Welding structure fatigue data enhancement method

Abstract

The invention relates to the technical field of welding data expansion, in particular to a welding structure fatigue data enhancement method, which comprises the following steps of: establishing a welding structure fatigue data set; and carrying out data expansion by using the improved CTGAN generation model, and carrying out normalization processing on the expanded data to obtain enhanced welding structure fatigue data. According to the method, the Peng model in the damage model is embedded into the loss function of the CTGAN generation model, the improved CTGAN generation model is obtained, data under the variable amplitude load can be subjected to accounting, fatigue data closer to the actual working condition can be obtained, the complexity of fatigue data collection in the prior art is reduced, and the fatigue data collection efficiency is improved. A large amount of fatigue data close to an actual working condition can be obtained through a small part of fatigue data, and the accuracy of a fatigue life prediction result is enhanced.

Description

Technical Field

[0001] The invention relates to the technical field of welding data expansion, and in particular to a method for enhancing fatigue data of a welded structure. Background Art

[0002] Machine learning has developed into an effective method for fatigue life prediction. However, due to the long cycle and high cost of variable amplitude loading fatigue tests, it is difficult to obtain a large number of reliable training samples, resulting in insufficient basic data samples for predicting fatigue life. Most of the fatigue data obtained by the model in the existing technology are fatigue data under constant amplitude load. However, in actual engineering, welded structures are mostly in a variable amplitude load environment, and fatigue data are complex and discrete. If the prediction of fatigue life is only based on fatigue data under constant amplitude load, the fatigue life prediction results will be inaccurate. Summary of the invention

[0003] In order to overcome the technical problems in the prior art that fatigue data sample size is low and the existing model has poor data expansion effect, the present invention provides a method for enhancing fatigue data of a welded structure.

[0004] The technical solution adopted by the present invention to achieve the above-mentioned purpose is: a method for enhancing fatigue data of a welded structure, characterized in that it comprises the following steps:

[0005] S1: Establish a fatigue data set for welded structures;

[0006] S2: Based on the constructed welded structure fatigue dataset, the improved CTGAN generation model is used for data augmentation;

[0007] S3: normalize the expanded data;

[0008] S4: Obtain fatigue data of the enhanced welded structure.

[0009] Preferably, the establishment of the welding structure fatigue data set in step S1 specifically includes: obtaining fatigue test data of welding machine fatigue analysis institutions and enterprises, conducting laboratory fatigue tests, and obtaining welding structure test data under variable amplitude loads. The welding structure fatigue data set includes the first-level stress level σ1 and the second-level stress level σ2, and the fatigue life N under the first-level stress level f1 and fatigue life N at the second stress level f2 , the number of cycles under the first stress level n 1p and the number of cycles n at the second stress level 2p .

[0010] Preferably, in step S2, using the improved CTGAN generation model to perform data expansion includes the following steps:

[0011] S201: The Peng model in the cumulative damage model is embedded into the loss function of the CTGAN generation model. After embedding, the total loss function expression of the CTGAN generation model is:

[0012] L generator =L adversarial +L discrete +L mmd +L physics

[0013] Among them, L generator is the total loss function, L adversarial To combat the loss function, L discrete is the discrete column loss function, L mmd is the MMD loss function, L physics is the physical loss function;

[0014] The expression of the MMD loss function is:

[0015]

[0016] Among them, L mmd is the MMD loss function, P is the probability distribution of the data generated by the generator, Q is the probability distribution of the real data, N and M are the number of samples sampled from P and Q respectively, y i and j are samples sampled from P and Q respectively, i and j are the serial numbers of the samples, and φ(*) is the feature mapping function;

[0017] The expression of the physical loss function is:

[0018]

[0019] Among them, L physics is the physical loss function, σ1 is the first-level stress level, σ2 is the second-level stress level, N f1 is the fatigue life under the first stress level, N f2 is the fatigue life under the second stress level, n1 is the number of cycles under the first stress level, and n2 is the number of cycles under the second stress level;

[0020] S202: Generate σ using the improved CTGAN generative model 1m , σ 2m 、n 1m 、n 2m ;

[0021] S203: Filter out fatigue data with the same stress level as the welded structure fatigue data set from the data output by the improved CTGAN generation model, i.e., σ 1m =σ1,σ2m = σ2, and improve the CTGAN generation model to generate σ 1m , σ 2m 、n 1m 、n 2m Insert the corresponding fatigue life N under the first stress level into the data f1 , fatigue life at the second stress level N f2 , get the complete expanded data σ1, σ2, N f1 、N f2 、n 1m 、n 2m .

[0022] Preferably, the expression for the normalization process in step S3 is:

[0023]

[0024] Among them, x new is the normalized data, x is the data before normalization, and x max is the maximum value in the data, x min is the minimum value in the data.

[0025] Compared with the prior art, the present invention embeds the Peng model in the damage model into the loss function of the CTGAN generation model to obtain an improved CTGAN generation model, which can account for data under variable amplitude loads and obtain fatigue data that is closer to the actual working conditions. It reduces the complexity of collecting fatigue data in the prior art, so that a large amount of fatigue data close to the actual working conditions can be obtained through a small part of fatigue data, thereby enhancing the accuracy of fatigue life prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 The present invention is a flow chart of a method for enhancing fatigue data of a welded structure. DETAILED DESCRIPTION

[0027] A specific implementation method of a welding structure fatigue data enhancement method of the present invention is as follows Figure 1 As shown, the following steps are included:

[0028] S1: Establish a fatigue data set for welded structures. The specific steps of establishing a fatigue data set for welded structures are: obtain fatigue test data from welding machine fatigue analysis institutions and enterprises, conduct laboratory fatigue tests, and obtain test data for welded structures under variable amplitude loads. The fatigue data set for welded structures includes the first-level stress level σ1 and the second-level stress level σ2. The fatigue life N under the first-level stress level f1 and fatigue life N at the second stress level f2 , the number of cycles under the first stress level n1p and the number of cycles n at the second stress level 2p ;

[0029] S2: Based on the constructed welded structure fatigue dataset, the improved CTGAN generation model is used for data expansion. The improved CTGAN generation model is used for data expansion, which includes the following steps:

[0030] S201: The Peng model in the cumulative damage model is embedded into the loss function of the CTGAN generation model. After embedding, the total loss function expression of the CTGAN generation model is:

[0031] L generator =L adversarial +L discrete +L mmd +L physics

[0032] Among them, L generator is the total loss function, L adversarial To combat the loss function, L discrete is the discrete column loss function, L mmd is the MMD loss function, L physics is the physical loss function;

[0033] The expression of the MMD loss function is:

[0034]

[0035] Among them, L mmd is the MMD loss function, P is the probability distribution of the data generated by the generator, Q is the probability distribution of the real data, N and M are the number of samples sampled from P and Q respectively, y i and j are samples sampled from P and Q respectively, i and j are the serial numbers of the samples, and φ(*) is the feature mapping function;

[0036] The expression of the physical loss function is:

[0037]

[0038] Among them, L physics is the physical loss function, σ1 is the first-level stress level, σ2 is the second-level stress level, N f1 is the fatigue life under the first stress level, N f2 is the fatigue life under the second stress level, n1 is the number of cycles under the first stress level, and n2 is the number of cycles under the second stress level;

[0039] S202: Generate σ using the improved CTGAN generative model1m , σ 2m 、n 1m 、n 2m ;

[0040] S203: Filter out fatigue data with the same stress level as the welded structure fatigue data set from the data output by the improved CTGAN generation model, i.e., σ 1m =σ1,σ 2m = σ2, and generate σ in the improved CTGAN generation model 1m , σ 2m 、n 1m 、n 2m Insert the corresponding fatigue life N under the first stress level into the data f1 , fatigue life at the second stress level N f2 , get the complete expanded data σ1, σ2, N f1 、N f2 、n 1m 、n 2m ;

[0041] S3: Normalize the expanded data. The normalization expression is:

[0042]

[0043] Among them, x new is the normalized data, x is the data before normalization, and x max is the maximum value in the data, x min is the minimum value in the data;

[0044] S4: Obtain fatigue data of the enhanced welded structure.

[0045] The present invention is described by way of embodiments, and those skilled in the art will appreciate that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application belong to the protection scope of the present invention.

Claims

1. A method for enhancing fatigue data of a welded structure, characterized in that: The following steps are involved: S1: Establish a fatigue data set for welded structures; S2: Based on the constructed welded structure fatigue dataset, the improved CTGAN generation model is used for data augmentation; S3: normalize the expanded data; S4: Obtain fatigue data of the enhanced welded structure.

2. A method for enhancing fatigue data of a welded structure according to claim 1, characterized in that: The establishment of the welding structure fatigue data set in step S1 is specifically as follows: obtaining fatigue test data of welding machine fatigue analysis institutions and enterprises, conducting laboratory fatigue tests, and obtaining welding structure test data under variable amplitude loads. The welding structure fatigue data set includes the first-level stress level σ1 and the second-level stress level σ2, and the fatigue life N under the first-level stress level f1 and fatigue life N at the second stress level f2 , the number of cycles under the first stress level n 1p and the number of cycles n at the second stress level 2p .

3. A method for enhancing fatigue data of a welded structure according to claim 1, characterized in that: In step S2, data augmentation using the improved CTGAN generation model includes the following steps: S201: The Peng model in the cumulative damage model is embedded into the loss function of the CTGAN generation model. After embedding, the total loss function expression of the CTGAN generation model is: L generator =L adversarial +L discrete +L mmd +L physics Among them, L generator is the total loss function, L adversarial To combat the loss function, L discrete is the discrete column loss function, L mmd is the MMD loss function, L physics is the physical loss function; The expression of the MMD loss function is: Among them, L mmd is the MMD loss function, P is the probability distribution of the data generated by the generator, Q is the probability distribution of the real data, N and M are the number of samples sampled from P and Q respectively, y i and j are samples sampled from P and Q respectively, i and j are the serial numbers of the samples, and φ(*) is the feature mapping function; The expression of the physical loss function is: Among them, L physics is the physical loss function, σ1 is the first-level stress level, σ2 is the second-level stress level, N f1 is the fatigue life under the first stress level, N f2 is the fatigue life under the second stress level, n1 is the number of cycles under the first stress level, and n2 is the number of cycles under the second stress level; S202: Generate σ using the improved CTGAN generative model 1m , σ 2m 、n 1m 、n 2m ; S203: Filter out fatigue data with the same stress level as the welded structure fatigue data set from the data output by the improved CTGAN generation model, i.e., σ 1m =σ1,σ 2m = σ2, and generate σ in the improved CTGAN generation model 1m , σ 2m 、n 1m 、n 2m Insert the corresponding fatigue life N under the first stress level into the data f1 , fatigue life at the second stress level N f2 , get the complete expanded data σ1, σ2, N f1 、N f2 、n 1m 、n 2m .

4. A method for enhancing fatigue data of a welded structure according to claim 1, characterized in that: The expression of the normalization process described in step S3 is: Among them, x new is the normalized data, x is the data before normalization, and x max is the maximum value in the data, x min is the minimum value in the data.

Images