S-N curve prediction method based on gated circulation network, medium, equipment and product

Abstract

The invention provides an S-N curve prediction method based on a gated circulation network, a medium, equipment and a product, and the method comprises the steps: obtaining fatigue performance data obtained through a fatigue performance test in advance, and the fatigue performance data comprise stress amplitude and circulation cycles; performing preset processing on the obtained fatigue performance data to obtain sample data for model training; performing gating circulation network model training based on the sample data to obtain an S-N curve prediction model; and performing loop prediction by using the obtained S-N curve prediction model, and outputting an S-N curve. According to the scheme provided by the invention, accurate prediction of the S-N curve can be realized based on a small amount of high-stress test data, and the time cost and the economic cost required by a fatigue test are greatly reduced.

Description

Technical Field

[0001] The present invention relates to the field of control, and in particular to a SN curve prediction method based on a gated cyclic network, a storage medium, a computing device and a program product. Background Art

[0002] High-end fields such as aviation, aerospace, and nuclear power have developed rapidly in recent years. As the key core components of high-end equipment, the service environment of bearings has become increasingly harsh. Fatigue failure is the most important failure form of bearings during service. In order to meet the service performance requirements of high fatigue life, fatigue performance evaluation and anti-fatigue design of bearing steel become very important.

[0003] The SN curve is a stress-life curve, which represents the relationship between the fatigue life and fatigue strength of a standard specimen under certain loading conditions. It is one of the most basic methods for material fatigue performance evaluation and fatigue resistance design. However, the method of obtaining the SN curve through fatigue testing is costly and time-consuming. If you want to meet the high-precision requirements, you need to obtain multiple sets of data through long-term fatigue testing to fit the SN curve, so it cannot meet the low-cost and high-efficiency evaluation requirements. Summary of the invention

[0004] The main purpose of the present invention is to overcome the defects of the above-mentioned related technologies and provide a SN curve prediction method, medium, equipment and product based on a gated cyclic network to solve the problem of high cost and long time consumption in the method of obtaining SN curve through fatigue test in the related technology.

[0005] On the one hand, the present invention provides an SN curve prediction method based on a gated recurrent network, comprising: obtaining fatigue performance data obtained in advance through a fatigue performance test, the fatigue performance data including: stress amplitude and cycle number; performing preset processing on the obtained fatigue performance data to obtain sample data for model training; performing gated recurrent network model training based on the sample data to obtain an SN curve prediction model; performing cyclic prediction using the obtained SN curve prediction model to output an SN curve.

[0006] Optionally, the fatigue performance test includes at least one of rotational bending fatigue, tension-compression fatigue and torsional fatigue.

[0007] Optionally, the acquired fatigue performance data is subjected to preset processing, including:

[0008] The stress amplitude data in the fatigue performance data are interpolated to unify the data format and expand the data volume; wherein the stress amplitude data obtained after the interpolation process are equally spaced in the direction of the cycle number; the stress amplitude data obtained after the interpolation process are normalized to be converted into dimensionless data; the stress amplitude data after the interpolation process and the normalization process are serialized, wherein the sequence length of the stress amplitude sequence is set according to the data volume.

[0009] Optionally, when performing interpolation processing, the range of the cycle number is set to 9×10 4 Up to 1×10 7 ; and / or, the amount of stress amplitude data obtained after interpolation processing is set to 800-1500; and / or, the sequence length of the stress amplitude sequence is set to 10-80.

[0010] Optionally, the interpolation processing includes: performing interpolation processing on the stress amplitude data using a linear interpolation method; and / or the normalization processing includes: mapping the range of the stress amplitude data to a range of 0 to 1 using a MinMax Scaler.

[0011] Optionally, it also includes: dividing the sample data into a training set, a validation set and a test set according to a preset ratio; training a gated recurrent network model based on the sample data to obtain an SN curve prediction model, including: constructing gated recurrent network models of different structures, and setting different numbers of neurons for the gated recurrent network models of different structures; using the training set to train the gated recurrent network models of different structures respectively; for the gated recurrent network models of different structures, by evaluating their performance on the validation set, determining the structure with the best performance and adjusting its hyperparameters; using the adjusted hyperparameters, reusing the training set and the validation set to perform model training on the structure with the best performance to obtain an SN curve prediction model.

[0012] Optionally, a gated recurrent network model is constructed, including: a loss function adopts a root mean square error; and / or a model optimizer adopts an Adam optimizer; and / or a GRU layer activation function adopts a hyperbolic tangent function tanh; and / or a GRU layer recurrent activation function adopts Sigmoid; and / or an output layer activation function adopts a linear activation function.

[0013] Optionally, it also includes: dividing the sample data into a training set, a validation set and a test set according to a preset ratio; using the obtained SN curve prediction model to perform cyclic prediction and output the SN curve, including: inputting the last stress amplitude sequence of the validation set with a length of L into the prediction model, and outputting the first stress amplitude of the predicted test set; sliding the window backward by a data length, inputting the last L-1 stress amplitudes of the validation set and the predicted first stress amplitude of the test set into the prediction model, and outputting the second stress amplitude of the predicted test set; and so on, cyclically outputting the prediction results of the prediction model for the test set until the target cycle number is reached to obtain the SN curve.

[0014] Another aspect of the present invention provides a storage medium having a computer program stored thereon, wherein the program implements the steps of any of the aforementioned methods when executed by a processor.

[0015] In another aspect, the present invention provides a computing device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the aforementioned methods when executing the program.

[0016] In another aspect, the present invention provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of any of the aforementioned methods are implemented.

[0017] According to the technical solution of the present invention, by learning the data under high cyclic stress with the help of GRU network, the accurate prediction of the trend of the later stage of SN curve can be achieved, thus greatly reducing the number of fatigue tests required. In addition, since the number of cycles in the later stage of SN curve is higher and the required test time is longer, the present invention can greatly reduce the time cost and economic cost required to obtain SN curve through fatigue test.

[0018] In addition, the present invention also includes the following beneficial effects:

[0019] (1) Compared with recurrent neural networks, gated recurrent units have a simpler structure, faster model convergence, and are easier to train.

[0020] (2) The fatigue SN curves of different types, different processing methods, different service environments and different loading forms are highly versatile.

[0021] (3) It can greatly reduce the time and economic costs required for fatigue testing, has the characteristics of high efficiency and high precision, and can provide reliable assistance for high-quality anti-fatigue design. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0023] Figure 1 is a method schematic diagram of an embodiment of the SN curve prediction method based on the gated recurrent network provided by the present invention;

[0024] Figure 2 This is the structural principle diagram of GRU;

[0025] Figure 3 A schematic diagram of the division of sample data is shown;

[0026] Figure 4 A flowchart of a specific implementation of the steps of training a gated recurrent network model based on the sample data to obtain an SN curve prediction model is shown;

[0027] Figure 5 A prediction result diagram of a specific example of the SN curve prediction method based on the gated recurrent network is shown;

[0028] Figure 6 A schematic diagram showing the evaluation results of the prediction performance of the SN curve prediction model in the test set according to a specific example of the present invention is shown;

[0029] Figure 7 1 is a schematic diagram of a method of a specific embodiment of the SN curve prediction method based on a gated recurrent network provided by the present invention;

[0030] Figure 8 It is a schematic diagram of the SN curve cycle prediction method. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. 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 making creative work are within the scope of protection of the present invention.

[0032] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0033] Since the announcement of the Materials Genome Project, methods for predicting and evaluating material fatigue performance have been continuously innovating and evolving. Many researchers have tried to combine artificial intelligence methods to replace traditional experiments and modeling to reduce the time and economic costs required for material performance testing and evaluation. At present, machine learning methods have been widely used in research in the field of materials science, and have successfully predicted the fatigue life, fatigue strength and fatigue crack growth rate of materials under given conditions, but there are still few prediction methods for SN curves.

[0034] Recurrent Neural Network (RNN) is an artificial neural network (ANN) that uses sequence data or time series data. Compared with ANN, there are connections between the nodes in the hidden layer of RNN, that is, the output of the network at the current moment is also related to the input at the previous moment. Specifically, the network will remember the previously input information and use it to calculate the current output. RNN is often used to solve sequential or time problems, such as natural language processing, speech recognition, language translation, etc. In theory, RNN can process sequence data of any length, but in practical applications, RNN has two serious problems, namely gradient disappearance and gradient explosion.

[0035] The Gated Recurrent Unit (GRU) is a variant of RNN that aims to solve the gradient problem in long-term memory and backpropagation in RNN. Figure 2The following is a schematic diagram of the structure of GRU. It uses a gating mechanism to control the steps of information input, memory, and forgetting. GRU has two gates, namely a reset gate and an update gate. The reset gate determines how to combine the new input information with the input information at the previous moment, and the update gate defines the amount of previous memory saved to the current moment. In particular, when the reset gate value of GRU is 1 and the update gate value is 0, it is a standard RNN model. Compared with other RNN variants, GRU has a simpler structure and is easier to train.

[0036] The present invention proposes an SN curve prediction method based on a gated cyclic network, which can effectively reduce the time cost and economic cost required for fatigue testing on the basis of achieving accurate prediction, and provide reliable assistance for the performance prediction and anti-fatigue design of high-quality bearing steel.

[0037] Figure 1 It is a method schematic diagram of an embodiment of the SN curve prediction method based on the gated recurrent network provided by the present invention.

[0038] like Figure 1 As shown, according to one embodiment of the present invention, the SN curve prediction method based on the gated recurrent network includes at least step S110, step S120, step S130 and step S140.

[0039] Step S110 , obtaining fatigue performance data obtained in advance through fatigue performance tests as sample data.

[0040] The fatigue performance test may specifically include: at least one of rotational bending fatigue, tension and compression fatigue and torsion fatigue. The fatigue performance data may specifically include: stress amplitude σ and cycle number N. The test material of the fatigue performance test may specifically be a metal material, such as bearing steel.

[0041] Specifically, a rotary bending fatigue test is performed in advance to obtain fatigue performance data. For example, the test material is M50 bearing steel, which is the main steel type for aviation bearings, has good high temperature stability and wear resistance, and is widely used. The rotary bending fatigue test can be performed at room temperature, the stress ratio is set to -1, the rotation speed is set to 5000r / min, and n pairs of data points are finally obtained. For example, n=15 pairs of data points are obtained through the rotary bending fatigue test.

[0042] Step S120, performing preset processing on the acquired fatigue performance data to obtain sample data for training a gated recurrent network model.

[0043] The preset processing may specifically include: interpolation processing, normalization processing and serialization processing.

[0044] (1) performing interpolation processing on the stress amplitude data in the fatigue performance data to unify the data format and expand the data volume;

[0045] When GRU is used to process time series data, it requires that the time series be equally spaced. Since the intervals between the data points obtained through fatigue tests are not equal, they cannot meet the input requirements of the gated recurrent network. Therefore, the interpolation method is used to unify the data format of the stress amplitude σ in the fatigue performance data and expand the data volume. Among them, the stress amplitude data obtained after interpolation processing is equally spaced in the direction of the cycle number; the method of interpolation processing may specifically include at least one of linear interpolation, parabolic interpolation, cubic spline interpolation and Newton interpolation. By comparing the interpolation effect, preferably, the linear interpolation method is used to interpolate the obtained stress amplitude σ data to unify the data format and expand the data volume. The linear interpolation formula is as follows:

[0046]

[0047] Where (x, y) is the interpolation point to be found, (x0, y0) and (x1, y1) are the known points on both sides of the interpolation point to be found.

[0048] Before interpolation, set the range of the loop number N. For example, the range of the loop number N can be set to 9×10 4 Up to 1×10 7 The stress amplitude σ data obtained after interpolation processing are equally spaced in the direction of the cycle number N. The amount of stress amplitude data obtained after interpolation processing can be set to 800 to 1500, for example, to 800.

[0049] (2) normalizing the stress amplitude data obtained after interpolation processing and converting it into dimensionless data;

[0050] GRU is sensitive to the range of input data. In order to make the sample data set more standardized and standard and the model converge quickly, the data needs to be normalized after interpolation. The normalization methods can include: MinMax normalization or z-score normalization.

[0051] MinMax normalization is also called deviation normalization, which maps the original sample data to the range of [0,1] after linear transformation. Z-score normalization standardizes the data according to the mean and standard deviation of the original data. The processed data conforms to the standard normal distribution, but when the original data is significantly different from the general normal distribution, the effect of standardization is not ideal. In a preferred embodiment, the MinMax Scaler method is used to map the range of stress amplitude data to the range of 0 to 1. Specifically, the MinMaxScaler function in the scikit-learn library in Python can be used for normalization. The default range of the normalized data is [0,1], and the conversion function is as follows:

[0052]

[0053] Among them, X is the data before normalization, X scaled is the normalized data, X max is the maximum value of the sample data, X min is the minimum value of the sample data.

[0054] (3) Serializing the stress amplitude data after interpolation and standardization;

[0055] GRU requires the input data to be serialized data, so it is necessary to input a certain length of stress amplitude sequence before the stress amplitude of the target cycle into the model, and the model outputs the predicted value of the target stress amplitude. The sequence length L can be set according to the amount of data. For example, depending on the amount of data, the value of the sequence length L can be 10 to 80, for example, L = 40. That is, the model learns the 40 stress amplitudes before the stress amplitude corresponding to the target cycle, and then outputs the predicted value of the target stress amplitude. The shape formats of the input and output data after serialization are X i×j×k and Y i×k , where i is the number of samples, for example, 760; j is the length of the sequence, for example, 40; and k is 1.

[0056] Step S130, performing gated recurrent network model training based on the sample data to obtain an SN curve prediction model.

[0057] First, the sample data obtained after the preset processing is divided into a training set, a validation set and a test set according to a preset ratio, so as to use the training set to train the gated recurrent network model, use the validation set to adjust the network structure and other hyperparameters of the trained gated recurrent network model, and then use the training set and the test set to retrain the gated recurrent network model.

[0058] Specifically, in order to improve the accuracy of model prediction, all sample data are divided into three data sets. The division diagram can be referred to Figure 3 As shown, Figure 3 A schematic diagram of the division of sample data is shown. For example, the division ratio of the training set, validation set, and test set is 4:1:5. For example, depending on the size of the data, the division ratio of the data set can be set to: the training set accounts for 40% to 60%, the validation set accounts for 10%, and the rest is the test set. For example, the first 40% of the high cycle stress data is used as a training set to train the gated cyclic network model; the middle 10% of the data is used as a validation set. By observing the performance of the model on the validation set after the training is completed, the network structure and other hyperparameters of the network can be adjusted to avoid overfitting and underfitting of the model; after the adjustment is completed, the model is retrained using the training set and validation set, and the generalization ability of the model is evaluated on the test set.

[0059] Figure 4 A flowchart of a specific implementation of the steps of training a gated recurrent network model based on the sample data to obtain an SN curve prediction model is shown.

[0060] like Figure 4 As shown, in a specific implementation, step S130 includes: step S131, step S132 and step S133.

[0061] Step S131, constructing gated recurrent network models with different structures, and setting different numbers of neurons for the gated recurrent network models with different structures.

[0062] In a specific implementation, the structure of the gated recurrent network includes a GRU layer, a fully connected layer, and an output layer. In order to find a suitable network structure and improve the accuracy of model prediction, gated recurrent network models with different structures can be constructed, and different numbers of neurons can be set for the gated recurrent network models with different structures.

[0063] The different structures include: the fully connected layer is a single-layer structure or the fully connected layer is a double-layer structure, and further, different numbers of neurons are set for different structures. For example, in a specific implementation, the number of neurons included in the GRU layer is uniformly set to 10, and the details of the constructed different network structures are shown in Table 2:

[0064] Table 2

[0065]

[0066] Generally speaking, increasing the depth and width of the network within a certain range can improve the model's ability to learn complex problems. Increasing the depth of the network can improve the model's nonlinear expression ability, allowing the model to learn more complex data patterns. In addition, deepening the network can also reduce the complexity of each layer, allowing the model to extract features layer by layer. But this does not mean that the deeper the model, the better the performance. Too much depth will lead to problems such as gradient instability and network degradation, which will deteriorate the model's performance. Similarly, increasing the width of the network within a certain range can extract richer features from each layer and improve the model's learning ability, while too much width will reduce the model's efficiency. Generally speaking, the cost of increasing the width of the network is much higher than increasing the depth of the network.

[0067] The model training process needs to calculate the error between the predicted value and the true value in each training cycle, so the choice of loss function is very critical. In specific problems, some loss functions calculate the interpolation gradient to decrease quickly, while some decrease slowly, so it is necessary to select a suitable loss function for a specific embodiment. In the embodiment of the present invention, the root mean square error is preferably used as the loss function to calculate the loss during the training process.

[0068] The selection of model optimizer and corresponding learning rate is also a very critical issue. Commonly used optimizers are Adam and SGDM. Among them, Adam is widely used because of its faster convergence speed and higher precision. In addition to Adam, SGMD is also a commonly used model optimizer. Generally speaking, its convergence speed is slower than Adam but its precision is higher. A larger learning rate will improve the training efficiency of the model in the early stage, but it will often lead to large fluctuations and difficulty in convergence in the later stage of model training; and a smaller learning rate will sometimes cause the model training to fall into a local minimum and fail to achieve the best training results. In the embodiment of the present invention, Adam is preferably used as the model optimizer, and the corresponding learning rate is set to 0.001.

[0069] Figure 2 This is the GRU principle diagram. Figure 2 As shown, it contains several important parameters, namely reset gate r, update gate z, hidden state h and candidate hidden state The GRU layer preferably uses the hyperbolic tangent function tanh as the activation function, and its output range is [-1, 1], and the expression is as follows:

[0070]

[0071] In the embodiment of the present invention, Sigmoid is preferably used as the recurrent activation function of GRU, and its output range is [0, 1], and the expression is as follows:

[0072]

[0073] Sigmoid acts on the reset gate to help capture short-term dependencies in the time series. The reset gate signal is used to calculate the candidate hidden state. The calculation formulas are:

[0074] r r =σ(W r ·[x t ,h t-1 ])

[0075]

[0076] Among them, σ is the Sigmoid function, tanh is the hyperbolic tangent function, W r is the weight matrix of the reset gate, W is the weight matrix of the candidate hidden state, x t is the stress amplitude at the current moment, h t-1 is the hidden state of σ at the previous moment, is the candidate hidden state of σ at the current moment, and ⊙ is the Hadamard product.

[0077] Sigmoid acts on the update gate to capture long-term dependencies in the time series. The update gate signal is used to calculate the hidden state h. The calculation formulas are:

[0078] z t =σ(W z ·[x t ,h t-1 ])

[0079]

[0080] Among them, W z is the weight matrix of the update gate, h t is the hidden state of σ at the current moment. In the above, through the two gating units selectively memorizing or ignoring the information in the input data, GRU can better capture the important patterns in the sequence.

[0081] Step S132, using the training set to train the gated recurrent network models of different structures respectively.

[0082] Step S133, for the gated recurrent network models with different structures, determine the structure with the best performance by evaluating their performance on the validation set, and adjust its hyperparameters.

[0083] Use the training set to train the model. By observing the performance of the model on the validation set after training, the network structure and other hyperparameters of the network can be adjusted. For models with different structures, after training on the training set, by comparing their performance on the validation set, the best performing structure can be selected, and its hyperparameters can be adjusted to obtain the best hyperparameters.

[0084] The evaluation index of the model can be, for example, the coefficient of determination R 2 , mean absolute error MAE and root mean square error RMSE. For example, the above three indicators are used to evaluate the accuracy and generalization ability of the model. The corresponding calculation formula is as follows:

[0085] Table 1

[0086]

[0087] In Table 1, y i,true is the true value, y i,pre is the predicted value of the gated recurrent network, for y i,true The average value of .

[0088] The training set is used to train networks with different structures, and the model performances of different structures are compared on the validation set. It is found that when the number of neurons is equal, the overall prediction error of the double hidden layer is smaller and the result is more stable.

[0089] Taking all the above factors into consideration, the present invention adopts a 10×15×15×1 network structure to construct an SN curve prediction model, in which the GRU layer contains 10 GRU units, the fully connected layer is a two-layer structure, each layer contains 15 neurons, and the output layer contains 1 neuron. After the model is constructed, the training set and the validation set are used to retrain the model.

[0090] Step S134, using the adjusted hyperparameters, reusing the training set and the validation set to perform model training on the structure with the best performance to obtain the SN curve prediction model.

[0091] After determining the best performing structure and adjusting the hyperparameters, the adjusted hyperparameters are used to re-use the training set and the validation set to train the model of the best performing structure to obtain the SN curve prediction model.

[0092] Step S140, using the obtained SN curve prediction model to perform cyclic prediction and output the SN curve.

[0093] After the model training is completed, the obtained SN curve prediction model is used to predict the test set and output a complete SN curve. Specifically, the SN curve is predicted within the test set, the complete SN curve is output, and the model performance is evaluated based on the prediction results.

[0094] The learning strategy of the GRU model is to read in the serialized stress amplitude data for learning, and obtain the stress amplitude of the next cycle. In the actual application process, the data under the low stress amplitude in the late stage of the curve is unknown, so the output strategy of the model is adjusted as follows: the last stress amplitude sequence of length L in the validation set is input into the prediction model, and the first stress amplitude data of the predicted test set is output; the last L-1 stress amplitude data of the validation set and the first stress amplitude data of the predicted test set are input into the prediction model, and the second stress amplitude data of the predicted test set is output, and so on, and the above cycle steps are repeated until the specified target cycle is reached. The method of applying the cyclic output outputs the prediction results of the test set, so the first result of the test set should be generated by the prediction of the last L data of the input validation set, and then the window moves back to generate the 2nd to nth results in turn.

[0095] For example, when the length of the stress amplitude sequence is L = 40, the last L data of the validation set are input into the model as a sequence, and the first point of the predicted test set is output; the window slides backward, and the last 39 stress amplitude data of the validation set and the first stress amplitude data of the predicted test set are input into the model, and the second data of the predicted test set is output, and the above cycle steps are repeated until the specified cycle number. The low stress amplitude data of the SN curve obtained under this strategy are all generated by the cyclic iterative prediction after the model learns the data under the high stress amplitude, and finally outputs the complete prediction result.

[0096] Figure 8 A schematic diagram of the SN curve cycle prediction method is shown. Figure 8 As shown, assuming that the stress amplitude sequence length L = 2, points 1 and 2 are the last two points of the validation set, the model reads the last two points of the validation set (i.e., points 1 and 2) and predicts the first point of the test set, i.e., point 3, and then the window slides back one (data) point, the model reads the last point of the validation set (i.e., point 2) and the first point of the test set predicted, the second point of the test set predicted, i.e., point 4, and then the window slides back one (data) point, the model reads the first two points of the predicted test set (i.e., points 3 and 4), the third point of the test set predicted, i.e., point 5, and so on.

[0097] The front data divided according to the preset ratio is the training set, which has high cycle stress σ and short cycle life N; the back data is the test set, which has low cycle stress σ and longer cycle life N. Therefore, if the trend of the front part of the learning curve can be predicted by the trend of the back part of the curve, a complete SN curve can be obtained. Therefore, only fatigue tests under high stress are needed, and the cycle life is short, which can greatly reduce the time required for the test. At the same time, compared with the test under the full stress range, the number of samples required for the test can be reduced.

[0098] Figure 5 FIG. 4 shows a prediction result diagram of a specific example of the SN curve prediction method based on the gated recurrent network. Figure 5 It can be seen that the prediction method provided by the present invention has extremely high accuracy. Only under extremely high cycle times, there is a slight deviation between the predicted SN curve and the true SN curve. This deviation is limited by the accumulation of errors in the cyclic iteration output process and can be ignored. Compared with the traditional method, the method provided by the present invention greatly reduces time and economic costs while taking into account accuracy.

[0099] Figure 6 FIG. 1 is a schematic diagram showing the evaluation results of the prediction performance of the SN curve prediction model based on the gated recurrent network in the test set according to a specific example of the present invention. Figure 6 As shown, the determination coefficient R 2 , mean absolute error (MAE) and root mean square error (RMSE). To avoid contingency during operation, all the above results are the average values ​​obtained after 5 independent runs of the model. The left ordinate is used to represent the determination coefficient R 2 The right ordinate is used to represent the mean absolute error MAE and the root mean square error RMSE, and the model determination coefficient R 2 The final score is 0.97, and the values ​​of MAE and RMSE are 0.99MPa and 1.18MPa respectively. It can be seen that the accuracy and reliability of this method for the SN curve prediction task are good.

[0100] Figure 7 FIG. 1 is a schematic diagram of a specific embodiment of the SN curve prediction method based on a gated recurrent network provided by the present invention. Figure 7As shown in the figure, fatigue performance data is obtained through fatigue test, and the stress amplitude σ data format is unified and the data volume is increased by using interpolation method; the above data are normalized and serialized to meet the requirements of the gated recurrent network model for the input data format; the above data are divided according to a certain ratio to obtain several training sets, validation sets and test sets; a gated recurrent network model is constructed, and the model is trained with the training set, and its hyperparameters are adjusted according to the performance of the model on the validation set to obtain the optimal hyperparameters, and then the model is retrained with the training set and validation set to obtain the required model; the obtained model is used for cyclic prediction, and finally a complete SN curve is output.

[0101] The present invention also provides a storage medium corresponding to the SN curve prediction method based on the gated recurrent network, on which a computer program is stored, and when the computer program is executed by a processor, the steps of any of the aforementioned methods are implemented.

[0102] The present invention also provides a computing device corresponding to the SN curve prediction method based on the gated recurrent network, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the aforementioned methods when executing the computer program.

[0103] The present invention also provides a computer program product corresponding to the SN curve prediction method based on the gated recurrent network, including a computer program, and when the computer program is executed by a processor, the steps of any of the aforementioned methods are implemented.

[0104] Accordingly, the scheme provided by the present invention, the SN curve prediction method based on the gated cyclic network can greatly reduce the time cost and economic cost of the traditional fatigue test method under the premise of ensuring accurate prediction. At the same time, the technical scheme of the present invention is not subject to many restrictions such as test parameters, research materials, fatigue forms, etc., and has extremely high versatility. In the scheme provided by the present invention, the prediction of the SN curve only needs to be carried out by high-stress fatigue testing, and its rules are learned to predict the complete SN curve, the number of test samples required is reduced, and the fatigue test time under high stress is shorter.

[0105] The functions described herein may be implemented in hardware, software executed by a processor, firmware, or any combination thereof. If implemented in software executed by a processor, the functions may be stored as one or more instructions or codes on a computer-readable medium or transmitted via a computer-readable medium. Other examples and implementations are within the scope and spirit of the present invention and the appended claims. For example, due to the nature of software, the functions described above may be implemented using software executed by a processor, hardware, firmware, hard wiring, or a combination of any of these. In addition, each functional unit may be integrated into a processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit.

[0106] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0107] The units described as separate components may or may not be physically separated, and the components of the control device may or may not be physical units, that is, they may be located in one place or distributed in multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0108] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the relevant technology or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0109] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the claims of the present invention.

Claims

1. A SN curve prediction method based on gated recurrent network, It is characterized in that include: Acquire fatigue performance data obtained in advance through fatigue performance tests, wherein the fatigue performance data includes: stress amplitude and cycle number; Performing preset processing on the acquired fatigue performance data to obtain sample data for model training; Perform gated recurrent network model training based on the sample data to obtain an SN curve prediction model; The obtained SN curve prediction model is used to perform cyclic prediction and output the SN curve.

2. The method according to claim 1, It is characterized in that The fatigue performance test includes at least one of rotational bending fatigue, tension and compression fatigue and torsional fatigue.

3. The method according to claim 1, It is characterized in that The acquired fatigue performance data is subjected to preset processing, including: Interpolation processing is performed on the stress amplitude data in the fatigue performance data to unify the data format and expand the data volume; wherein the stress amplitude data obtained after the interpolation processing is distributed at equal intervals in the direction of the cycle number; The stress amplitude data obtained after interpolation processing is normalized and converted into dimensionless data; The stress amplitude data after interpolation and normalization are serialized, wherein the sequence length of the stress amplitude sequence is set according to the amount of data.

4. The method according to claim 3, It is characterized in that When performing interpolation processing, the range of the cycle is set to 9×10 4 Up to 1×10 7 and / or, The amount of stress amplitude data obtained after interpolation processing is set to 800 to 1500; and / or, The sequence length of the stress amplitude sequence is set to 10-80.

5. The method according to claim 3, It is characterized in that The interpolation processing includes: performing interpolation processing on the stress amplitude data using a linear interpolation method; and / or, The normalization process includes: using MinMax Scaler to map the range of the stress amplitude data to a range of 0 to 1.

6. The method according to claim 1, It is characterized in that Also includes: Dividing the sample data into a training set, a validation set and a test set according to a preset ratio; A gated recurrent network model is trained based on the sample data to obtain an SN curve prediction model, including: Constructing gated recurrent network models with different structures, and setting different numbers of neurons for the gated recurrent network models with different structures; Using the training set to train the gated recurrent network models with different structures respectively; For the gated recurrent network models with different structures, determine the structure with the best performance by evaluating their performance on the validation set, and adjust their hyperparameters; The adjusted hyperparameters are used to reuse the training set and the validation set to perform model training on the structure with the best performance, thereby obtaining an SN curve prediction model.

7. The method according to claim 6, It is characterized in that The constructed gated recurrent network model includes a GRU layer, a fully connected layer and an output layer; Among them, the number of neurons in the GRU layer is 10; the fully connected layer is a two-layer structure, the number of neurons in each layer is 15, and the number of neurons in the output layer is 1.

8. The method according to any one of claims 1 to 7, It is characterized in that Also includes: Dividing the sample data into a training set, a validation set and a test set according to a preset ratio; The obtained SN curve prediction model is used to perform cyclic prediction and output the SN curve, including: Input the last stress amplitude sequence of the validation set with a length of L into the prediction model, and output the first stress amplitude of the predicted test set; Input the last L-1 stress amplitudes of the validation set and the predicted first stress amplitude of the test set into the prediction model, and output the predicted second stress amplitude of the test set; By analogy, the prediction results of the prediction model for the test set are output cyclically until the target cycle number is reached to obtain the SN curve.

9. A computer-readable storage medium, It is characterized in that A computer program is stored thereon, and when the program is executed by a processor, the steps of any method described in claims 1-8 are implemented.

10. A computing device, It is characterized in that The method comprises a processor, a memory and a computer program stored in the memory and executable on the processor, wherein the steps of the method according to any one of claims 1 to 8 are implemented when the processor executes the program.

11. A computer program product, It is characterized in that The invention comprises a computer program, which implements the steps of any one of the methods of claims 1 to 8 when executed by a processor.

Images