order to achieve the above purpose, the present invention adopts the following technical scheme.
The invention provides a data and physical dual-drive low cycle fatigue life prediction method for mechanical parts, which supports low cycle fatigue life assessment of mechanical parts of an aeroengine. In order to achieve the above purpose, the invention adopts the following technical scheme:
Firstly, determining stress and strain distribution of dangerous parts through finite element analysis, and obtaining stress, strain and low cycle fatigue cycle data through carrying out a low cycle fatigue test of a mechanical part material test piece, then carrying out physical knowledge analysis of material low cycle fatigue by combining stock data to obtain physical knowledge of fatigue life and stress, and finally, predicting the low cycle fatigue life of the mechanical part by establishing a physical information neural network model, wherein the method comprises the following steps:
Step S1, determining stress and strain distribution of a dangerous part, carrying out finite element analysis of a mechanical part to obtain stress and strain distribution, and determining maximum stress and strain of the dangerous part and the value of the maximum stress and strain;
S2, acquiring data, namely designing and developing a low cycle fatigue test of the mechanical part material test piece by taking the stress and the strain in the step S1 as input, and acquiring stress, strain and low cycle fatigue cycle test data;
S3, acquiring physical knowledge, carrying out material low-cycle fatigue physical knowledge analysis by combining stock data, and acquiring physical knowledge of fatigue life and stress;
step S4, establishing a physical information neural network model, wherein test data and physical knowledge are respectively determined through the step S2 and the step S3;
and S5, predicting the low cycle fatigue life of the mechanical part, wherein the lowest life of the dangerous part is used as the life of the mechanical part.
Further, the specific steps of the step S1 are as follows:
Step S11, firstly, establishing a three-dimensional model according to a two-dimensional design drawing of a mechanical part;
and step S12, setting material properties including elastic modulus, poisson ratio, yield strength and the like in finite element analysis software. If the nonlinear behaviour of the material is significant in low cycle fatigue, plastic models, such as bilinear or multi-linear hardening models, are also considered.
And S13, carrying out grid division on the parts. To ensure computational accuracy, a denser grid is employed, particularly in areas of stress concentration (e.g., holes, grooves, abrupt geometry, etc.).
And S14, setting boundary conditions according to the actual load condition of the mechanical part, establishing a Finite Element Model (FEM), and applying static or dynamic load to the finite element model according to the actual working condition. This may include axial loads, torsional loads, bending loads, or combined loads. The loading process may simulate a single large load or multiple cyclic loads to observe the stress-strain response of the mechanical part. If the part is subjected to multiaxial loads (e.g., a stretch-twist combination) in actual use, multiaxial stress analysis is required to more accurately reflect the stress state.
And S15, running finite element simulation to obtain stress strain distribution. The results include key parameters such as von Mises stress, maximum principal stress, shear stress, plastic strain, etc. And (3) carrying out visual analysis on the results through a post-processing tool, generating a stress cloud chart, a deformation cloud chart and the like, and determining the concentration areas of stress and strain in the part. These areas are potential fatigue hazard sites. And extracting stress and strain values of the dangerous parts to be used as input data for subsequent experiments and model analysis. If the hazard site involves multiple regions, the stress-strain characteristics of each region should be recorded separately.
And comparing the finite element simulation result with actual working conditions or known test data. If there is a deviation, it may be necessary to redefine the boundary conditions, optimize the meshing, or adjust the material model. The simulation result is ensured to be in a reasonable range, and particularly the stress-strain distribution of dangerous parts accords with a physical rule, so that excessive simplification or model distortion is avoided.
And the stress concentration area and dangerous parts of the mechanical part can be identified through finite element analysis results, so that a basis is provided for the subsequent fatigue test scheme design and fatigue life prediction.
Further, the specific steps of the step S2 are as follows:
step S21, firstly determining test standards and specifications according to which tests are required to be carried out, thereby preparing a test scheme.
And S22, determining test conditions according to the dangerous part stress and strain values determined in the step S15. The test design should simulate the load and environment under actual working conditions as much as possible. The appropriate material sample is selected and the sample shape may be a standard tensile sample, a flat plate sample or a specially machined feature simulator sample. The size of the sample meets the requirement of the testing machine and ensures even stress distribution. The loading mode and the loading level are determined. Low cycle fatigue tests typically apply cyclic loads, varying from tens to thousands of cycles, depending on material properties and design requirements.
Step S23, applying a cyclic load to the sample on the testing machine. The testing machine should be equipped with high precision strain gages, strain gages or displacement sensors to monitor the strain response of the test specimen in real time. In the test process, the stress and the strain value of each loading cycle and the corresponding cycle times are recorded until the sample is broken or obviously damaged. The raw data obtained in the test were collected, including stress-strain curves, number of cycles, etc. for each cycle. The data is processed and consolidated to generate stress-strain curves, fatigue life curves (e.g., S-N curves), and other key data graphs. Data smoothing is performed as necessary to eliminate noise and errors in the test. If the test samples are more, the data under different conditions can be subjected to statistical analysis, and the distribution characteristics of the fatigue life, such as average life, standard deviation, confidence interval and the like, can be extracted.
Further, the specific steps of the step S3 are as follows:
Step S31, firstly, carrying out stock data analysis, and collecting historical fatigue data and literature data related to current materials and working conditions, wherein the historical fatigue data and literature data comprise published low-cycle fatigue test results, fatigue curves, material constitutive models and the like. Existing data are analyzed to extract fatigue behavior under stress and strain ranges similar to those of current research. For example, the fatigue limit, fatigue crack growth law, etc. of the same kind of material are analyzed.
Step S32, combining the test data and the stock data, and analyzing the fatigue behavior of the material by using classical fatigue theory (such as Coffin-Manson physical model and Basquin physical model). The material constants and fatigue parameters in these physical models were determined by fitting the experimental data. If the material exhibits significant nonlinear behavior, it may be desirable to use a nonlinear fatigue model or introduce fracture mechanics theory to more accurately describe the fatigue process. The fatigue life at different stress levels is predicted using a physical model, forming a stress-life (S-N) curve, a strain-life (epsilon-N) curve, etc.
And step S33, combining the test data with the result of the physical model to form a comprehensive low-cycle fatigue life knowledge base. The knowledge base should contain fatigue life information under different stress levels, strain amplitudes, loading frequencies and other conditions, as well as details of crack growth rates, fatigue fracture mechanisms and the like. And analyzing the relation among stress, strain, cycle number and fatigue life, extracting a key physical rule and an empirical formula, and providing physical constraint conditions for the subsequent neural network model construction.
Further, the specific steps of the step S4 are as follows:
And S41, designing a proper neural network model architecture according to the data characteristics and the prediction task. And (3) parameter adjustment is carried out on the model structure, such as network layer number, neuron number, activation function, learning rate and the like, so that the model is ensured to have enough expression capacity to capture complex fatigue behaviors.
Step S42, introducing physical constraint or priori knowledge in the design of the neural network model. And taking the physical knowledge as an input characteristic of the neural network model, or introducing physical consistency constraint through regularization terms to ensure that the output of the neural network model accords with a physical rule.
In order to combine the physical knowledge, the training of the neural network is expressed as a constraint optimization problem in step S43.
And S44, customizing a loss function, and converting the constraint optimization problem into an unconstrained optimization problem.
And step S45, preprocessing the test data obtained in the step S2. Including data cleansing (removing outliers and noise), normalization (scaling the data to the same range) to ensure that the data is easier to learn in the neural network model. According to physical laws and fatigue theory, characteristic parameters such as maximum stress, strain amplitude, plastic strain range and the like which have influence on fatigue life prediction are extracted from data such as stress, strain, cycle times and the like. The model was trained using the pre-processed test data. During training, the model continuously adjusts internal parameters (such as weights and biases) through a back propagation algorithm to minimize the prediction error. By adopting the cross verification method, the neural network model is ensured to perform well on the training set and the verification set, and overfitting is avoided. In the training process, indexes such as a loss function, accuracy and the like are monitored in real time, and the convergence condition of the model is judged. If the model performs poorly, the model may be improved by adding training data, introducing data enhancement techniques (e.g., adding noise, smoothing) or modifying the network architecture.
And analyzing the prediction result of the model, calculating evaluation indexes such as prediction error, correlation coefficient, mean Square Error (MSE) and the like, and judging the generalization capability of the model. If the model is subject to large errors under certain specific conditions, it may be necessary to further optimize the model or adjust the data set.
Further, the specific steps of the step S5 are as follows:
and S51, inputting the test data of the dangerous part determined in the step S1 into the neural network model trained in the step S4, and carrying out life prediction. For different stress levels and strain amplitudes, the model will output corresponding fatigue life predictions. In general, the overall life of a part is considered to be the lowest life of its most dangerous location (location of greatest stress strain).
And S52, comparing the prediction result of the neural network model with the service life of the mechanical part in actual use, and analyzing the prediction precision and reliability of the neural network model. If there is a significant deviation of the actual life from the predicted life, it may be necessary to retrain the model or adjust the parameters. The model is gradually optimized through multiple applications and feedback, so that the prediction result is more accurate and stable.
Finally, a systematic fatigue life prediction flow is formed, and a reliable technical support is provided for the design and maintenance of mechanical parts by combining data and a physical dual-drive model.
The technical scheme of the invention is further described below with reference to the accompanying drawings.
As shown in FIG. 1, the method for predicting the low cycle fatigue life of the mechanical part by data and physical double driving comprises the following steps:
and S1, carrying out finite element analysis on the mechanical parts to obtain stress and strain distribution, and determining the maximum stress and strain parts and the values thereof.
Taking a certain turbine disc as an example, carrying out finite element analysis based on UG and ABAQUS software, carrying out three-dimensional modeling on the turbine disc in UG according to a two-dimensional drawing, and then importing the three-dimensional model into ABAQUS for further finite element analysis.
Step S12, setting the turbine disk material as GH720Li, setting the material properties as multi-section linear follow-up reinforcement in ABAQUS, and setting other material properties including elastic modulus, poisson' S ratio, yield strength and the like.
Step S13, meshing the turbine disk, and adopting a dense grid in particular in a stress concentration area (such as holes, mortises, geometric shapes with abrupt change and the like) in order to ensure calculation accuracy.
In the finite element analysis, the boundary conditions of the turbine disc should be as close as possible to the load state in the service environment, and the boundary conditions are set as follows:
1) The temperature field distribution of the turbine disk is applied according to a function obtained by fitting a heat transfer calculation result;
2) Applying a corresponding rotational angular velocity according to the rotational speed of the turbine disc;
3) Centrifugal force generated by the turbine disk blades is applied to the units of the upper half part of the mortise in the form of pressure;
4) The axial and circumferential displacements of the turbine disc are constrained by applying UZ, uy=0 at the mounting edge.
And S15, calculating through ABAQUS to obtain finite element analysis stress distribution. When the stress analysis is carried out on the turbine disk, not only the stress strain condition of dangerous parts is required to be concerned, but also the stress strain of stress concentration areas (such as holes, mortises, geometric shapes with abrupt changes and the like) are required to be extracted for analysis.
And S2, acquiring data. And carrying out a low cycle fatigue test of the mechanical part material test piece to obtain stress, strain and low cycle fatigue cycle data. The method comprises the following specific steps:
the low cycle fatigue test is carried out, so that the stress strain data and the low cycle fatigue life of the mechanical part material are obtained, and data support is provided for the prediction of the low cycle fatigue life of the mechanical part.
And S21, the test is carried out by referring to the standards of GB/T15248-2008, GB/T26077-2010, HB 5195-1996, metal high temperature tensile test method and the like, and the contents of a clamping mode, a loading mode and the like are adjusted according to the actual condition of a test piece.
And S22, taking a sample from GH720Li blank discs in the same batch, designing the sample according to GJB, and adopting an isothermal forging mode, wherein the heat treatment process is (1080-1110) DEGC X (2-4) h/oil quenching +650 ℃ X24 h/air cooling +760 ℃ X16 h/air cooling. Based on the result of the finite element analysis in step S1, the test conditions are determined as follows:
The turbine disk samples in the same batch are sampled at the temperature of 650 ℃ and the stress ratio of 0.05, the loads are respectively 800MPa, 900MPa, 1000MPa, 1100MPa and 1200MPa, and the number of test pieces is 5 under each load condition;
And S23, all samples are carried out on the same MTS 370.10 hydraulic servo fatigue tester, the control condition is stress control, the load form is triangular wave, and the loading frequency is 5Hz. Strain data are measured in the whole course by adopting an extensometer in the test process, and the cycle time and the test time are recorded after the test is finished.
Aiming at the test data obtained in the step S2, the method can be used for training a physical information neural network and can be used for predicting the low cycle fatigue life of subsequent mechanical parts.
And S3, acquiring physical knowledge. And carrying out physical knowledge analysis of low-cycle fatigue of the material by combining the stock data to obtain physical knowledge of fatigue life and stress. The method comprises the following specific steps:
Step S31, firstly, carrying out stock data analysis, and collecting historical fatigue data and literature data related to current materials and working conditions, wherein the historical fatigue data and literature data comprise published low-cycle fatigue test results, fatigue curves, material constitutive models and the like. The existing data are analyzed to extract fatigue behavior under stress and strain ranges similar to those of the mechanical parts in the invention.
And S32, analyzing fatigue behavior of the material by combining the test data and the stock data. The material constants and fatigue parameters in these models were determined by fitting the experimental data. The fatigue life at different stress levels is predicted using a physical model, forming a stress-life (S-N) curve, a strain-life (epsilon-N) curve, etc.
And step S33, combining the test data with the result of the physical model to form a comprehensive low-cycle fatigue life knowledge base. And analyzing the relation among stress, strain, cycle times and fatigue life, and extracting a key physical rule and an empirical formula. The method mainly comprises the steps of increasing the variance of the fatigue life along with the reduction of the stress amplitude, enabling the fatigue life to have a monotonically decreasing trend along with the increase of the stress amplitude, and enabling the curvature of the S-N curve to decrease along with the reduction of the stress when the stress amplitude is reduced to the fatigue limit.
And S4, building a physical information neural network model, wherein test data and physical knowledge are respectively determined through the step S2 and the step S3. The method comprises the following specific steps:
Step S41. The physical neural network model architecture used in the present invention is shown in FIG. 2, and the network comprises three layers, namely an input layer, a hidden layer and an output layer. The choice of the number of hidden layers is determined by the complexity of the problem, and a network structure with one hidden layer is used in the present invention. Unlike conventional neural networks that learn the average of the outputs only from the collected distributed data, the output layer has two output neurons, average and standard deviation. The loss function uses a negative log-likelihood function, namely:
(1)
wherein n is the number of training data, x and y are input and output variables, respectively, θ is the set of neural network parameters, Representing the conditional probability density function of the output variable y given the input variable x and the neural network parameter θ,Representing the cumulative distribution function of the output variable y given the input variable x and the neural network parameter theta,Is a failure indicator, expressed as:
(2)
For the input layer, the neural network definition does not use an activation function. The neurons of the hidden layer are subjected to hyperbolic tangent activation function (tanh), and the expression of the tanh activation function is:
(3)
where z is the output of the corresponding neuron.
The mean value of the output layer uses a linear (i.e., identity) activation function, the expression of which is:
(4)
since the standard deviation is non-negative, an exponential linear unit (elu) activation function is selected, the expression for the elu activation function is:
(5)
in step S42, the input of the network is stress amplitude x, and the output is average mu and standard deviation sigma of predicted fatigue life y. In order to ensure that the physical index is consistent with the underlying physical knowledge, the physical knowledge determined in step S3 is expressed as a physical constraint as follows by calculating the derivative of the physical index with respect to the stress amplitude and imposing a condition to be satisfied, i.e., the physical constraint:
(1) Stress amplitude is reduced, variance of fatigue life is increased, and the first derivative of standard deviation is negative:
(6)
(2) The fatigue life monotonically decreases with increasing stress amplitude, with the mean first derivative being negative:
(7)
(3) The curvature of the stress life curve decreases with decreasing stress amplitude, the mean second derivative being a positive number:
(8)
In order to combine the physical knowledge, the training of the neural network is expressed as a constraint optimization problem in step S43.
(9)
Note that the above three constraints need to be satisfied simultaneously.
Step S44, solving the optimization problem in the formula (9) by adopting a penalty function method, and converting the constrained optimization problem into an unconstrained optimization problem, thereby defining a composite loss function as:
(10)
Where L 0 (θ) is a negative log-likelihood function in equation (1), L 1 (θ) is a loss function corresponding to the first physical constraint, and the expression is:
(11)
L 2 (θ) is a loss function corresponding to the second physical constraint, and the expression is:
(12)
L 3 (θ) is a loss function corresponding to the third physical constraint, and the expression is:
(13)
Penalty factor ,,To balance between data fitting and physical constraints, physical constraints are enforced, only as close to zero as possible. Typically, penalty factors,,Too small a value to guarantee physical constraints, penalty factors,,Too large a value may in turn lead to an overfitting. Thus, by selecting an appropriate penalty factor,,Better fitting results can be achieved for the values.
In summary, as shown in fig. 3, the training flowchart of the physical neural network in step S45 is as follows:
step S451, dividing the original data into a training set, a testing set and a verification set, and starting training the neural network. Penalty factor ,,Initially 1. When the data fitting loss converges, neural network training ceases. The physical constraint loss at the end of step S451 may not be zero.
Step S452, judging whether the physical constraint is satisfied, and terminating the whole process when the physical constraint is satisfied. If not, a penalty factor is increased to incrementally adjust the neural network to meet the physical constraint while maintaining the data fit loss unchanged significantly. The learning rate is reduced, and the weight and deviation of the updated neural network are prevented from being too different from the result of the previous training. The initial weights and deviations are reassigned to the values.
And step S453, the neural network resumes training after parameter updating, and physical constraint loss is monitored. When the physical constraint loss converges, neural network training ceases. And then goes to step S452 to continue the calculation.
And S5, carrying out low cycle fatigue life prediction of the mechanical parts. The lowest life of the dangerous part is taken as the life of the mechanical part. The method comprises the following specific steps:
And S51, taking the test data obtained in the step S2 as input, introducing the test data into the physical neural network model trained in the step S4 to predict the low cycle fatigue life, and taking the lowest life of all dangerous parts as the life of the mechanical parts.
And S52, comparing the prediction result of the model with the service life of the mechanical part in actual use, and analyzing the prediction precision and reliability of the neural network model. If there is a significant deviation of the actual life from the predicted life, it may be necessary to retrain the neural network model or adjust the parameters. The model is gradually optimized through multiple applications and feedback, so that the prediction result is more accurate and stable.
It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (7)
1. A method for predicting the low cycle fatigue life of a mechanical part driven by data and physics is characterized by comprising the following steps:
step S1, determining stress-strain distribution of a dangerous part, which comprises the steps of carrying out finite element analysis of a mechanical part to obtain stress-strain distribution, and determining maximum stress and strain values of the dangerous part;
s2, acquiring data, namely, taking stress and strain as input, designing and developing a low cycle fatigue test of a mechanical part material test piece, and acquiring test stress, strain and low cycle fatigue cycle test data;
s3, acquiring physical knowledge of low cycle fatigue failure, wherein the physical knowledge comprises the steps of carrying out material low cycle fatigue physical knowledge analysis by combining stock data to acquire physical knowledge of fatigue life and stress;
S4, establishing a physical information neural network model, and inputting test data and physical knowledge;
And S5, predicting the low cycle fatigue life of multiple dangerous parts of the mechanical part, wherein the lowest life of the dangerous parts is used as the life of the mechanical part.
2. The method for predicting the low cycle fatigue life of a mechanical part driven by both data and physical as set forth in claim 1, wherein the specific steps of step S1 are as follows:
s11, establishing a three-dimensional model of the mechanical part according to a two-dimensional design drawing of the mechanical part;
step S12, setting material properties of the mechanical parts in finite element analysis software, wherein the material properties comprise elastic modulus, poisson ratio and yield strength;
Step S13, carrying out grid division on the parts, wherein a denser grid is adopted in a stress concentration area, and the stress concentration area comprises holes, grooves and geometric shapes which change rapidly;
Step S14, setting boundary conditions according to the actual load condition of the mechanical part, establishing a finite element model, and applying static or dynamic load to the finite element model according to the actual working condition;
Step S15, running finite element simulation to obtain stress-strain distribution, wherein the result comprises key parameters of von Mises stress, maximum principal stress, shear stress and plastic strain, and performing visual analysis on the result through a post-processing tool to generate a stress cloud image and a deformation cloud image, determining concentration areas of stress and strain in the mechanical part, wherein the areas are potential fatigue dangerous parts, extracting stress and strain values of the dangerous parts as input data, and respectively recording stress-strain characteristics of each area if the dangerous parts relate to a plurality of areas.
3. The method for predicting the low cycle fatigue life of a mechanical part driven by both data and physical as set forth in claim 2, wherein the specific steps of step S2 are as follows:
step S21, firstly determining test standards and specifications required to be relied on by a test, thereby preparing a test scheme;
S22, determining test conditions according to the determined stress and strain values of dangerous parts, testing and simulating loads and environments under actual working conditions, selecting a proper material sample, wherein the shape of the sample is a standard tensile sample, a flat sample or a specially processed characteristic simulation piece sample, the size of the sample meets the requirements of a testing machine, the stress distribution is uniform, the loading mode and the loading level are determined, and the cyclic load is applied in a low-cycle fatigue test;
And S23, applying a cyclic load to the sample on a testing machine, wherein the testing machine is provided with a strain gauge, a strain gauge or a displacement sensor so as to monitor the strain response of the sample in real time, recording the stress and the strain value of each loading cycle and the corresponding cycle times in the test process until the sample breaks or is obviously damaged, collecting the original data obtained in the test, including the stress-strain curve and the cycle times of each cycle, processing and sorting the data to generate a stress-strain curve and a fatigue life curve, and carrying out statistical analysis on the data under different conditions to extract the distribution characteristics of the fatigue life, including the average life, the standard deviation and the confidence interval.
4. A method for predicting low cycle fatigue life of a mechanical part driven by both data and physical as set forth in claim 3, wherein the specific steps of step S3 are as follows:
Step S31, firstly, carrying out stock data analysis, collecting historical fatigue data and literature data related to current materials and working conditions, including low-cycle fatigue test results, fatigue curves and material constitutive models, analyzing the data, and extracting fatigue behaviors of mechanical parts in stress and strain ranges;
Step S32, analyzing the fatigue behavior of the material by combining the test data and the stock data, determining the material constant and the fatigue parameter in the physical model by fitting the test data, and predicting the fatigue life under different stress levels by using the physical model to form a stress-life curve, namely an S-N curve, and a strain-life curve, namely an epsilon-N curve;
Step S33, combining test data with the result of the physical model to form a low-cycle fatigue life knowledge base, analyzing the relation among stress, strain, cycle times and fatigue life, and extracting a key physical rule, wherein the method comprises the steps of increasing the variance of the fatigue life along with the decrease of the stress amplitude, enabling the fatigue life to be in a monotonically decreasing trend along with the increase of the stress amplitude, and decreasing the curvature of the S-N curve along with the decrease of the stress when the stress amplitude is reduced to the fatigue limit.
5. The method for predicting the low cycle fatigue life of a mechanical part driven by both data and physical as set forth in claim 4, wherein the specific steps of step S4 are as follows:
Step S41, setting a network architecture comprising three layers, namely an input layer, a hidden layer and an output layer, wherein the number of the hidden layers is set to be 1, the output layer is provided with two output neurons, namely an average value and a standard deviation, and a loss function uses a negative log likelihood function, namely:
wherein n is the number of training data, x and y are input and output variables, respectively, θ is the set of neural network parameters, Representing the conditional probability density function of the output variable y given the input variable x and the neural network parameter θ,Representing the cumulative distribution function of the output variable y given the input variable x and the neural network parameter theta,Is a failure indicator, expressed as:
For the input layer, the neural network is defined without using an activation function, and for the neurons of the hidden layer, a hyperbolic tangent activation function tanh is used, and the expression of the tanh activation function is as follows:
wherein z is the output of the corresponding neuron;
the average value of the output layer uses a linear activation function, and the expression of the linear activation function is:
the exponential linear unit elu is selected to activate the function, and the expression of the elu activation function is:
step S42, the input variable x of the network is stress amplitude, and the output variable y is the mean value mu and standard deviation sigma of the predicted fatigue life;
the physical knowledge determined in step S3 is expressed as a physical constraint by calculating the derivative of the physical index with respect to the stress amplitude and applying the condition to be satisfied, i.e., the physical constraint, specifically as follows:
1) Stress amplitude is reduced, variance of fatigue life is increased, and the first derivative of standard deviation is negative:
2) The fatigue life monotonically decreases with increasing stress amplitude, with the mean first derivative being negative:
3) The curvature of the stress life curve decreases with decreasing stress amplitude, the mean second derivative being a positive number:
Step S43, in order to combine physical knowledge, training of the neural network is expressed as a constraint optimization problem:
The three constraint conditions need to be met simultaneously;
Step S44, solving the optimization problem in the formula (9) by adopting a penalty function method, and converting the constrained optimization problem into an unconstrained optimization problem, thereby defining a composite loss function as:
Where L 0 (θ) is a negative log-likelihood function in equation (1), L 1 (θ) is a loss function corresponding to the first physical constraint, and the expression is:
L 2 (θ) is a loss function corresponding to the second physical constraint, and the expression is:
L 3 (θ) is a loss function corresponding to the third physical constraint, and the expression is:
Wherein, ,,Is a penalty factor;
And step S45, training a physical information neural network according to the network setting, and predicting the low cycle fatigue life of the subsequent mechanical parts.
6. The method for predicting low cycle fatigue life of a mechanical part driven by both data and physical as set forth in claim 5, wherein the specific steps of step S45 are as follows:
step S451, dividing the original data into a training set, a testing set and a verification set, starting training the neural network, and punishing factors ,,Initial 1, when the data fitting loss converges, the neural network training is stopped;
Step S452, judging whether the physical constraint is met, and ending the whole process when the physical constraint is met, if the physical constraint is not met, increasing a punishment factor to gradually adjust the neural network to meet the physical constraint, simultaneously keeping the data fitting loss not to change significantly, reducing the learning rate, and preventing the weight and deviation of the updated neural network from being excessively different from the previous training result, wherein the initial weight and deviation are reassigned to values;
and step S453, the neural network resumes training after the parameter update, monitors the physical constraint loss, stops training the neural network when the physical constraint loss converges, and then goes to step S452 to continue calculation.
7. The method for predicting the low cycle fatigue life of a mechanical part driven by both data and physical as set forth in claim 6, wherein the specific steps of step S5 are as follows:
S51, taking the test data obtained in the step S2 as input, leading the test data into the physical neural network trained in the step S4 to predict the low cycle fatigue life, and taking the lowest life of all dangerous parts as the life of mechanical parts;
And step S52, comparing the prediction result of the neural network model with the life of a mechanical part in actual use, analyzing the prediction accuracy and reliability of the neural network model, and if the actual life has obvious deviation from the predicted life, retraining or adjusting parameters of the neural network model, and gradually optimizing the neural network model through multiple applications and feedback.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411805294.5A CN119670499B (en) | 2024-12-10 | 2024-12-10 | Method for predicting low cycle fatigue life of mechanical part driven by data and physics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411805294.5A CN119670499B (en) | 2024-12-10 | 2024-12-10 | Method for predicting low cycle fatigue life of mechanical part driven by data and physics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN119670499A true CN119670499A (en) | 2025-03-21 |
| CN119670499B CN119670499B (en) | 2025-10-28 |
Family
ID=94997354
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202411805294.5A Active CN119670499B (en) | 2024-12-10 | 2024-12-10 | Method for predicting low cycle fatigue life of mechanical part driven by data and physics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN119670499B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119889548A (en) * | 2025-03-31 | 2025-04-25 | 湖南科技大学 | Rubber material fatigue life prediction method based on physical-data fusion driving |
| CN120404359A (en) * | 2025-06-26 | 2025-08-01 | 西北工业大学 | Method for obtaining failure rate of pressure compensator bladder of underwater electro-hydraulic actuator |
| CN120724863A (en) * | 2025-08-27 | 2025-09-30 | 浙江大学 | Probabilistic fatigue life prediction method for complex equipment based on physical information neural network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112100766A (en) * | 2020-08-28 | 2020-12-18 | 北京航空航天大学 | Fatigue life prediction method for cold extrusion reinforced hole structure |
| US20220026326A1 (en) * | 2019-01-11 | 2022-01-27 | East China University Of Science And Technology | A Multiaxial Creep-Fatigue Prediction Method Based On ABAQUS |
| CN119047348A (en) * | 2024-11-01 | 2024-11-29 | 北京航空航天大学 | Machine learning prediction method for complex fatigue life of material based on simple test data |
-
2024
- 2024-12-10 CN CN202411805294.5A patent/CN119670499B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220026326A1 (en) * | 2019-01-11 | 2022-01-27 | East China University Of Science And Technology | A Multiaxial Creep-Fatigue Prediction Method Based On ABAQUS |
| CN112100766A (en) * | 2020-08-28 | 2020-12-18 | 北京航空航天大学 | Fatigue life prediction method for cold extrusion reinforced hole structure |
| CN119047348A (en) * | 2024-11-01 | 2024-11-29 | 北京航空航天大学 | Machine learning prediction method for complex fatigue life of material based on simple test data |
Non-Patent Citations (2)
| Title |
|---|
| SP ZHU: "Probabilistic Low Cycle Fatigue Life Prediction Using an Energy-Based Damage Parameter and Accounting for Model Uncertainty", 《INTERNATIONAL JOURNAL OF DAMAGE MECHANICSINTERNATIONAL JOURNAL OF DAMAGE MECHANICS》, 31 December 2012 (2012-12-31) * |
| 王宇堂;张伟;胡磊;: "基于损伤力学的TB6冲击缺陷影响系数", 中国科技信息, no. 22, 19 November 2018 (2018-11-19) * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119889548A (en) * | 2025-03-31 | 2025-04-25 | 湖南科技大学 | Rubber material fatigue life prediction method based on physical-data fusion driving |
| CN120404359A (en) * | 2025-06-26 | 2025-08-01 | 西北工业大学 | Method for obtaining failure rate of pressure compensator bladder of underwater electro-hydraulic actuator |
| CN120724863A (en) * | 2025-08-27 | 2025-09-30 | 浙江大学 | Probabilistic fatigue life prediction method for complex equipment based on physical information neural network |
| CN120724863B (en) * | 2025-08-27 | 2025-11-25 | 浙江大学 | Complex equipment probability fatigue life prediction method based on physical information neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN119670499B (en) | 2025-10-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN119670499B (en) | Method for predicting low cycle fatigue life of mechanical part driven by data and physics | |
| Fei | Springback prediction for incremental sheet forming based on FEM-PSONN technology | |
| CN106096073A (en) | A kind of metal fatigue crackle life-cycle predictor method analyzing model based on the non-probability interval of damage mechanics | |
| CN117892583A (en) | Intelligent stamping control method and system | |
| Nikitin et al. | Multimode model for fatigue damage development | |
| CN117034699A (en) | A metal fatigue S-N curve prediction method based on tensile testing | |
| CN115310311A (en) | Method for calculating fatigue life of metal material under spectral load | |
| CN120106692A (en) | Finished product quality evaluation method of prefabricated beam bridge integrating fuzzy reasoning and neural network | |
| Guan-feng et al. | Constitutive model of 25CrMo4 steel based on IPSO-SVR and its application in finite element simulation | |
| Chang et al. | Damage identification and localization of pultruded FRP composites based on convolutional recurrent neural network and metaheuristic intelligent algorithms | |
| Yang et al. | A novel multi-asperity-based dynamic (MABD) model for piezoelectric actuator: Theory, numerical framework, and experimental validation | |
| CN120951574A (en) | A method for predicting the lifetime of defective materials based on M-integral physical information neural networks | |
| CN119047376B (en) | Pressure vessel safety evaluation early warning method based on digital twinning | |
| CN119247852B (en) | Vacuum glue filling control method and system based on machine learning | |
| Zhou et al. | Inversion of the fracture toughness of zirconium alloy cladding interface in nuclear fuel using splitting method via general regression neural network | |
| CN119673336A (en) | Prediction method of constitutive relationship of rheological behavior of 6061 aluminum alloy based on physical information machine learning | |
| CN114186461B (en) | Multiaxial fatigue life prediction method | |
| CN115488213B (en) | Prediction method for sheet forming performance | |
| Yu et al. | Application of machine learning modeling for predicting the reliability of solder joints under thermal cycling | |
| Chao et al. | Artificial intelligence model of complicated flow behaviors for Ti− 13Nb− 13Zr alloy and relevant applications | |
| WO2021240900A1 (en) | Steel pipe collapse strength prediction model generation method, steel pipe collapse strength prediction method, steel pipe manufacturing characteristics determination method, and steel pipe manufacturing method | |
| CN119598817B (en) | Standard component design effectiveness evaluation method | |
| Guo et al. | Neural network approaches for real-time fatigue life Estimation by surrogating the rainflow counting method | |
| CN120086955B (en) | ETFE hyperbolic negative Gaussian film digital construction method and system based on big data | |
| CN119558119B (en) | Shape optimization method and system for high-temperature alloy disc type pre-forgings |
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 |