Multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure

Abstract

The invention discloses a multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure, is applied to the field of fatigue reliability, and aims at the limitation of metal fatigue life prediction under defect-induced failure in the prior art. According to the method, uncertainty quantification of fatigue performance and fitting of low-fidelity fatigue data with physical consistency are realized through the physically guided Wasserstein generative adversarial network. The concept of transfer learning is introduced, and multi-fidelity fatigue data is allowed to be used for training the physical information neural network in the training process. The influence of manufacturing defects on fatigue performance is embedded as a physical constraint, so that the physical consistency of the prediction result of the multi-fidelity physical information neural network is ensured.

Description

Technical Field

[0001] The invention belongs to the field of fatigue reliability, and particularly relates to a metal fatigue life prediction technology under defect-induced failure. Background Art

[0002] With the rapid development of industrialization, major equipment is developing towards precision, complexity, high reliability and high safety. The fatigue performance of components and structures is usually controlled by defects that are inevitably introduced during the manufacturing process (casting, forging and additive manufacturing technology). These defects significantly reduce the fatigue strength and overall life relative to the theoretical performance of a given microstructure and are also one of the main sources of life dispersion. Additive manufacturing technology has become one of the cutting-edge technologies that promote the leapfrog development of major equipment in the aviation and aerospace fields due to its significant advantages in the forming and manufacturing of large and complex components. However, the random manufacturing defects introduced by additive manufacturing technology still make it a challenge to adopt it under fatigue load conditions. As major equipment develops towards high reliability and high safety, a comprehensive assessment of the fatigue performance degradation of key components under defect-induced failure is crucial for its engineering application.

[0003] So far, there has been a deep accumulation of research on fatigue failure behavior induced by defects such as additively manufactured metals. Researchers hope to find a method that can reasonably and reliably predict the fatigue life of metals under defect-induced failure. Thanks to the refined characteristics of manufacturing defects provided by computed tomography, studies on the effects of defects on fatigue performance have shown that the size, shape and position of defects are the main factors affecting fatigue performance. This means that defect-sensitive fatigue modeling is crucial for quantifying the uncertainty of fatigue performance under defect-induced failure. However, in reality, there are a large number of manufacturing defects with different shapes, uneven sizes, and irregular distribution in parts and components, and there are complex interactions between them. It is difficult for existing fracture mechanics models to correctly evaluate the impact of defects on fatigue performance. It is urgent to establish a high-precision fatigue life assessment method to ensure the service safety of key components.

[0004] Thanks to the rapid development of data-driven methods, machine learning algorithms with powerful data analysis and fitting capabilities provide a new perspective for exploring defect-induced fatigue failure behavior. However, the data-driven nature of machine learning models determines that their prediction and generalization performance depends on the size and quality of the constructed training set. Considering that fatigue testing is economically expensive and time-consuming, it is difficult to provide a large enough data set to cover the entire observation range in engineering practice. For unknown scenarios outside the prediction training data set, purely data-driven machine learning models are difficult to provide reliable and accurate fatigue life predictions. More importantly, it is difficult for purely data-driven machine learning models to correctly understand fatigue failure behavior from a small amount of experimental data, which may result in their fatigue life predictions often not being in line with the laws of physics. Summary of the invention

[0005] In order to solve the above technical problems, the present invention proposes a multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure, which has significant advantages in improving the prediction performance, generalization ability and effectiveness of the fatigue life prediction model.

[0006] The technical solution adopted by the present invention is: a multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure, comprising:

[0007] S1. Obtain the test conditions, critical defect geometric characteristics, static mechanical properties and corresponding fatigue life test results of the same grade series of metals to form an original fatigue database;

[0008] S2. Expand and preprocess the data in the fatigue database using physical information feature engineering to obtain an expanded fatigue database, and divide the data in the expanded fatigue database into a training set and a test set;

[0009] S3. Use the fracture mechanics model as a physical constraint to construct a physically guided Wasserstein generative adversarial network, and use the data in the training set to train the physically guided Wasserstein generative adversarial network to obtain a low-fidelity fatigue dataset.

[0010] S4, building a deep neural network and pre-training the deep neural network based on the low-fidelity fatigue dataset;

[0011] S5. Introduce the influence of critical defect geometry characteristics on fatigue performance as physical constraints in deep neural networks to construct physical information neural networks;

[0012] S6. Using the concept of transfer learning, the training set data is used as high-fidelity fatigue data to fine-tune the parameters of the physical information neural network. The multi-fidelity physical information neural network is constructed through pre-training with low-fidelity fatigue data and fine-tuning with high-fidelity fatigue data. The prediction effect of the multi-fidelity physical information neural network is tested using the test set data to obtain the optimized multi-fidelity physical information neural network.

[0013] S7. Use an optimized multi-fidelity physical information neural network to input the test conditions, critical defect geometric characteristics, and static mechanical properties of a metal to be tested, and output its fatigue life prediction results.

[0014] Beneficial effects of the present invention: uncertainty quantification of fatigue performance and fitting of low-fidelity fatigue data with physical consistency are achieved through a physically guided Wasserstein generative adversarial network. The concept of transfer learning is introduced, allowing the use of multi-fidelity fatigue data to train physical information neural networks during the training process. The impact of manufacturing defects on fatigue performance is embedded as a physical constraint, ensuring the physical consistency of the prediction results of the multi-fidelity physical information neural network. Compared with the prior art, the method of the present invention has the following advantages:

[0015] (1) The present invention introduces the concept of transfer learning to establish a multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure. Based on a generative machine learning algorithm, low-fidelity fatigue data is fitted at low cost, and the test data is used as high-fidelity fatigue data. The influence of defect geometric characteristics on fatigue is introduced as a physical constraint, and a fatigue life prediction process based on metal defect characteristics under small sample conditions is established;

[0016] (2) The present invention introduces a fracture mechanics model as a physical constraint to construct a physically guided Wasserstein generative adversarial network, and uses the adversarial learning framework of the Wasserstein generative adversarial network to achieve uncertainty quantification of metal fatigue performance under defect-induced failure and capture of the potential distribution of critical defect geometric features. The physics-based discriminator established through the fracture mechanics model ensures the physical consistency of the fitted low-fidelity fatigue data, thus overcoming the inherent disadvantages of traditional models in generating low-fidelity data.

[0017] (3) The present invention makes comprehensive use of low-fidelity and high-fidelity fatigue data, so that the life prediction results of metals containing manufacturing defects have small dispersion and high accuracy. The introduced physical constraints further enhance the generalization ability and prediction stability of the present invention when facing data outside the training set and ensure the physical consistency of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of the multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure provided by the present invention;

[0019] Figure 2 This is a schematic diagram of the structure of a physically guided Wasserstein generative adversarial network in an embodiment of the present invention;

[0020] Figure 3 This is a schematic diagram of the physical information neural network structure in an embodiment of the present invention;

[0021] Figure 4 It is a physical loss history curve diagram of the physical information neural network in an embodiment of the present invention;

[0022] Figure 5Schematic diagram comparing the prediction results and test results of the additively manufactured specimen using the method of the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] like Figure 1 As shown, the embodiment of the present invention discloses a multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure, comprising the following steps:

[0025] S1. Obtain the test conditions, critical defect geometric characteristics, static mechanical properties and corresponding fatigue life test results of the same grade series of metals to form an original fatigue database;

[0026] S2. Expand and preprocess the data in the fatigue database using physical information feature engineering, and divide the data in the fatigue database into a training set and a test set;

[0027] S3. Use the fracture mechanics model as a physical constraint to construct a physically guided Wasserstein generative adversarial network, configure the network structure and hyperparameters, and use the training set data to train the physically guided Wasserstein generative adversarial network to obtain a low-fidelity fatigue dataset.

[0028] S4, building a deep neural network, and pre-training the deep neural network based on the low-fidelity fatigue dataset;

[0029] S5. Introduce the influence of critical defect geometry characteristics on fatigue performance as physical constraints in deep neural networks to construct physical information neural networks;

[0030] S6. Using the concept of transfer learning, the training set data is used as high-fidelity fatigue data to fine-tune the parameters of the physical information neural network. The multi-fidelity physical information neural network is constructed through pre-training with low-fidelity fatigue data and fine-tuning with high-fidelity fatigue data. The prediction effect of the multi-fidelity physical information neural network is tested using the test set data to obtain the optimized multi-fidelity physical information neural network.

[0031] S7. Use an optimized multi-fidelity physical information neural network to input the test conditions, critical defect geometric characteristics, and static mechanical properties of a metal to be tested, and output its fatigue life prediction results.

[0032] Below, each step of the present invention is further described.

[0033] S1. Obtain the test conditions, critical defect geometric characteristics, static mechanical properties and corresponding fatigue life test results of the same grade series of metals to form an original fatigue database. In this embodiment, the material is selected as additive manufacturing Ti-6Al-4V, and the construction of the original fatigue database specifically includes:

[0034] S11. Define the defect size as the square root of the projection area of ​​the defect perpendicular to the load direction. Area represents the projection area of ​​the defect perpendicular to the load direction;

[0035] S12. Calculate the relative position of the defect using the defect size calculated by S11. The function expression of the relative position of the defect is:

[0036]

[0037] Where d represents the shortest distance between the defect and the free surface of the specimen;

[0038] S13. Using the defect size calculated by S11, the defect shape is characterized by the defect roundness c. The defect roundness is calculated as follows:

[0039]

[0040] Where R represents the maximum distance from the geometric center of the projected area to the boundary. The original fatigue database constructed for this material is shown in Table 1.

[0041] Table 1 Fatigue database of additively manufactured Ti-6Al-4V

[0042]

[0043] In Table 1, σ u is the ultimate tensile strength; in this embodiment, considering that there are many data points in the data set, Table 1 uses the minimum value, maximum value, average value and standard deviation of each feature to characterize the fatigue database.

[0044] S21. Use stress intensity factor to characterize the local stress concentration caused by defects and use stress intensity factor as an extended feature;

[0045] Furthermore, the stress intensity factor is calculated as:

[0046]

[0047] Among them, △σ represents the stress range and Y represents the position factor.

[0048] S22. Using the normalized SN curve as the driving force for crack extension, the normalized SN curve is used as the extension feature. The function expression of the normalized SN curve is:

[0049]

[0050] Among them, N f represents fatigue life, σ w represents fatigue limit, and A and B represent material parameters.

[0051] The expression of fatigue limit is:

[0052]

[0053] Where HV represents Vickers hardness, σ w represents fatigue limit, A and B represent material parameters, and the expression of fatigue limit is:

[0054]

[0055] Where HV represents Vickers hardness and μ represents position factor. The extended fatigue database constructed for this material is shown in Table 2.

[0056] Table 2 Extended fatigue database for additively manufactured Ti-6Al-4V

[0057] feature Minimum Maximum average value Standard Deviation △K 8μm 331μm 66.459μm 50.807μm <![CDATA[△σ / σ w ]]> 427.5MPa 799MPa 636.476MPa 100.528MPa Y 0.5 0.65 0.546 0.070

[0058] S23, compare the test conditions (△σ), critical defect geometry characteristics ( h, c, d), static mechanical properties (σ u ), extended characteristics (△K, △σ / σ w , Y) are normalized to the interval of [-1,1] as input features for constructing a multi-fidelity physical information neural network;

[0059] S24. Logarithmically normalize the fatigue life to the interval [-1,1], where the normalized function expression is:

[0060]

[0061] S3. The fracture mechanics model is used as physical information to construct a physics-based discriminator. Specifically, the Paris equation is introduced as the fracture mechanics model, and its expression is:

[0062]

[0063] Where a is the crack length, C and m are material parameters. The physics-based discriminator is introduced into the Wasserstein generative adversarial network structure, such as Figure 2 As shown in Figure 2, the physical constraints construct the optimization objective of the physical guidance Wasserstein generative adversarial network:

[0064]

[0065]

[0066] Among them, G is the generator, D p is the physics-based discriminator, D d is a purely data-driven discriminator, V W (·) is the Wasserstein distance, V PW (·) is the optimization objective of the physical guided Wasserstein generative adversarial network. The structure and parameter settings of the constructed physical guided Wasserstein generative adversarial network are shown in Table 3.

[0067] Table 3 Structure and parameter settings of the physical guided Wasserstein generative adversarial network

[0068]

[0069] The activation function in the physical-guided Wasserstein generative adversarial network adopts the Leaky-ReLU function, and its function expression is:

[0070]

[0071] Among them, α is a coefficient greater than zero, and x is the input of the neurons in the physically guided Wasserstein generative adversarial network.

[0072] The physics-guided Wasserstein generative adversarial network constructed in step S3 is trained using the training set obtained in step S2 to fit the test conditions, critical defect geometric characteristics, static mechanical properties and the potential distribution of the corresponding fatigue life, and to achieve uncertainty quantification of fatigue life. After the physics-guided Wasserstein generative adversarial network is fully trained, random noise is used as the input of the generator, and the generator outputs the simulated test conditions, critical defect geometric characteristics, static mechanical properties and their corresponding fatigue life, thereby constructing a low-fidelity fatigue data set.

[0073] S4. Build a deep neural network, configure the network structure and hyperparameters, and pre-train the deep neural network based on the low-fidelity fatigue dataset, including:

[0074] S41. The number of hidden layers of the deep neural network is configured to be 6. Each hidden layer in the network structure contains 144 neurons, the learning rate is set to 0.001, the regularization coefficient is set to 0, and the activation function uses the Tanh function, whose function expression is:

[0075]

[0076] S42, using low-fidelity fatigue data to optimize the parameters of the deep neural network, and the number of training rounds is set to 500;

[0077] S43. After the deep neural network is fully pre-trained on the low-fidelity fatigue dataset, the network structure and parameters of the pre-trained deep neural network are saved locally.

[0078] S5. The effects of defect size, defect location and defect shape on fatigue performance are summarized and extracted into a programmable partial differential inequality form, as shown in Table 4.

[0079] Table 4 Effect of defect geometric characteristics on fatigue performance and its mathematical description

[0080]

[0081] The effects of defect size, defect location and defect shape on fatigue performance are used as constraints in the deep neural network optimization process:

[0082] minLoss mse

[0083]

[0084] Among them, Loss mse is the optimization target of the deep neural network, and its calculation method is:

[0085]

[0086] Among them, N p,i is the fatigue life predicted by the deep neural network corresponding to the i-th training data, and n is the number of fatigue data used for model training. Furthermore, the exterior point penalty function method is used to transform this constrained optimization problem into an unconstrained optimization problem to construct a physical information neural network:

[0087] minLoss P

[0088] Loss P =(1-λ)Loss mse +λLoss phy

[0089] Among them, λ is the relaxation factor, Loss P is the loss function of the physical information neural network, Loss phy is the physical loss function. Furthermore, based on the exterior point penalty function method, the physical loss function is calculated as follows:

[0090]

[0091]

[0092] Where j is the number of high-fidelity fatigue data in the training set. The constructed physical information neural network is as follows: Figure 3 Shown

[0093] S61, setting the network structure and activation function of the physical information neural network to be consistent with the pre-trained deep neural network;

[0094] S62, using the concept of transfer learning, allowing the physical information neural network to read the network structure and parameters of the deep neural network after pre-training in step S4 stored locally, and limiting the number of hidden layers that can be updated to the last layer to retain the feature information extracted from the low-fidelity fatigue data;

[0095] S63, configure the learning rate of the physical information neural network to be 0.001, the regularization coefficient to be 0, and the physical constraint strength to be 0.05;

[0096] S64, using the fatigue data of the training set as high-fidelity fatigue data, fine-tuning the parameters of the updateable hidden layer, so as to realize the construction of a multi-fidelity physical information neural network by pre-training the deep neural network with low-fidelity fatigue data and fine-tuning the physical information neural network with high-fidelity fatigue data. P As the optimization target, when the model optimization target no longer decreases, it is considered that the prediction performance of the constructed multi-fidelity physical information neural network meets the requirements. The test set data is used to verify the extrapolation performance of the fine-tuned multi-fidelity physical information neural network. Further, as Figure 4 As shown, the introduced physical loss ensures the physical consistency of the predictions of the multi-fidelity physics-informed neural network.

[0097] S7. The test conditions, critical defect geometric characteristics, and static mechanical properties of the additively manufactured Ti-6Al-4V to be tested are first input into a multi-fidelity physical information neural network for prediction with feature expansion, and the predicted value of the fatigue life of the additively manufactured alloy to be tested is output.

[0098] Figure 5 The figure is a comparison of the prediction results and the experimental results according to the method of the present invention. The comparison between the experimental results and the prediction results shows that the multi-fidelity physical information neural network is used to couple low-fidelity and high-fidelity fatigue data, so that the dispersion of the metal fatigue life prediction results under defect-induced failure is small and the accuracy is high. The physical constraints introduced further improve the generalization ability and prediction stability of the present invention when facing data outside the training set and ensure the physical consistency of the prediction results. It also illustrates the rationality and effectiveness of the multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure.

[0099] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. For those skilled in the art, the present invention may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure, characterized in that: include: S1. Obtain the test conditions, critical defect geometric characteristics, static mechanical properties and corresponding fatigue life test results of the same grade series of metals to form an original fatigue database; S2. Expand and preprocess the data in the fatigue database using physical information feature engineering to obtain an expanded fatigue database, and divide the data in the expanded fatigue database into a training set and a test set; S3. Use the fracture mechanics model as a physical constraint to construct a physically guided Wasserstein generative adversarial network, and use the data in the training set to train the physically guided Wasserstein generative adversarial network to obtain a low-fidelity fatigue dataset. S4, building a deep neural network and pre-training the deep neural network based on the low-fidelity fatigue dataset; S5. Introduce the influence of critical defect geometry characteristics on fatigue performance as physical constraints in deep neural networks to construct physical information neural networks; S6. Using the concept of transfer learning, the training set data is used as high-fidelity fatigue data to fine-tune the parameters of the physical information neural network. The multi-fidelity physical information neural network is constructed through pre-training with low-fidelity fatigue data and fine-tuning with high-fidelity fatigue data. The prediction effect of the multi-fidelity physical information neural network is tested using the test set data to obtain the optimized multi-fidelity physical information neural network. S7. Use an optimized multi-fidelity physical information neural network to input the test conditions, critical defect geometric characteristics, and static mechanical properties of a metal to be tested, and output its fatigue life prediction results.

2. The multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure according to claim 1 is characterized in that: The test conditions include a stress range Δσ; Critical defect geometry features include: defect size The relative position of the defect h, the roundness of the defect c, and the closest distance d between the defect and the free surface of the specimen; Static mechanical properties include ultimate tensile strength σ u ; The fatigue life test results are specifically stress range △σ, defect size The relative position of the defect h, the roundness of the defect c, the closest distance between the defect and the free surface of the specimen d, and the ultimate tensile strength σ u The corresponding fatigue life N f .

3. The multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure according to claim 2 is characterized in that: The process of expanding the data in the fatigue database is as follows: First, according to the stress range △σ, defect size Calculate the stress intensity factor △K: Where Y represents the position factor; The normalized SN curve is used as the driving force of crack extension and the normalized SN curve is used as the extension feature. The function expression of the normalized SN curve is: Among them, N f represents fatigue life, σ w represents fatigue limit, and A and B represent material parameters.

4. The multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure according to claim 3 is characterized in that: The extended features included in the expanded fatigue database are △K, △σ / σ w , Y.

5. The multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure according to claim 4 is characterized in that: The specific implementation process of step S3 is: The Paris equation is introduced as the fracture mechanics model, and its expression is: Among them, a is the crack length, C and m are material parameters; Fracture mechanics models as physics-based discriminators; The physics-based discriminator is introduced into the Wasserstein generative adversarial network structure, and the physical constraints construct the optimization goal of the physically guided Wasserstein generative adversarial network: Among them, G is the generator, D p is the physics-based discriminator, D d is a purely data-driven discriminator, V W (·) is the Wasserstein distance, V PW (·) Optimization objectives for physically guided Wasserstein generative adversarial networks; The activation function in the physical-guided Wasserstein generative adversarial network adopts the Leaky-ReLU function, and its function expression is: Among them, α is a coefficient greater than zero, and x is the input of the neuron in the physical guidance Wasserstein generative adversarial network The constructed physics-guided Wasserstein generative adversarial network is trained using the training set obtained in step S2 to fit the potential distribution of test conditions, critical defect geometric characteristics, static mechanical properties and corresponding fatigue life, and to realize the uncertainty quantification of fatigue life. After the training of the physics-guided Wasserstein generative adversarial network is completed, random noise is used as the input of the generator, and the generator outputs the simulated test conditions, critical defect geometric characteristics, static mechanical properties and corresponding fatigue life, thereby constructing a low-fidelity fatigue data set.

6. The multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure according to claim 5 is characterized in that: Step S4 specifically includes the following sub-steps: S41. Configure the number of hidden layers of the deep neural network, the number of neurons in each hidden layer of the network structure, set the learning rate, and the regularization coefficient. The activation function uses the Tanh function, and its function expression is: S42. Optimize the parameters of the deep neural network using the low-fidelity fatigue dataset and set the training rounds; S43. After the deep neural network is trained on the low-fidelity fatigue dataset, the network structure and parameters of the trained deep neural network are saved locally.

7. The multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure according to claim 6 is characterized in that: The physical constraints described in step S5 include: The optimization problem based on physical constraints is: Among them, Loss mse is the optimization target of the deep neural network, and its calculation method is: Among them, N p,i Fatigue life prediction for deep neural networks; The exterior point penalty function method is used to transform the optimization problem based on physical constraints into an unconstrained optimization problem to construct a physical information neural network: minLoss P Loss P =(1-λ)Loss mse +λLoss phy Among them, λ is the relaxation factor, Loss P is the loss function of the physical information neural network, Loss phy is the physical loss function.

8. The multi-fidelity physical information neural network fatigue life prediction method under defect-induced failure according to claim 7 is characterized in that: Step S6 includes the following sub-steps: S61, using the concept of transfer learning, allowing the physical information neural network to read the network structure and parameters of the deep neural network trained in step S4 stored locally, and limiting the number of updateable hidden layers to the last layer to retain the feature information extracted from the low-fidelity fatigue data; S62. Configure the learning rate, regularization coefficient, and physical constraint strength of the physical information neural network; S63. Use the fatigue data of the training set as high-fidelity fatigue data, and fine-tune the parameters of the updateable hidden layer until the prediction performance of the multi-fidelity physical information neural network meets the requirements.

Images