Transfer learning fatigue life calculation method and system

Abstract

The invention provides a transfer learning fatigue life calculation method and system, and the method comprises the steps: obtaining training data, and determining key feature information corresponding to a key cycle number; constructing a training data set, taking the key feature information as a source domain data set, and dividing the target domain data set into the training data set and a test data set; establishing a main model and an auxiliary model, and sequentially carrying out coupling cross training on the main model and the auxiliary model by utilizing the source domain data set and the target domain data set; and calculating a difference value between the prediction results of the main model and the auxiliary model, and outputting a minimum training model and a corresponding prediction result when the difference value is smaller than a target difference value. According to the method, the comprehensive model of the main model and the auxiliary model is established, the comprehensive model is high in prediction precision, effectively prevents excessive generalization and is suitable for fatigue life calculation under the small sample condition, the main model and the auxiliary model are used for cross training, the training prediction difference value of the main model and the auxiliary model is minimum as the target, repeated iteration training is carried out, and fatigue life calculation is more accurate. And an optimal solution can be obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of fatigue testing of aero-engine components, and particularly to a method and system for calculating fatigue life by transfer learning. Background Art

[0002] The fatigue phenomenon widely exists in industries such as aviation, transportation, and energy. Fatigue failure of components often causes great harm. It is statistically shown that 70% of mechanical failures are caused by fatigue. Therefore, testing the fatigue performance of materials or structures and evaluating their fatigue life are extremely important for preventing mechanical equipment failures and ensuring the safety of people's lives and property.

[0003] However, due to the complex fatigue working conditions of components, multiple loading conditions, high equipment requirements, long testing time, and high testing costs. Therefore, people have been exploring better fatigue life evaluation methods to minimize test samples as much as possible.

[0004] With the development of artificial intelligence technology, the prediction of fatigue life of materials and structures using data-driven models has attracted more and more researchers' attention, providing a bright prospect for more accurate prediction of fatigue life. However, the existing methods are mainly based on pure data-driven models, failing to effectively incorporate physical constraints into the prediction model, failing to effectively solve the problem of accurate prediction of fatigue life under a small number of samples, and being prone to causing the problem of over-generalization of the model, resulting in low prediction accuracy and even extreme outliers.

[0005] Based on this, the inventors of the present application propose a method and system for calculating fatigue life by transfer learning in order to solve the above technical problems. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the defect of easy over-generalization of the model in the fatigue life prediction process of the pure data-driven learning model in the prior art, and to provide a method and system for calculating fatigue life by transfer learning.

[0007] The present invention solves the above technical problems by the following technical solutions:

[0008] The present invention provides a method for calculating fatigue life by transfer learning, which is characterized by including:

[0009] Step 1, obtaining training data and determining key feature information corresponding to key cycle numbers;

[0010] Step 2, constructing a training data set, using the key feature information as the source domain data set, and dividing the target domain data set into a training data set and a test data set;

[0011] Step 3: Establish a main model and an auxiliary model, and sequentially perform coupled cross-training on the main model and the auxiliary model by using the source domain dataset and the target domain dataset;

[0012] Step 4: Calculate the difference between the prediction results of the main model and the auxiliary model, and output the minimum training model and the corresponding prediction results when the difference is less than the target difference.

[0013] According to an embodiment of the present invention, the step 1 includes:

[0014] Obtain material fatigue test data through experiments or collection, including but not limited to: peak stress, peak strain, number of cycles, strain ratio / stress ratio.

[0015] According to an embodiment of the present invention, the step 2 includes:

[0016] Adopt the PLS regression method to compare the importance of the peak stress, the peak strain, the number of cycles, and the strain ratio / stress ratio on the fatigue life at different early cycles, and use the characteristic information of the most important cycle as the key characteristic information.

[0017] According to an embodiment of the present invention, the step 3 includes:

[0018] 31. First, call the main model, input the source domain dataset and the first target domain training dataset for training. After training, input the target domain test data to obtain the first target domain prediction dataset;

[0019] 32. Then combine the first target domain training dataset and the first target domain test dataset to form the second target domain training dataset;

[0020] 33. Input the second target domain training dataset and the source domain dataset into the auxiliary model for training. After training, input the target domain test data to obtain the second target domain prediction dataset;

[0021] 34. Then combine the second target domain prediction dataset and the first target domain training dataset to form the third target domain training dataset;

[0022] 35. Call the main model, and input the source domain dataset and the third target domain training dataset into the main model for training. After training, input the target domain test data to obtain the third target prediction dataset;

[0023] 36. Calculate the difference between the second target domain prediction data set and the third target domain prediction data set, and compare the difference with the target difference. If the difference is less than the target difference, the calculation ends; if the difference is greater than the target difference, loop through steps 32 - 36.

[0024] According to an embodiment of the present invention, when the difference is less than the target difference, output the main model and the first prediction result or the second prediction result.

[0025] According to an embodiment of the present invention, step 36 further includes:

[0026] If the difference is greater than the target difference, replace the main model and adjust the training parameters.

[0027] According to an embodiment of the present invention, the main model and the auxiliary model are respectively a transfer learning model or a physical model of a weak learner among a support vector machine, a decision tree, and a neural network model.

[0028] The present invention also provides a transfer learning fatigue life calculation system, characterized in that the transfer learning fatigue life calculation system adopts the transfer learning fatigue life calculation method as described above, and the calculation system includes:

[0029] An acquisition module, configured to acquire training data and determine key feature information corresponding to key cycle numbers;

[0030] A construction module, configured to construct a training data set, use the key feature information as the source domain data set, and divide the target domain data set into a training data set and a test data set;

[0031] A training module, configured to establish a main model and an auxiliary model, and sequentially perform coupled cross-training on the main model and the auxiliary model by using the source domain data set and the target domain data set;

[0032] An output module, configured to calculate the difference between the prediction results of the main model and the auxiliary model, and output the minimum training model and the corresponding prediction result when the difference is less than the target difference.

[0033] The present invention also provides an electronic device, characterized in that it includes: a processor and a memory, the memory stores a program or instruction that can run on the processor, and the program or instruction is executed by the processor to implement the transfer learning fatigue life calculation method according to any one of claims 1 - 7.

[0034] The present invention also provides a readable storage medium, characterized in that a program or instructions are stored on the readable storage medium, and when the program or instructions are executed by a processor, the migration learning fatigue life calculation method according to any one of claims 1-7 is implemented.

[0035] The positive and progressive effects of the present invention are as follows:

[0036] For the migration learning fatigue life calculation method of the present invention, a comprehensive model of a main model and an auxiliary model is established. This comprehensive model has high prediction accuracy, effectively prevents over-generalization, and is suitable for fatigue life calculation in the case of small samples. By cross-training the main model and the auxiliary model and aiming at the minimum difference value of the training predictions of the two, multiple iterative trainings are carried out, which is conducive to obtaining an optimal solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The above and other features, properties, and advantages of the present invention will become more apparent from the following description in conjunction with the drawings and embodiments, wherein:

[0038] Figure 1 is a calculation flowchart of an exemplary migration learning fatigue life calculation method of the present invention;

[0039] Figure 2 is a flowchart of an exemplary migration learning fatigue life calculation method of the present invention;

[0040] Figure 3 is a structural view of an electronic device of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following detailed description of the specific embodiments of the present invention is made in conjunction with the drawings.

[0042] Now, the embodiments of the present invention will be described in detail with reference to the drawings. Now, the preferred embodiments of the present invention will be described in detail, and examples thereof are shown in the drawings. In any possible case, the same reference numerals will be used throughout the drawings to represent the same or similar parts. In addition, although the terms used in the present invention are selected from well-known and commonly used terms, some of the terms mentioned in the specification of the present invention may be selected by the applicant according to his or her judgment, and their detailed meanings are described in the relevant parts of the description herein. In addition, it is required to understand the present invention not only by the actual terms used, but also by the meaning implied by each term.

[0043] The following are the explanations of professional terms in the art:

[0044] Machine learning: Machine learning is a branch of artificial intelligence that uses computer algorithms to automatically identify data patterns and make predictions or decisions.

[0045] Transfer learning: It refers to using the knowledge and experience learned in other tasks to accelerate and improve the training and performance of a model when solving a specific task.

[0046] Source domain: It refers to the domain or task to which a trained model or dataset belongs.

[0047] Target domain: It refers to the new domain or task that needs to be solved.

[0048] Refer to Figure 1 and Figure 2 , the present invention proposes a transfer learning fatigue life calculation method, including:

[0049] Step 1: Obtain training data and determine the key feature information corresponding to the key cycle number.

[0050] It should be noted that the material fatigue test data is obtained through experiments or collection, including but not limited to peak stress, peak strain, cycle number, strain ratio / stress ratio.

[0051] The present invention uses methods such as PLS regression to compare the importance of peak stress, peak strain, cycle number, and strain ratio / stress ratio on fatigue life in different early cycles, for example, within 50 cycles, and uses the feature information of the most important cycle as the input for subsequent training, corresponding to the above-mentioned key feature information.

[0052] Step 2: Construct a training dataset, use the key feature information as the source domain dataset, and divide the target domain dataset into a training dataset and a test dataset.

[0053] For example, if the input data is X and the life cycle number is Y, use the above key feature information data {X, Y}(R = a) as the source domain dataset T source , plus a small amount of data T in the target domain target {X, Y}(R = b, b ≠ a) as the training dataset, and the remaining data in the target domain as the test dataset. Where R is the stress ratio.

[0054] Then perform normalization processing on the input data, that is, perform normalization processing on the input data and perform log10 processing on the output data.

[0055] Step 3: Establish a main model and an auxiliary model, and sequentially perform coupled cross-training on the main model and the auxiliary model using the source domain dataset and the target domain dataset.

[0056] Specifically, step 3 further includes:

[0057] 31. First, call the main model, input the source domain dataset and the first target domain training dataset for training. After training, input the target domain test data to obtain the first target domain prediction dataset.

[0058] 32. Then, combine the first target domain training dataset and the first target domain test dataset to form the second target domain training dataset.

[0059] 33. Input the second target domain training dataset and the source domain dataset into the auxiliary model for training. After training, input the target domain test data to obtain the second target domain prediction dataset.

[0060] 34. Then, combine the second target domain prediction dataset and the first target domain training dataset to form the third target domain training dataset.

[0061] 35. Call the main model, and input the source domain dataset and the third target domain training dataset into the main model for training. After training, input the target domain test data to obtain the third target prediction dataset.

[0062] 36. Calculate the difference between the second target domain prediction dataset and the third target domain prediction dataset, and compare the difference with the target difference. If the difference is less than the target difference, the calculation ends; if the difference is greater than the target difference, loop through steps 32 - 36.

[0063] For example, establish the main model A and the auxiliary model B. The main model A and the auxiliary model B are transfer learning models or physical models with weak learners such as support vector machines, decision trees, neural network models, etc.

[0064] The coupling algorithm of the model is as follows:

[0065] Call model A, input the source domain data {X s , Y s}, and the target domain training data {X t , Y t} for training;

[0066] After training, input X p , and obtain the prediction result as Y pA .

[0067] Combine the target domain training data and the target domain test data {X t , Y t}, {X p , Y pA} into a new target domain training dataset {X t' , Y t'}; call model B, input the source domain data and the new target domain training dataset {X t' , Yt' , perform training.

[0068] After training, input the test data X of the target domain p , and obtain the prediction result Y pB .

[0069] Combine the data sets {X t , Y t}, {X p , Y pB} to form a new target domain training data set {X t' , Y t'}, call model A, and input {X s , Y s}, {X t' , Y t'} for training.

[0070] After training, input X p , and obtain the prediction result Y pA .

[0071] Step 4: Calculate the difference between the prediction results of the main model and the auxiliary model, and output the minimum training model and the corresponding prediction result when the difference is less than the target difference.

[0072] That is, calculate the prediction error ε = ||Y pA - Y pB ||. If ε < ε 0 , the calculation ends; otherwise, repeat steps 32 - 36 to adjust the training parameters. Or repeat steps 31 - 36 to replace the main model A and adjust the training parameters.

[0073] After the training ends, output: the minimum training model A and the corresponding prediction result Y pA .

[0074] Then organize the newly input data, and obtain the material fatigue test data through experiments or collection, including peak stress, peak strain, number of cycles, strain ratio / stress ratio, as the input for fatigue life prediction.

[0075] Then organize the newly input data according to the format of the key feature information, and input the organized data into the minimum training model to obtain the fatigue life prediction result.

[0076] In summary, the transfer learning fatigue life calculation method of the present invention establishes a comprehensive model of a main model and an auxiliary model. This comprehensive model has high prediction accuracy, effectively prevents over - generalization, and is suitable for fatigue life calculation in the case of small samples. It uses cross - training of the main model and the auxiliary model, and aims at the minimum difference value between the two training predictions, and performs multiple iterative trainings, which is conducive to obtaining the optimal solution.

[0077] The present invention also provides a transfer learning fatigue life calculation system, which adopts the above transfer learning fatigue life calculation method. The calculation system includes:

[0078] An acquisition module, configured to acquire training data and determine key feature information corresponding to key cycle numbers;

[0079] A construction module, configured to construct a training data set, use the key feature information as a source domain data set, and divide the target domain data set into a training data set and a test data set;

[0080] A training module, configured to establish a main model and an auxiliary model, and sequentially perform coupled cross-training on the main model and the auxiliary model by using the source domain data set and the target domain data set;

[0081] An output module, configured to calculate the difference between the prediction results of the main model and the auxiliary model, and output the minimum training model and the corresponding prediction result when the difference is less than the target difference.

[0082] Referring to Figure 3 , the present invention also provides an electronic device 900, including: a processor 901 and a memory 902. The memory 902 stores a program or instruction that can run on the processor 901, and the program or instruction is executed by the processor 901 to perform the above transfer learning fatigue life calculation method. When the program or instruction is executed by the processor 901, it implements each process of the above implementation manner of the transfer learning fatigue life calculation method, and can achieve the same technical effect. To avoid repetition, it will not be described in detail here.

[0083] The present invention also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, it implements the transfer learning fatigue life calculation method as described above. When the program or instruction is executed by the processor, it implements each process of the above implementation manner of the transfer learning fatigue life calculation method, and can achieve the same technical effect. To avoid repetition, it will not be described in detail here.

[0084] For those skilled in the art, the above disclosure of the invention is only an example and does not constitute a limitation to this application. Although not explicitly stated here, those skilled in the art may make various modifications, improvements, and corrections to this application. Such modifications, improvements, and corrections are proposed in this application, so such modifications, improvements, and corrections still fall within the spirit and scope of the exemplary implementation manner of this application.

[0085] Meanwhile, this application uses specific terms to describe the embodiments of this application. For example, "one embodiment", "an embodiment", and / or "some embodiments" mean a certain feature, structure, or characteristic related to at least one embodiment of this application. Therefore, it should be emphasized and noted that the "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or more at different positions in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of this application can be appropriately combined.

[0086] Some aspects of this application can be executed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above-mentioned hardware or software can all be referred to as "data blocks", "modules", "engines", "units", "components", or "systems". The processor can be one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DAPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, or combinations thereof. In addition, aspects of this application may be embodied as a computer product located in one or more computer-readable media, which includes computer-readable program code. For example, the computer-readable media may include, but are not limited to, magnetic storage devices (such as hard disks, floppy disks, magnetic tapes...), optical discs (such as compact discs CD, digital versatile discs DVD...), smart cards, and flash memory devices (such as cards, sticks, key drives...).

[0087] The computer-readable media may contain a propagated data signal that contains computer program code, such as on a baseband or as part of a carrier wave. This propagated signal may have various forms of manifestation, including electromagnetic form, optical form, etc., or a suitable combination of forms. The computer-readable media can be any computer-readable media other than computer-readable storage media, which can be connected to an instruction execution system, apparatus, or device to implement communication, propagation, or transmission for use of the program. The program code located on the computer-readable media can be propagated through any suitable media, including radio, cable, fiber optic cable, radio frequency signal, or similar media, or any combination of the above media.

[0088] Similarly, it should be noted that, in order to simplify the description disclosed in this application and thus assist in the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of this application, sometimes multiple features are merged into one embodiment, drawing, or description thereof. However, this disclosure method does not mean that the features required by the subject matter of this application are more than those mentioned in the claims. In fact, the features of the embodiments are fewer than all the features of the single embodiment disclosed above. In some embodiments, numbers describing components and attribute quantities are used. It should be understood that such numbers used for the description of embodiments are, in some examples, modified by the modifiers "about", "approximately", or "substantially". Unless otherwise specified, "about", "approximately", or "substantially" indicate that the number allows a variation of ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, and this approximate value can be changed according to the characteristics required by individual embodiments. In some embodiments, the numerical parameters should consider the specified significant digits and adopt the method of retaining the general number of digits. Although the numerical ranges and parameters used in some embodiments of this application to confirm the breadth of their scope are approximate values, in specific embodiments, the setting of such numerical values is as precise as possible within the feasible range.

[0089] Although the present invention is disclosed above in its preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, all modifications, equivalent changes, and decorations made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for calculating fatigue life using transfer learning, characterized in that, it includes: Step 1: Obtain training data and determine the key feature information corresponding to the key cycle number; Step 2: Construct a training data set, use the key feature information as the source domain data set, and divide the target domain data set into a training data set and a test data set; Step 3: Establish a main model and an auxiliary model, and use the source domain data set and the target domain data set to perform coupled cross-training on the main model and the auxiliary model in sequence; Step 4: Calculate the difference between the prediction results of the main model and the auxiliary model, and output the minimum training model and the corresponding prediction result when the difference is less than the target difference.

2. The method for calculating fatigue life using transfer learning according to claim 1, characterized in that, Step 1 includes: Obtain material fatigue test data through experiments or collection, including but not limited to: peak stress, peak strain, cycle number, strain ratio / stress ratio.

3. The method for calculating fatigue life using transfer learning according to claim 2, characterized in that, Step 2 includes: Use the PLS regression method to compare the importance of the peak stress, the peak strain, the cycle number, and the strain ratio / stress ratio on the fatigue life at different early cycle numbers, and use the feature information of the most important cycle number as the key feature information.

4. The method for calculating fatigue life using transfer learning according to claim 1, characterized in that, Step 3 includes:

31. First, call the main model, input the source domain data set and the first target domain training data set for training. After training, input the target domain test data to obtain the first target domain prediction data set; 32. Then combine the first target domain training data set and the first target domain test data set to form a second target domain training data set; 33. Input the second target domain training data set and the source domain data set into the auxiliary model for training. After training, input the target domain test data to obtain the second target domain prediction data set; 34. Then combine the second target domain prediction data set and the first target domain training data set to form a third target domain training data set; 35. Call the main model, and input the source domain data set and the third target domain training data set into the main model for training. After training, input the target domain test data to obtain the third target prediction data set; 36. Calculate the difference between the second target domain prediction data set and the third target domain prediction data set and compare the difference with the target difference. If the difference is less than the target difference, the calculation ends; if the difference is greater than the target difference, loop through steps 32 - 36.

5. The method for calculating fatigue life using transfer learning according to claim 4, characterized in that, When the difference is less than the target difference, output the main model and the second target domain prediction data set or the third target domain prediction data set.

6. The method for calculating fatigue life using transfer learning according to claim 4, characterized in that, Step 36 further includes: If the difference is greater than the target difference, replace the main model and adjust the training parameters.

7. The transfer learning fatigue life calculation method according to claim 1, wherein, the main model and the auxiliary model are respectively a transfer learning model or a physical model of a weak learner among a support vector machine, a decision tree, and a neural network model.

8. A transfer learning fatigue life calculation system, wherein, the transfer learning fatigue life calculation system adopts the transfer learning fatigue life calculation method according to any one of claims 1-7, and the calculation system includes: an acquisition module, configured to acquire training data and determine key feature information corresponding to key cycle numbers; a construction module, configured to construct a training data set, use the key feature information as a source domain data set, and divide the target domain data set into a training data set and a test data set; a training module, configured to establish a main model and an auxiliary model, and sequentially perform coupled cross-training on the main model and the auxiliary model by using the source domain data set and the target domain data set; an output module, configured to calculate the difference between the prediction results of the main model and the auxiliary model, and output the minimum training model and the corresponding prediction result when the difference is less than the target difference.

9. An electronic device, wherein, it includes: a processor and a memory, the memory stores a program or instruction that can run on the processor, and the program or instruction is executed by the processor to implement the transfer learning fatigue life calculation method according to any one of claims 1-7.

10. A readable storage medium, wherein, the readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, it implements the transfer learning fatigue life calculation method according to any one of claims 1-7.

Images