A method for identifying crack characteristics of a metal structure based on measured strain data

Abstract

The application provides a metal structure crack feature recognition method based on measured strain data, utilizes the strain data collected on the surface of a structure, adopts a machine learning method to establish a quantitative relationship model of the measured strain data and crack features, and further realizes the purpose of directly predicting the crack position and crack length through the strain data. The application realizes real-time detection of the fatigue cracks of the aircraft metal structure, provides support for ensuring the service safety of the aircraft, and compared with the crack detection methods of other sensors, utilizes the strain field singularity principle near the crack tip, uses the strain data collected on the surface of the structure for real-time detection of the cracks, the strain collection method is mature, the data source is highly reliable, the weight of the aircraft is not significantly increased, the structural functional characteristics are not affected, and the final precision error of the crack feature detection is less than 1mm.

Description

Technical Field

[0001] This invention belongs to the field of structural mechanics fatigue crack measurement, and specifically relates to a method for identifying crack features in metal structures based on measured strain data. Background Technology

[0002] When aircraft metal structures are subjected to repeated loading, fatigue damage is inevitable, leading to the initiation of fatigue cracks in weak points of the structure, which then slowly propagate during subsequent repeated loading. Timely detection and identification of fatigue cracks, followed by appropriate repair measures, are crucial for ensuring the operational safety and reliability of aircraft.

[0003] As the complexity of advanced aircraft continues to increase, the cost of fatigue crack inspection and maintenance in their structures is rising. Traditional methods, primarily based on maintenance programs, involve disassembling the aircraft structure and directly detecting cracks at predetermined maintenance intervals. This approach suffers from high costs, long cycles, and significant challenges, and is susceptible to issues such as unavailable detection methods or human error leading to missed detections, threatening the operational safety of the aircraft. Due to the variability in structural and material properties, current life management methods are often conservative, potentially resulting in unnecessary maintenance and return of aircraft to the factory, impacting operational efficiency. Therefore, providing an accurate and convenient real-time crack detection method is crucial for ensuring aircraft operational safety and improving equipment uptime.

[0004] Although various real-time crack sensors based on different principles (such as piezoelectric sensors and eddy current sensors) have been proposed, they all suffer from problems such as insufficient detection accuracy, limited crack types detected, false alarms, and potential impact on the original structural function. Current crack identification sensor methods include active Lamb wave methods, eddy current array sensor detection methods, piezoelectric impedance methods, and fiber optic sensor detection methods. However, the reliability, lifespan, lightweight design, and air-to-ground consistency of these methods still require extensive engineering verification and are not yet widely used in actual structures. Strain gauges are a mature technology in engineering, providing highly reliable detection results. Aircraft structures widely use bonded strain gauges to collect structural strain histories during flight to understand structural load changes, but this method has not yet been used for crack identification. Further application of this collected strain data to fatigue crack identification could achieve real-time fatigue crack detection without significantly increasing the aircraft's weight. With the development of artificial intelligence algorithms, machine learning methods are becoming increasingly sophisticated, abundant, and accurate. Researchers have applied machine learning methods to flight load inversion, flight maneuver identification, and crack prediction, achieving high accuracy. By utilizing machine learning methods to establish the correlation between measured strain data and structural fatigue crack characteristics, a structural fatigue crack feature identification method based on measured strain can be formed. This can effectively reduce the detection cost of fatigue cracks and improve the crack detection accuracy. However, in crack prediction, there is still no mature and complete work on how to establish the correlation between strain data and structural cracks. The difficulty lies in how to determine the location and quantity of strain monitoring, how to correlate strain data with crack length, and how to improve the accuracy of machine learning methods in predicting crack length. Summary of the Invention

[0005] To address the problems of existing technologies, this invention proposes a method for identifying crack features in metal structures based on measured strain data. This method can determine the location and quantity of strain monitoring, select a machine learning method, and improve the prediction accuracy of the machine learning method. This invention utilizes strain data collected from the structural surface and employs a machine learning method to establish a quantitative relationship model between measured strain data and crack features, thereby achieving the goal of directly predicting the crack location and crack length through strain data. This invention enables real-time detection of fatigue cracks in aircraft metal structures, providing support for ensuring the safe operation of aircraft.

[0006] The technical solution of this invention is as follows.

[0007] A method for identifying crack features in metal structures based on measured strain data includes the following steps.

[0008] Step 1: Establish a finite element calculation model of the metal structure to obtain the stress concentration points.

[0009] Step 2: Establish a finite element calculation model of the cracked structure to obtain the structural strain field under different crack locations and different length combinations.

[0010] Step 3: Calculate the strain variance of the structural surface, determine the location of the strain measurement points on the structural surface based on the strain variance, calculate the strain value at the measurement points and perform equivalent transformation.

[0011] Step 4: Establish a machine learning model, using the strain value after equivalent transformation in Step 3 as the model input and the crack length at the corresponding measurement point as the output, to train the model and obtain the model accuracy.

[0012] Step 5: Use the model accuracy obtained in Step 4 to determine whether the measurement points used to train the machine learning model are suitable as input to the machine learning model. If the accuracy meets the requirements, the trained machine learning model is obtained. If the accuracy does not meet the requirements, adjust the position of the strain measurement points on the structural surface determined in Step 3. The adjustment principle is to reduce the distance between the measurement point and the crack and increase the number of strain measurement points, but the measurement points should not interfere with each other. Then, perform the equivalent transformation again and train the machine learning model until the accuracy requirements are met.

[0013] Step 6: Apply the model obtained in Step 5 to the process of real-time crack length prediction. Input the strain values ​​of the measured points collected in real time into the model to obtain the predicted value of crack length.

[0014] Furthermore, the specific process of step 1 is as follows.

[0015] For a given metal structure, a finite element model of the structure in its intact state is established based on its geometric dimensions, material properties, boundary conditions, and loading methods. The stress field of the structure is obtained through finite element calculations, thereby yielding... n There are several stress concentration sites, which are considered as potential fatigue cracking sites. Each cracking site is numbered as P1, P2, ..., Pn.

[0016] Furthermore, step 2 specifically includes the following sub-steps.

[0017] Step 2.1: In n In each stress concentration location, assuming that it may occur m There are 10 cracks, and each crack occurs at a stress concentration point; then, select from P1, P2, ..., Pn. m Each stress concentration point is considered a fatigue cracking site (1≤ m ≤ n ),exist There are two cases, each numbered Q. mj ,in j express m The first cracked section jIn this case, 1≤ j ≤ .

[0018] Step 2.2: For a given fatigue cracking condition Q mj Based on the extent and form of crack expansion, in its m Different combinations of crack lengths were defined for each cracked location, and a finite element model of the cracked structure was established. The strain field distribution of the structure under different combinations of crack lengths was obtained through finite element calculation.

[0019] Step 2.3 for m The first cracked part j In this case, j =1, 2, …, Repeat step 2.2 to calculate... m The results of the structural strain field distribution under all combinations of fatigue crack locations.

[0020] Step 2.4: Change the number of fatigue crack locations, and take... m= 1,2, …,n Repeat steps 2.2 to 2.3 to calculate the structural strain field distribution for all combinations of fatigue cracking sites.

[0021] Furthermore, step 3 includes the following sub-steps.

[0022] Step 3.1: Based on the structural strain field results in Step 2.4, calculate the strain variance at a certain location on the structural surface under all combinations of crack locations and crack lengths.

[0023] 。

[0024] In the formula, For strain variance, Let be the equivalent variable value for the k-th case. N The total number of all possible combinations. Let be the average equivalent variable for all combinations, where xyz Represents the coordinates of the structural surface. Indicates the first k The strain value at a certain location under this condition is calculated as follows.

[0025] .

[0026] Step 3.2: Reselect all surface locations of the structure point by point, repeat step 3.1, and calculate the strain variance of all surface locations until the strain variance of all locations on the structural surface is obtained. Plot the strain variance distribution map of the structural surface. The selection of measurement points should be based on the principle that the strain variance of the measurement points is greater than the set value, and that the installation of strain gauges at the measurement points does not interfere with each other or with crack propagation. K These locations are designated as strain measurement points. K ≥ 3.

[0027] Step 3.3: Perform an equivalent transformation on the strain values ​​at the measuring points: The strain values ​​at the strain monitoring points are linearly related to the external load. The strain value at the kth measuring point after the equivalent transformation... It is expressed as.

[0028] .

[0029] in This represents the strain value at the far end, which is the equivalent strain value at the measuring point when the crack length remains unchanged. No change.

[0030] Furthermore, step 4 includes the following sub-steps.

[0031] Step 4.1: Use a deep learning neural network as a machine learning model, take the strain value after the equivalent change in step 3.3 as input, and the crack length as output, and use the neural network algorithm to establish the relationship between input and output. Select the Adam optimizer, ReLU activation function, and MSE loss function.

[0032] Step 4.2: Based on the determination in step 3.2 K The location of each strain measurement point, combined with the strain data calculated in step 2.4, will... K The strain data from each strain measurement point is fed into the machine learning model to calculate the model's accuracy.

[0033] Invention Effects

[0034] The technical advantages of this invention are as follows: This invention proposes a method for identifying fatigue crack features in metal structures based on measured strain data. It employs machine learning to establish a model of the quantitative relationship between measured strain data and the location and length of cracks in the metal structure. This allows for real-time prediction of the crack location and length using measured strain data, achieving accurate, simple, and real-time detection of fatigue cracks in aircraft metal structures. Compared to existing crack detection methods using other sensors, this invention utilizes the singularity principle of the strain field near the crack tip, applying strain data collected from the structural surface to real-time crack detection. The strain acquisition method is mature, the data source is highly reliable, it does not significantly increase the weight of the aircraft, does not affect the structural functional characteristics, and the final accuracy error of crack feature detection is less than 1 mm. Attached Figure Description

[0035] Figure 1 This is a flowchart of the implementation process of the present invention.

[0036] Figure 2 This is a Mises result cloud diagram of the finite element calculation results for an example of the present invention.

[0037] Figure 3 This is a sensitivity analysis diagram for an example of the present invention.

[0038] Figure 4 This indicates the proposed installation location for the strain sensor.

[0039] Figure 5 This relates the combination of strain sensors to the loss value of the model loss function.

[0040] Figure 6 This is a schematic diagram of a neural network. Detailed Implementation

[0041] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0042] See Figures 1-6 A method for identifying crack features in metal structures based on measured strain data mainly includes the following steps.

[0043] Step 1: Establish a finite element model of the metal structure to obtain stress concentration points.

[0044] For a given metal structure, a finite element model of the metal structure is established based on its geometric dimensions, material properties, boundary conditions, and loading methods; the stress field of the structure is obtained through finite element calculations, and thus the stress field is obtained. n There are several stress concentration sites, which are considered as potential fatigue cracking sites. Each cracking site is numbered as P1, P2, …, Pn.

[0045] Step 2: Establish a finite element calculation model of the cracked structure to obtain the structural strain field under different crack locations and different length combinations.

[0046] Step 2.1: List the possible combinations of crack locations.

[0047] In structure n In each stress concentration location, assuming that it may occur m There are 10 cracks, each located at a stress concentration point, i.e., selected from P1, P2, …, Pn. m Each stress concentration point is considered a fatigue cracking site (1≤ m ≤ n ),exist There are two cases, each numbered Q. mj ,in j express m The first cracked section j In this case, 1≤ j ≤ .

[0048] Step 2.2: For a given fatigue cracking condition Q mj Based on the extent and form of crack expansion, in its m Different combinations of crack lengths were defined for each cracked location, and a finite element model of the cracked structure was established. The strain field distribution of the structure under different combinations of crack lengths was obtained through finite element calculation.

[0049] Step 2.3: Calculation m Strain field distribution of the structure under the condition of cracked location.

[0050] for m The first cracked part j This situation ( j =1, 2, …, Repeat step 2.2 to calculate all m The results of the structural strain field distribution under the condition of a combination of fatigue cracking sites.

[0051] Step 2.4: Calculate the structural strain field distribution for all combinations of fatigue crack locations.

[0052] Changing the number of fatigue crack sites, i.e., changing m The number of cracked areas ( m= 1,2, …,n Repeat steps 2.2 to 2.3 to calculate the structural strain field distribution for all combinations of fatigue cracking sites.

[0053] Step 3: Calculate the strain variance of the structural surface, determine the location of the strain measurement points on the structural surface based on the strain variance, calculate the strain value at the measurement points and perform equivalent transformation.

[0054] Step 3.1: Based on the structural strain field results in Step 2.4, calculate the strain variance at a certain location on the structural surface under all combinations of crack locations and crack lengths according to Equation (1).

[0055] (1).

[0056] In the formula, xyz Represents the coordinates of the structural surface. For strain variance, Let be the equivalent variable value for the k-th case. N The total number of all possible combinations. The average equivalent strain for all combinations is calculated using equation (2). Indicates the first k The strain value at a certain location under certain conditions.

[0057] (2).

[0058] Step 3.2: Preliminarily determine the location of strain measurement points on the structural surface.

[0059] Reselect all surface locations of the structure point by point (i.e., select xyz points point by point), repeat step 3.1, calculate the strain variance at each point, until the strain variance at all locations on the structural surface is obtained, and plot the strain variance distribution map of the structural surface. Locations with larger strain variances indicate that the strain data are more sensitive to changes in crack characteristics. Considering the feasibility of strain gauge installation, the initial selection is based on the principle that strain gauges can be installed at each crack location where the variance is larger, without interfering with crack propagation. K These locations are designated as strain measurement points. K ≥ 3).

[0060] Step 3.3: Perform an equivalent transformation on the strain values ​​at the measuring points.

[0061] When the crack length does not propagate but the external load changes, the strain value at the measuring point will change. If only the strain value is used as input, different combinations of strain values ​​at the measuring point will appear for the same crack, increasing the amount of training data required. This invention performs an equivalent transformation on the strain value at the measuring point, assuming a small-scale yielding assumption at the crack tip, and that the strain value at the strain monitoring point is linearly related to the external load. The equivalent transformation... k strain values ​​at each measuring point Perform the transformation of formula (3). Where... This is the strain value at the far end, which can be used to reflect changes in external load. When the crack length remains unchanged, the equivalent strain value at the measuring point is... No change.

[0062] (3).

[0063] Step 4: Establish a machine learning model, using the strain value after equivalent transformation in Step 3 as the model input and the crack length at the corresponding measurement point as the output, to train the model and obtain the model accuracy.

[0064] Step 4.1: Employ a deep learning neural network as the machine learning model. Compared to other machine learning methods, deep learning neural networks offer the advantage of increasing accuracy by deepening the network, reducing the requirements for the number and location of strain measurement points in Step 3.2, and improving the method's adaptability. Using the strain value after the equivalent transformation in Step 3.3 as input and the crack length as output, a neural network algorithm is used to establish the relationship between input and output. The Adam optimizer, ReLU activation function, and MSE loss function (Formula 4, where...) are selected. y i Represents the actual value. Represents the predicted value, total N (sample).

[0065] (4).

[0066] Step 4.2: Optimization of the number and location of strain sensors based on machine learning models.

[0067] Based on the preliminary determination in step 3.2 K The location of each strain measurement point, combined with the strain data calculated in step 2.4, will... K The strain data at each strain measurement point is fed into the machine learning model to calculate the model accuracy (i.e., the loss function value).

[0068] Step 5: Use the model accuracy obtained in Step 4 to determine whether the measurement points used to train the machine learning model are suitable as inputs. If the accuracy meets the requirements (usually, the accuracy requirement means that the difference between the measured crack length and the predicted crack length is less than 2mm), then the trained machine learning model is obtained. If the accuracy does not meet the requirements, adjust the positions of the strain measurement points on the structural surface determined in Step 3. The adjustment principle is to reduce the distance between the measurement points and the crack and increase the number of strain measurement points, but the measurement points should not interfere with each other. Then, perform the equivalent transformation again and train the machine learning model until the accuracy requirements are met, thus obtaining the final machine learning model and the required positions and number of inputs.

[0069] Step 6: Predict the crack length.

[0070] By using a machine learning model, the strain at the measurement points during the fatigue crack length test is collected in real time based on the measured strain data and then substituted into the model to achieve real-time prediction.

[0071] The present invention will be further described in detail below with reference to specific embodiments. These descriptions are for explanation purposes only and are not intended to limit the scope of the invention.

[0072] The technical process of this invention is as follows: Figure 1 As shown. To better understand the technical solution of the present invention, the above method is applied to the fatigue crack detection of perforated aluminum alloy plates.

[0073] The specific process of this embodiment includes the following steps.

[0074] Step 1: Establish a finite element model of the structure to obtain the stress concentration points.

[0075] For a given metal structure, a finite element analysis model is established based on its geometric dimensions, material properties, boundary conditions, and loading conditions to determine its intact state. The stress field of the structure is obtained through finite element calculation, and all stress concentration points and potential fatigue cracking points are identified.

[0076] Step 2: Establish a finite element calculation model of the cracked structure to obtain the structural strain field under different crack locations and different length combinations.

[0077] Fatigue cracks are introduced at stress concentration points in the finite element model. By combining different crack locations and establishing finite element models with different crack sizes, the finite element strain field results under different crack lengths are obtained through finite element analysis.

[0078] Establishing a finite element model is not necessary. If the strain values ​​of the structure can be obtained through experiments or practical applications, then establishing a finite element model is unnecessary. The reason for establishing a finite element simulation in this invention example is that it is impossible to obtain a large amount of data through numerous experiments. It should be noted that the use of finite element modeling to obtain the relationship between strain values ​​and crack values ​​in this example is not a limitation of this invention. If other personnel in the art do not use the finite element method to obtain data, it is merely a substitution for part of the process of this invention, and does not cause the essence of the corresponding technical solution to deviate from the scope of the technical solutions of the embodiments of this application.

[0079] like Figure 2 The image shown is a Mises contour plot of one of the finite element models established in this invention example. This invention example was calculated using the commercial finite element calculation software Abaqus. This example uses a perforated plate with cracks at both ends. Various finite element models are established by varying the crack lengths at both ends.

[0080] like Figure 2As shown, the crack generated directly above the perforated plate is called the upper crack, the crack generated directly below the perforated plate is called the lower crack, and the model structure on the left side of the perforated plate is called the left side of the perforated plate. It should be understood that this terminology is only for the convenience of describing the present invention and simplifying the description, and is not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0081] This example uses a perforated plate with a length of 130mm, a width of 54mm, and a central hole diameter of 5mm. The upper crack in this example follows an arithmetic progression with a minimum length of 0.2mm, a maximum length of 16mm, and a step size of 0.2mm, resulting in 80 possible crack lengths. Similarly, the lower crack also follows an arithmetic progression with a minimum length of 0.2mm, a maximum length of 16mm, and a step size of 0.2mm, again resulting in 80 possible crack lengths. Therefore, there are a total of 6400 possible crack lengths for both the upper and lower plates, requiring the creation of 6400 finite element models.

[0082] Step 3: Calculate the strain variance of the structural surface, determine the location of the strain measurement points on the structural surface based on the strain variance, calculate the strain value at the measurement points and perform equivalent transformation.

[0083] like Figure 3 As shown, the strain values ​​of all the established finite element model surfaces are output. The variance of the same location is calculated according to formula (1), and then the relationship between the strain value of the model and the crack length is obtained. The larger the variance, the more sensitive the strain at that point is to the crack length. The smaller the variance, the less sensitive the strain at that point is to the crack length.

[0084] like Figure 3 As shown, in this invention example, only the sensitivity analysis diagram of the right model with the perforated plate is produced. The reason is that the example has a left-right symmetrical structure, so the sensitivity analysis structure on the left side should be the same as the analysis structure on the right side. Figure 3 It can be seen that the closer to X=0, the more sensitive it is to the crack length (i.e., the closer to the crack, the more sensitive it is).

[0085] like Figure 4 As shown, considering the effects of crack propagation and interference between strain sensors, some points are roughly selected. The X-axis values ​​are 6, 8, 10, 12.5, 15, and 17.5. The Y-axis values ​​are -20, -15, -10, -5, 20, 15, 10, and 5. These straight lines are drawn in the figure, and the point where two lines intersect is selected as the strain sensor installation point. (See figure). Figure 4 The "x" indicates that a strain sensor is installed here to react to the load condition. This is denoted as... .like Figure 5As shown in the table, the strain sensor combinations are as follows. Specifically, (6,0) means that a strain sensor is installed at x=6, y=0. Regardless of the combination chosen, a strain sensor must be installed at position "x". Due to the symmetrical structure of this invention, symmetrical strain gauges are attached to the upper and lower surfaces. Figure 5 The combination of strain sensors should be used to account for negative Y values. Taking the first group as an example, to predict cracks, strain sensors need to be attached to seven locations: (6,20), (6,15), (6,10), (6,0), (6,-10), (6,-15), and (6,-20). Then, the possible strain values ​​are extracted from all finite element models.

[0086] Step 4, as follows Figure 6 As shown, a simple machine learning model is first established. The strain values ​​obtained in step 2 are substituted into the machine learning model according to the strain sensor combination forms in Table 5. This example uses a neural network model to obtain the loss value for each combination form. The neural network model has an input layer, hidden layers, and an output layer. The number of nodes in the input layer is equal to the number of input data, the output layer is the crack value, and the hidden layers are hyperparameters that can be adjusted, including the number of hidden layers and the number of nodes in each layer.

[0087] Furthermore, the embodiment of the present invention uses the mean squared error loss function, the calculation equation of which is as follows.

[0088] .

[0089] Furthermore, the input to the machine learning model needs to be processed by dividing the strain value in the form of a combination of strain sensors by the strain value at "x" to reduce the impact when the crack strength factor does not meet the propagation requirements but the external load changes. A normalization-like operation can also be performed to improve the accuracy of machine learning predictions.

[0090] Furthermore, it is worth noting that while this machine learning model uses a neural network to predict crack length, it is not limited to neural network models. Linear regression, ridge regression, decision tree regression, SVM, and other machine learning models can also be used to build the prediction model. Any predictions made by other individuals without using neural networks are considered within the scope of this invention.

[0091] Step 5: Using the model accuracy obtained in Step 4 (i.e., the lower the loss value, the higher the accuracy), determine whether these points are suitable as input to the machine learning model. If none are suitable, or only a few are suitable, select several more combinations of strain sensors and substitute them into the machine learning model to obtain the accuracy. From these, select combinations with high accuracy and easy installation of the strain sensors.

[0092] To prevent overfitting, a test dataset should be created. The test dataset consists of random combinations of upper and lower cracks. A finite element model is calculated, and the data from the input location of the machine learning model is extracted and input into the machine learning model to obtain the crack length prediction result. The loss value is calculated by comparing the predicted structure with the actual test data result, and this is used as the criterion for evaluating the quality of the machine learning model.

[0093] The machine learning model parameters were adjusted again using the obtained strain sensor combination to achieve the optimal machine learning model.

[0094] Depending on the machine learning model, different parameters need to be adjusted. In this example, which uses a neural network, the number of layers, the number of nodes in each layer, and the form of the activation function need to be adjusted. If a decision tree regression model is used, the parameters to adjust are the maximum depth and the minimum number of leaf nodes.

[0095] In this example, the strain sensor combination was ultimately adopted as (6,10), (6,20), (12.5,10), (12.5,20) because the accuracy requirements were met, and the strain sensors did not interfere with each other or with crack propagation.

[0096] Step 6: Through multiple iterations in step 5, the final machine learning model is obtained, which can be used to predict fatigue crack length.

[0097] The above description is merely one example of the present invention and is not intended to limit the present invention in any way. Although the process of one example of the present invention is disclosed above, it is not the only solution of the present invention. Any changes or modifications that achieve the same or similar functions as the present invention without departing from the solution of the present invention are within the protection scope of the new technical solution.

Claims

1. A method for identifying crack features in metal structures based on measured strain data, characterized in that, Includes the following steps: Step 1: Establish a finite element calculation model of the metal structure to obtain stress concentration locations; Step 2: Establish a finite element calculation model of the cracked structure to obtain the structural strain field under different crack locations and different length combinations; Includes the following sub-steps: Step 2.1: Among n stress concentration sites, assume that m cracks may occur, and each crack occurs at a stress concentration site; then select m stress concentration sites from P1, P2, ..., Pn as fatigue cracking sites, where 1 ≤ m ≤ n. There are two cases, each numbered Q. mj Where j represents the j-th case of m cracked parts, 1≤j≤ ; Step 2.2: For a given fatigue cracking condition Q mj Based on the crack expansion range and form, different crack length combinations are defined at each of the m cracking locations. A finite element model of the cracked structure is established, and the structural strain field distribution under different crack length combinations is obtained through finite element calculation. Step 2.3 For the j-th case with m crack locations, j=1, 2, …, Repeat step 2.2 to calculate the structural strain field distribution for all combinations of m fatigue crack locations. Step 2.4: Change the number of fatigue crack locations, taking m=1,2,…,n, and repeat steps 2.2~2.3 to calculate the corresponding structural strain field distribution results for all combinations of fatigue crack locations; Step 3: Calculate the strain variance of the structural surface, determine the location of the strain measurement points on the structural surface based on the strain variance, calculate the strain value at the measurement points and perform equivalent transformation; Includes the following sub-steps: Step 3.1: Based on the structural strain field results from Step 2.4, calculate the strain variance at a certain location on the structural surface for all combinations of crack locations and crack lengths. In the formula, For strain variance, Let be the equivalent variable value for the k-th case, and N be the total number of all combinations. Let xyz be the average equivalent strain for all combinations, where x, y, and z represent the surface coordinates of the structure. Let S represent the strain value at a certain location in the k-th case, calculated as follows: Step 3.2: Select all surface locations of the structure point by point again, repeat step 3.1, calculate the strain variance of all surface locations of the structure until the strain variance of all locations on the structure surface is obtained, and draw the strain variance distribution map of the structure surface. K locations were selected as strain measurement points based on the principle that the strain variance at the measurement point location is greater than the set value, and that the installation of strain gauges at the measurement points does not interfere with each other or with crack propagation. K ≥ 3. Step 3.3: Perform an equivalent transformation on the strain values ​​at the measuring points: The strain values ​​at the strain monitoring points are linearly related to the external load. The strain value at the kth measuring point after the equivalent transformation... Expressed as: in This represents the strain value at the far end, which is the equivalent strain value at the measuring point when the crack length remains unchanged. No change; Step 4: Establish a machine learning model, using the strain value after equivalent transformation in Step 3 as the model input and the crack length at the corresponding measurement point as the output, to train the model and obtain the model accuracy; Step 5: Use the model accuracy obtained in Step 4 to determine whether the measurement points used to train the machine learning model are suitable as input. If the accuracy meets the requirements, the trained machine learning model is obtained. If the accuracy does not meet the requirements, adjust the positions of the strain measurement points on the structural surface determined in Step 3. The adjustment principle is to reduce the distance between the measurement point and the crack and increase the number of strain measurement points, but the measurement points should not interfere with each other. Then, perform the equivalent transformation again and train the machine learning model until the accuracy requirements are met. Step 6: Apply the model obtained in Step 5 to the process of real-time crack length prediction. Input the strain values ​​of the measured points collected in real time into the model to obtain the predicted value of crack length.

2. The method for identifying crack features in metal structures based on measured strain data as described in claim 1, characterized in that, The specific process of step 1 is as follows: For a given metal structure, a finite element model of the metal structure in its intact state is established based on its geometric dimensions, material properties, boundary conditions, and loading form. The stress field of the structure is obtained through finite element calculation, and then n stress concentration sites are obtained. These stress concentration sites are regarded as potential fatigue cracking sites, and each cracking site is numbered as P1, P2, …, Pn.

3. The method for identifying crack features in metal structures based on measured strain data as described in claim 1, characterized in that, Step 4 includes the following sub-steps: Step 4.1: Use a deep learning neural network as a machine learning model, take the strain value after the equivalent change in step 3.3 as input, and the crack length as output. Use the neural network algorithm to establish the relationship between input and output, and select Adam optimizer, ReLU activation function, and MSE loss function. Step 4.2: Based on the K strain measurement point locations determined in Step 3.2, and combined with the strain data calculated in Step 2.4, substitute the strain data of the K strain measurement point locations into the machine learning model and calculate the model accuracy.

Images