Method, system and equipment for predicting service life of high-speed train structure and storage medium

Abstract

The invention relates to the technical field of train safety monitoring, in particular to a high-speed train structure life prediction method, system and device and a storage medium. The prediction method comprises the following steps: acquiring Lamb wave response data; inputting the Lamb wave response data into the damage diagnosis model to obtain the current crack length; and inputting the current crack length into a pre-constructed service life prediction model of the high-speed train structure, predicting the residual service life of the high-speed train structure, and outputting the residual service life. The prediction accuracy and credibility are improved, the interpretability of the model is enhanced, the maintenance cost is reduced, the driving safety is improved, the service life is prolonged, the adaptability to environmental changes is high, a large amount of monitoring data can be rapidly processed, real-time service life prediction information is provided for train operators, and the method is suitable for popularization and application. And the structural integrity and passenger safety of the high-speed train are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of train safety monitoring, and in particular to a high-speed train structure life prediction method, system, device and storage medium. BACKGROUND

[0002] The rapid development of high-speed train technology puts higher requirements on the performance of train structure materials. Due to the influence of fatigue stress, fatigue cracks are easily generated in the structure materials during use, which not only shortens the service life of the materials and increases the maintenance cost, but also may cause safety hazards. Therefore, carrying out monitoring of material damage and accurate prediction of remaining life is of great importance to ensure the safe operation of key mechanical structures such as high-speed trains.

[0003] Life prediction is a key link in structural health monitoring (SHM), which aims to predict the remaining effective use time of equipment by analyzing real-time monitoring data. Main methods include prediction methods based on fracture mechanics, such as evaluating crack propagation using Paris law; S-N curve method based on stress / strain; statistical model-based methods, such as gamma process; and artificial intelligence-based methods, including machine learning and deep learning techniques. These methods are suitable for different scenarios and conditions, but all aim to improve the accuracy and reliability of prediction. Among them, deep learning methods have advantages in automatic feature extraction, but they are often regarded as "black boxes", and their results lack interpretability.

[0004] Therefore, it is of great practical significance and application value to propose a high-speed train structure life prediction method based on physical information neural network. SUMMARY

[0005] The purpose of the present application is to provide a high-speed train structure life prediction method, system, device and storage medium, which solves the technical problems existing in the existing train structure life prediction method.

[0006] The present application is realized by the following technical solutions:

[0007] The present application discloses a high-speed train structure life prediction method, comprising the following steps:

[0008] S1, acquiring Lamb wave response data;

[0009] S2, inputting the Lamb wave response data into a damage diagnosis model to obtain the current crack length;

[0010] S3, inputting the current crack length into a pre-constructed high-speed train structure life prediction model to predict the remaining life of the high-speed train structure, and outputting the remaining service life.

[0011] Further, in S1, the Lamb wave response data is acquired by a high-speed train structure health monitoring system.

[0012] Further, in S3, the construction process of the pre-constructed high-speed train structure life prediction model is as follows:

[0013] According to the preset size, a cyclic load is applied on the target structure, the generated crack size is gradually measured, and the Lamb wave response data corresponding to the crack size is acquired at the same time;

[0014] A normalized energy damage factor is proposed based on the Lamb wave response data, and a damage diagnosis model is constructed based on the normalized energy damage factor;

[0015] Based on the damage diagnosis model and the Paris formula, an empirical degradation model is established;

[0016] The empirical degradation model is integrated into a physical information neural network model to obtain a fusion model, the fusion model is trained and optimized to obtain the pre-constructed high-speed train structure life prediction model.

[0017] Further, the expression of the damage diagnosis model is as follows:

[0018] a=w·DI 3 +p·DI 2 +q·DI+d

[0019] Wherein, a is the crack length, DI is the normalized energy damage factor, w, p, q, d are all fitting parameters.

[0020] Further, the expression of the empirical degradation model is as follows:

[0021]

[0022] Wherein, a is the crack length, N is the number of fatigue cycles, α1, β1, c1 are all fitting parameters.

[0023] Further, the loss function of the fusion model is composed of the loss of actual samples and the loss of empirical physical samples, and the expression is as follows:

[0024] Loss total =αLoss real +(1-α)Loss phy

[0025] Wherein, Loss real is the loss of actual samples, Loss phy is the loss of empirical physical samples, and α is a weight coefficient.

[0026] The loss Loss phyThe prediction loss and the degradation feature loss are combined, and the expression is:

[0027] Loss phy =βLoss phy,model +(1-β)Loss phy,character

[0028]

[0029]

[0030] Wherein, beta is used to adjust the proportion between the prediction loss and the degradation feature loss, Loss phy,model is the prediction loss, Loss phy,character is the degradation feature loss;

[0031] Delta x phy is the label of the physical knowledge model sample, is the prediction value of the fusion model.

[0032] Further, the current crack length is input into the high-speed train structure life prediction model constructed in advance, the remaining life of the high-speed train structure is predicted, and the remaining service life is output, and the specific process is:

[0033] The current characteristic crack length is input into the high-speed train structure life prediction model constructed in advance, the prediction value of the degradation rate is obtained, and then the current characteristic crack length and the prediction value of the degradation rate are added to obtain the characteristic crack length at the next moment;

[0034] The characteristic crack length at the next moment is compared with the crack propagation size threshold value, if the characteristic crack length at the next moment is less than the crack propagation size threshold value, the next iteration calculation is continued;

[0035] Until the characteristic crack length at the next moment reaches the crack propagation size threshold value, the iteration number corresponding to the remaining service life.

[0036] The application discloses a high-speed train structure life prediction system, comprising:

[0037] A data acquisition module is used for acquiring Lamb wave response data;

[0038] A damage diagnosis module is used for inputting the Lamb wave response data into a damage diagnosis model to obtain a current crack length;

[0039] A life prediction module is used for inputting the current crack length into a high-speed train structure life prediction model constructed in advance to predict the remaining life of the high-speed train structure and output the remaining service life.

[0040] The application further discloses a computer device, including a memory, a processor and a computer program stored in the memory and running on the processor, and the processor implements the steps of the high-speed train structure life prediction method when the computer program is executed.

[0041] The application further discloses a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the high-speed train structure life prediction method when executed by a processor.

[0042] Compared with the prior art, the application has the following beneficial technical effects:

[0043] The application provides a high-speed train structure life prediction method and system based on a physical information neural network, which extracts a normalized energy damage factor from a Lamb wave signal, inputs the damage diagnosis model to calculate a crack length, inputs the calculated crack length into a high-speed train structure life prediction model constructed in advance to obtain a predicted value of a degradation rate, then adds and calculates the current characteristic crack length and the predicted value of the degradation rate to obtain a characteristic crack length at the next moment. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 It is a flow framework diagram of the high-speed train structure life prediction method based on the physical information neural network of the application.

[0045] Figure 2 It is an experimental aluminum plate 6061 and an experimental overall arrangement of the application.

[0046] Figure 3 It is an experimental overall arrangement diagram of the application.

[0047] Figure 4 It is a Lamb wave signal schematic diagram of the application.

[0048] Figure 5 It is a normalized energy damage factor diagram provided by the application.

[0049] Figure 6 It is a crack propagation observation model result diagram provided by the application.

[0050] Figure 7The training effect of a high-speed train structure life prediction method based on a physical information neural network;

[0051] Figure 8 The performance of a high-speed train structure life prediction method based on a physical information neural network in life prediction;

[0052] Fig. (a) is an iterative prediction curve of the high-speed train structure life prediction model with a starting point of 10009 cycles;

[0053] Fig. (b) is an iterative prediction curve of the physical model with a starting point of 10009 cycles;

[0054] Fig. (c) is an iterative prediction curve of the high-speed train structure life prediction model with a starting point of 22510 cycles;

[0055] Fig. (d) is an iterative prediction curve of the physical model with a starting point of 22510 cycles;

[0056] Fig. (e) is an iterative prediction curve of the high-speed train structure life prediction model with a starting point of 30020 cycles;

[0057] Fig. (f) is an iterative prediction curve of the physical model with a starting point of 30020 cycles;

[0058] Fig. (g) is an iterative prediction curve of the high-speed train structure life prediction model with a starting point of 38102 cycles;

[0059] Fig. (h) is an iterative prediction curve of the physical model with a starting point of 38102 cycles. DETAILED DESCRIPTION

[0060] In order to make the objectives, technical solutions and advantages of the present application clearer, further detailed description will be made in combination with the accompanying drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application, and are not used to limit the present application, that is, the described examples are only a part of the examples of the present application, but not all examples.

[0061] The components described and shown in the accompanying drawings and examples of the present application can be arranged and designed in various different configurations, therefore, the detailed description of the examples of the present application provided in the following accompanying drawings is not intended to limit the scope of the claimed present application, but only to represent a selected embodiment of the present application. Based on the accompanying drawings and examples of the present application, all other examples obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0062] It is to be understood that the terms "comprising," "including," and other correlative terms are intended to be open-ended, and specifically mean "comprising, but not limited to."

[0063] The features and performances of the present application are further described in detail below in combination with embodiments.

[0064] It should be noted that due to the complex operating environment of the train, factors such as fatigue load and harsh working environment can cause micro-damage to the structure and threaten the safety of train operation, so it is necessary to predict the service life of each structure of the high-speed train. The technology lacks explainability in the field of structure life prediction, and cannot accurately monitor and predict the fatigue crack propagation in the structure of the high-speed train. Based on this, the present application can use physical information neural network to fuse physical information, make up for the defect of unexplainable deep learning, and further predict the remaining service life of the high-speed train structure, providing a basis for maintenance support.

[0065] Embodiment 1

[0066] Please refer to Figure 1 From the design point of view, the construction process of the high-speed train structure life prediction model is described, which specifically includes the following processes:

[0067] N sensors and a processing center are arranged at the target structure of the train, N being a positive integer;

[0068] The processing center is used to:

[0069] A cyclic load is applied to the target structure, and the crack size generated is measured step by step;

[0070] The sensor is used to obtain Lamb wave response data of the high-speed train structure health monitoring system;

[0071] A normalized energy damage factor is proposed based on the Lamb wave response data, a damage diagnosis model is constructed, and a mapping relationship between damage and Lamb wave signal is established;

[0072] In combination with the established damage diagnosis model and the Paris formula, the Paris formula is simplified and deduced, and an empirical degradation model is established;

[0073] An empirical physical loss and an actual sample loss function are constructed, the empirical degradation model is integrated into a physical information neural network (PINN), a fusion model is obtained, the fusion model is trained and optimized, and a constructed high-speed train structure life prediction model is obtained.

[0074] The remaining life of the high-speed train structure is predicted by using the built high-speed train structure life prediction model, and the remaining service life of the structure is output, thereby providing a basis for the use of the structure.

[0075] Lamb wave generation principle: when a certain frequency of alternating load is applied to the thin plate, the particles in the thin plate will oscillate around their equilibrium position, and this vibration will propagate in the form of waves in the thin plate, thereby forming Lamb waves. In essence, Lamb waves are formed by the mutual interference and superposition of longitudinal waves and transverse waves between the two surfaces of the thin plate. In the thin plate, the longitudinal wave makes the particles vibrate along the propagation direction of the wave, and the transverse wave makes the particles vibrate perpendicular to the propagation direction of the wave. The combination of the two waves makes the Lamb wave have unique propagation characteristics.

[0076] In the embodiment, the target structure refers to the object to be measured.

[0077] Embodiment 2

[0078] On the basis of embodiment 1, the establishment process of the damage diagnosis model is introduced, which specifically includes the following steps:

[0079] The normalized energy damage factor reflects the change rule of the energy of the damage scattering signal with the crack size, wherein the damage scattering signal is obtained by comparing the change of the Lamb wave signals before and after damage. The calculation formula of the normalized energy damage factor DI is as follows:

[0080]

[0081] Wherein, u0(t) is the undamaged signal, u i (t) is the damage scattering signal, t1 and t2 are the start time and end time of the time window respectively.

[0082] The normalized energy damage factor DI is used to analyze the quantitative characterization of the fatigue crack damage information of the Lamb wave signal. A polynomial is defined for fitting, thereby constructing a damage diagnosis model, and the expression is as follows:

[0083] a=w·DI 3 +p·DI 2 +q·DI+d

[0084] Wherein, a is the crack length, DI is the normalized energy damage factor, w, p, q, d are all fitting parameters.

[0085] Embodiment 3

[0086] On the basis of embodiment 1, an empirical degradation model is established by combining the damage diagnosis model with the Paris formula.

[0087] Paris formula is an important formula in the field of fatigue crack propagation, which is used to describe the relationship between crack propagation rate and stress intensity factor. Paris formula is:

[0088]

[0089] wherein: a is the crack length; da represents the crack propagation rate, i.e. the increment of crack length per unit stress cycle; C and m are material-related constants, which need to be determined through experiments. is a parameter related to the material of the test piece, which can be obtained through experiments;

[0090] Let Therefore:

[0091]

[0092] By variable separation of this simplified formula, we can get:

[0093] a -β da=αdN

[0094] Then, according to the integration rule, integrate both sides of the equation to get:

[0095]

[0096] wherein c is a constant term. Through exponential transformation operation, the relationship between crack length and cycle number is obtained:

[0097]

[0098] In order to facilitate subsequent calculation and fitting, the formula is simplified, only the highest power of N and the constant term are retained, and the empirical degradation model is obtained, the expression is:

[0099]

[0100] wherein a is the crack length, N is the fatigue cycle number, and α1, β1, c1 are fitting parameters.

[0101] After obtaining the simplified physical expression, a part of the actual observation data is taken out to fit the formula, and the parameters of the empirical degradation model are obtained by using the nonlinear least squares method.

[0102] The empirical degradation model can approximately reflect the degradation process in the fatigue crack propagation process of structural materials, and it can be applied to the training process of the fusion model as an empirical physical law, and the interpretability of the fusion model can be given.

[0103] Example 4

[0104] Based on the embodiment 1, the process of constructing the experience physical loss and actual sample loss function, integrating the experience degradation model into the physical information neural network model (PINN), obtaining the fusion model, training and optimizing the fusion model, and obtaining the high-speed train structure life prediction model is introduced.

[0105] The adopted physical information neural network model is composed of full connection layer and physical information layer, wherein the output of the network is predicted through three linear full connection layers. In the process of network training, the total loss of the network is composed of two parts, namely the loss of actual sample and the loss of experience physical sample, and the relationship is as follows:

[0106] Loss total =αLoss real +(1-α)Loss phy

[0107] Wherein, Loss real is the loss of actual sample, Loss phy is the loss of experience physical sample, and α is the weight coefficient.

[0108] In the loss formula, the weight of the two component losses can be adjusted by α. The loss of experience physical sample Loss phy is also composed of prediction loss and degradation feature loss. It is not difficult to understand that in the process of fatigue crack propagation, the structure state will only get worse with the growth of fatigue crack, and without external artificial intervention, the structure state will only develop towards the result of fracture, so it can be known that its degradation rate is always positive from the appearance of fatigue crack to fracture. The loss function expression based on physical knowledge is:

[0109] Loss phy =βLoss phy,model +(1-β)Loss phy,character

[0110]

[0111]

[0112] Wherein, β is also used to adjust the proportion of the two component losses. Loss phy,model is the prediction loss, and Loss phy,character is the degradation feature loss.

[0113] The calculation value of Loss phy,model is the square of the difference between the label Δx phy of the physical knowledge model sample and the predicted value of the high-speed train structure life prediction model. Loss phy,characterThe role of the network is to punish the parameters of the network when the degradation rate of the network output is negative.

[0114] During the training of the fusion model, the crack length feature x of the training set is input into the fusion model, and then the output corresponding to x is calculated After obtaining the output, the physical knowledge model loss is calculated using it, and then the prediction loss Loss of the fusion model is calculated phy,model and degradation feature loss Loss phy,character . Then the total loss of network training is obtained by adding the calculated physical knowledge model loss to the actual sample prediction loss. Finally, the model parameter learning is performed using the back propagation algorithm. At this point, the physical knowledge has been integrated into the neural network for constraint, which makes the network model have internal interpretability. When the network needs to be optimized, the process can be completed by adjusting the physical information equation inside the network.

[0115] Finally, the constructed high-speed train structure life prediction model is used to predict the remaining life of the high-speed train structure. The feature crack length corresponding to the current cycle number is input into the trained network to obtain the predicted value of the current degradation rate. Then the feature crack length corresponding to the current cycle number is added to the predicted degradation rate obtained by the network to obtain the feature corresponding to the next cycle.

[0116]

[0117] where a i is the feature crack length corresponding to the current cycle number, a i+1 is the feature crack length corresponding to the next cycle number, is the predicted value of the degradation rate.

[0118] By continuously iterating, the feature change trend before the test piece breaks can be predicted. When the feature crack length corresponding to the next cycle number reaches the crack propagation size threshold of the test piece, the corresponding iteration number is the remaining service life.

[0119] Wherein, the physical test verification method is as follows:

[0120] Al6061 aluminum alloy has been widely used in the field of manufacturing high-speed train accessories, so Al6061 aluminum alloy as shown in Figure 2 is selected as the fatigue tensile test piece. A total of 6 test materials are prepared, denoted as T1-T6 aluminum plate physical objects. The test aluminum plate size is 240mm x 45mm x 4mm. A pre-crack with a length of about 2mm is cut on one side of the center position of the aluminum plate, which can ensure the starting position and propagation direction of the crack. The specific structure design and sensor arrangement position are the same as those inFigure 2 The same.

[0121] Table 1: Mechanical parameters of aluminum plate 6061

[0122]

[0123] The overall experimental arrangement is shown in Figure 3 , which consists of a fatigue testing machine (MTS Landmark), a structural health monitoring scanning system, and a microscope. In the test, the loading frequency is set to 10 Hz, the maximum loading fatigue load is set to 32 kN of sinusoidal tensile fatigue load, and the fatigue stress is set in the range of 8 Mpa-80 Mpa. Due to the size limitation, in order to weaken the mutual influence between different modes of Lamb waves, the signal frequency is close to the center frequency, and the frequency of the Lamb wave signal is set to 140 kHz. The monitoring signal length of the structural health monitoring scanning system is recorded as 3000, and the sampling frequency is set to 10 MHz. After the tensile fracture test starts, according to the set fatigue loading times, every time the set value is loaded, the tensile process is paused, and the crack length information at this time is measured and recorded through the microscope and the computer connected to the hydraulic fatigue testing machine. Repeat this process until the set stress maximum loading times or the specimen is broken and record the corresponding stress loading times. A total of six specimens, the same experimental operation is carried out, and the stress loading times before the specimen is broken and the corresponding crack length and the Lamb wave signal in the whole process are recorded. Figure 4 The amplitude change of the Lamb wave with the change of the sampling point when passing through different fatigue crack lengths is recorded, and it can be seen that the amplitude and phase of the signal at the same sampling point between different cracks change regularly. The normalized energy damage factor corresponding to the six experimental specimens is calculated, the Lamb wave signal is analyzed, and the results are shown in Figure 5 . In this invention, a polynomial is defined for fitting:

[0124] a = w·DI 3 + p·DI 2 + q·DI + d

[0125] This formula is a cubic polynomial, where a is the crack length, DI is the normalized energy damage factor, and w, p, q, and d are fitting parameters. The feasibility of the above method is studied by the damage factors of the six specimens. Take specimen T6 as the test piece, and use the damage indices of specimens T1, T2, T3, T4, and T5 to perform polynomial fitting. The parameters of the fitted expression are shown in Table 2:

[0126] Table 2: Polynomial parameter value table

[0127]

[0128] Then the damage factor distribution of T6 is drawn by combining the fitting curve of the damage factor of T1-T5, that is Figure 6 It can be seen that the damage factor of T6 is distributed around the fitting curve, and its change trend with the increase of crack length is consistent with the fitting curve of the damage factor shown in the figure. The above results show that the normalized energy damage factor of the specimen can be obtained from the Lamb signal data to quantitatively characterize the propagation process of the fatigue crack of the structure material, and it is feasible to establish the observation equation. Figure 6

[0129] By using the observation data of the fatigue crack length a and the cycle number N of the fatigue crack propagation stage of the specimens T1-T5, the a-N relationship equation, i.e. formula (1) can be obtained by empirical physical model fitting:

[0130] a = 8.02322 x 10 -10 N + 0.5033

[0131] On this basis, the final empirical degradation equation can be obtained by using the derivation rule:

[0132]

[0133] The above formula can well represent the propagation process of the structure fatigue crack with the increase of the stress cycle number. Through the derivation formula, the physical meaning represented by it is the propagation rate of the fatigue crack length with the increase of the stress cycle number, that is, the degradation rate of the structure material, so theoretically, the formula can also meet the propagation trend of the fatigue crack length in the degradation stage to a certain extent, and can well characterize the propagation process of the fatigue crack. In addition, it quantitatively describes the degradation process of the training specimen, successfully provides an explanation based on accurate physical knowledge for the specimens with similar degradation process, and due to the influence of noise and environmental factors, there are certain differences in the degradation process, so the specimens with similar degradation rules mostly meet the physical equation in the degradation process. Therefore, the physical equation can be used as priori physical knowledge to be integrated into the neural network for the next stage of fatigue life prediction process.

[0134] The crack length information in the training data set is taken as the input x of the network, and the approximate degradation rate obtained by piecewise linear fitting is taken as the real training sample (x, x real ) of the network output. Then the crack length input is fused into the model to obtain the degradation rate, which is taken as the physical model training sample (x, Δx phy ) of the fusion model output. In order to obtain a relatively smooth degradation rate curve and a more accurate prediction value, the loss weights of the network should be reasonably distributed to obtain smaller errors when assigning the loss weights of the real sample and the physical model sample.

[0135] ​In the process of training the fusion model, the real data samples and the physical knowledge samples obtained by the a-N relationship equation are input into the fusion model at the same time, the real sample loss and the physical model loss are calculated by using the output of the fusion model, and then the network total loss is obtained, and then the network is trained to obtain the ideal high-speed train structure life prediction model. The final hyperparameters in the training process are as shown in Table 3.

[0136] Table 3: Fusion model training parameters

[0137]

[0138] After assigning the optimal weight, the prediction data of the degradation rate of the degradation process of the training sample is obtained by the fusion model, and the real degradation rate obtained by the segmented linear fitting and the prediction result of the degradation rate of the training sample by the physical model are recorded. As shown in FIG. 6, it can be seen that the smoothness of the degradation rate curve is improved, and the change trend of the degradation rate predicted by the network model is similar to that of the physical model in the early stage of degradation, but in the later stage of degradation, the prediction of the degradation rate of the network model starts to rise, which is similar to the change characteristics shown by the observation data, so the whole network follows the physical change law and covers the degradation rate change information shown by the real data samples. Figure 7

[0139] In addition, the fatigue crack length prediction can be obtained by using the degradation rate of the first crack length accumulation after all cycle times of single-step prediction. The specific method is to predict the degradation rate by using the current fatigue crack length, and then obtain the fatigue crack length corresponding to the next cycle, and then the change trend of the fatigue crack can be obtained by cycle iteration. Figure 8 The prediction results of the crack length by the high-speed train structure life prediction model and the physical model constructed by the present application are shown when the starting prediction point is at 10009, 22510, 30020 and 38102 stress cycles. The prediction results of the structure material residual life and the corresponding crack error of the high-speed train structure life prediction model and the physical model constructed by the present application are calculated, as shown in Table 4.

[0140] Table 4: Comparison of high-speed train structure life prediction model and physical model prediction

[0141]

[0142] ​It can be seen from the combination of the chart analysis that, when only considering the starting prediction point, whether it is the high-speed train structure life prediction model or the physical model, when predicting the fatigue crack propagation, the error of the starting prediction point is smaller at 10009 and 30020 cycles, which can also be seen from the comparison prediction chart. When predicting the remaining life, the starting prediction point at 22510 cycles is the most optimal, and the prediction effect in the early stage is more consistent with the true data. When the specimen is close to the end of life, although the trend is the same, it deviates from the true data, and a certain error is generated in the prediction of the final remaining life.

[0143] Based on the same starting prediction point, it can be seen from the image that the prediction result of the high-speed train structure life prediction model is closer to the real crack propagation process. It can also be seen from the data comparison in Table 4 that, whether in the fatigue crack propagation prediction or the remaining life prediction of the structural material, the prediction result error of the high-speed train structure life prediction model is smaller, and it can better reflect the real situation. The reason is that the parameters of the physical model are relatively fixed, and are obtained from a set of observation training data, and the change of the prediction result is relatively single. The high-speed train structure life prediction model not only integrates the physical model as a constraint, but also automatically extracts relevant features from the observation data. In combination with this point, the prediction effect of the high-speed train structure life prediction model is relatively better.

[0144] In summary, the high-speed train structure life prediction model and the physical model can complete the prediction of the fatigue crack propagation process and the remaining life prediction task of the structural material to a certain extent, but overall, the prediction effect of the high-speed train structure life prediction model is better than that of the physical model, and the prediction accuracy is higher.

[0145] The high-speed train structure life prediction system based on the physical information neural network described above can realize each embodiment of the high-speed train structure life prediction method based on the physical information neural network described above and achieve the same beneficial effects, which will not be repeated here.

[0146] The embodiment of the present application also provides a high-speed train structure life prediction system based on a physical information neural network, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above method when executing the computer program. The damage detection system described above can realize each embodiment of the high-speed train structure life prediction method based on the physical information neural network described above and achieve the same beneficial effects, which will not be repeated here.

[0147] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize the method steps described above. The readable storage medium can realize each embodiment of the high-speed train structure life prediction method based on the physical information neural network described above, and can achieve the same beneficial effects, which will not be described here.

[0148] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application rather than limit them, although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that: the specific embodiments of the present application can be modified or replaced by the equivalent, without departing from the spirit and scope of the present application, any modification or equivalent replacement, which should be covered within the protection scope of the present application.

Claims

1. A method for predicting the structural life of high-speed trains, characterized in that, Includes the following steps: S1. Obtain Lamb wave response data; S2. Input the Lamb wave response data into the damage diagnosis model to obtain the current crack length; S3. Input the current crack length into the pre-built high-speed train structure life prediction model to predict the remaining life of the high-speed train structure and output the remaining service life.

2. The method for predicting the structural life of a high-speed train according to claim 1, characterized in that, In S1, the Lamb wave response data is acquired through the high-speed train structural health monitoring system.

3. The method for predicting the structural life of a high-speed train according to claim 1, characterized in that, In S3, the construction process of the pre-built high-speed train structural life prediction model is as follows: Cyclic loads are applied to the target structure according to the preset dimensions, and the resulting crack size is measured step by step. At the same time, Lamb wave response data corresponding to the crack size are obtained. A normalized energy damage factor is proposed based on Lamb wave response data, and a damage diagnosis model is constructed based on the normalized energy damage factor. An empirical degradation model is established based on the damage diagnosis model and the Paris formula. By integrating the empirical degradation model into the physical information neural network model, a fusion model is obtained. The fusion model is then trained and optimized to obtain a pre-constructed high-speed train structural life prediction model.

4. The method for predicting the structural life of a high-speed train according to claim 3, characterized in that, The expression for the injury diagnosis model is: a=w·DI 3 +p·IN 2 +q·DI+d Where a is the crack length, DI is the normalized energy damage factor, and w, p, q, and d are all fitting parameters.

5. The method for predicting the structural life of a high-speed train according to claim 3, characterized in that, The expression for the empirical degradation model is: Where a is the crack length, N is the number of fatigue cycles, and α1, β1, and c1 are all fitting parameters.

6. The method for predicting the structural life of a high-speed train according to claim 3, characterized in that, The loss function of the fusion model consists of the loss of the actual samples and the loss of the empirical physical samples, and its expression is: Loss total =αLoss real +(1-α)Loss phy Among them, Loss real Loss is the loss for the actual sample. phy The loss is for empirical physical samples, and α is the weighting coefficient. Loss of empirical physical samples phy It consists of prediction loss and degradation feature loss, and its expression is: Loss phy =βLoss phy,model +(1-β)Loss phy,character Where β is used to adjust the weighting between prediction loss and degradation feature loss, and Loss phy,model To predict losses, Loss phy,character For degradation feature loss; Δx phy Labels for physical knowledge model samples. These are the predicted values ​​from the fusion model.

7. The method for predicting the structural life of a high-speed train according to claim 1, characterized in that, The current crack length is input into a pre-built high-speed train structure life prediction model to predict the remaining life of the high-speed train structure and output the remaining service life. The specific process is as follows: The current characteristic crack length is input into the pre-built high-speed train structure life prediction model to obtain the predicted value of the degradation rate. Then, the current characteristic crack length and the predicted value of the degradation rate are added together to calculate the characteristic crack length at the next moment. The characteristic crack length at the next time step is compared with the crack propagation size threshold. If the characteristic crack length at the next time step is less than the crack propagation size threshold, the next iteration calculation continues. The remaining service life is determined by the number of iterations until the feature crack length reaches the crack propagation size threshold at the next moment.

8. A high-speed train structural life prediction system, characterized in that, include: The data acquisition module is used to acquire Lamb wave response data; The damage diagnosis module is used to input Lamb wave response data into the damage diagnosis model to obtain the current crack length; The life prediction module is used to input the current crack length into a pre-built high-speed train structure life prediction model to predict the remaining life of the high-speed train structure and output the remaining service life.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the high-speed train structure life prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the high-speed train structure life prediction method as described in any one of claims 1 to 7.

Images