A method and apparatus for measuring residual stress in shot peening based on neural networks

By using a neural network-based method, a non-contact laser ultrasonic testing system, and an arithmetic optimization algorithm, the problem of surface parameter interference in laser ultrasonic testing was solved, and the accurate measurement of residual stress in shot peening was achieved.

CN117288365BActive Publication Date: 2026-11-10SHENZHEN POLYTECHNIC +1
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202311274681.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-09-26
Publication Date
2026-11-10
Estimated Expiration
2043-09-26

AI Technical Summary

Technical Problem

Existing laser ultrasonic testing methods are subject to interference from the surface physical parameters of the tested component during shot peening, resulting in inaccurate residual stress measurement results.

Method used

A neural network-based approach was adopted to measure the ultrasonic surface wave velocity using a non-contact laser ultrasonic testing system. Surface roughness, surface hardening layer depth, and average grain size were used as input parameters to establish a shot peening strengthening finite element model and a laser ultrasonic testing finite element model. The neural network model was trained, and the arithmetic optimization algorithm was optimized to improve measurement accuracy.

Benefits of technology

It enables accurate prediction of ultrasonic surface wave velocity affected only by residual stress, improving the accuracy and precision of residual stress measurement.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117288365B_ABST
    Figure CN117288365B_ABST
Patent Text Reader

Abstract

This invention provides a method and apparatus for measuring residual stress in shot peening based on a neural network. The method includes: acquiring physical parameters of a titanium alloy aerospace component obtained through a target shot peening process; wherein the physical parameters include surface wave velocity, surface roughness of the shot-peened sample, depth of the surface hardened layer, and average grain size; inputting the physical parameters into a pre-trained surface wave velocity prediction model, and outputting the surface wave velocity of the titanium alloy aerospace component affected only by residual stress; wherein the surface wave velocity prediction model is obtained by training a pre-set neural network using a known sample set, which includes a target input sample set and a target output sample set; and calculating the residual stress of the titanium alloy aerospace component based on the surface wave velocity affected only by residual stress. This solution uses a neural network to predict the surface wave velocity affected only by residual stress, improving the accuracy of residual stress measurement for titanium alloy aerospace components.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing equipment technology, and in particular to a method and device for measuring residual stress in shot peening reinforcement based on neural networks. Background Technology

[0002] In the manufacturing of aerospace components from titanium alloys, shot peening is often required to strengthen the surface of the component, creating residual compressive stress of a certain depth and magnitude on and near the surface. This residual compressive stress generated by shot peening can counteract the tensile stress caused by milling and uneven temperature gradients, reducing the crack propagation rate on the component surface and thus improving the service life and fatigue resistance of the tested component.

[0003] In related technologies, laser ultrasound is commonly used to detect residual stress. However, existing laser ultrasound testing is often affected by the physical parameters of the surface of the component being tested, resulting in inaccurate measurement results.

[0004] Based on this, this patent proposes a method and device for measuring residual stress in shot peening reinforcement based on neural networks to solve the above-mentioned technical problems. Summary of the Invention

[0005] To effectively improve the accuracy of residual stress measurement results, this invention provides a method and apparatus for measuring residual stress in shot peening reinforcement based on neural networks.

[0006] In a first aspect, embodiments of the present invention provide a method for measuring residual stress in shot peening based on neural networks, comprising: acquiring physical parameters of a titanium alloy aerospace component obtained by target shot peening process; wherein the physical parameters include surface wave velocity, surface roughness, surface hardening layer depth and average grain size;

[0007] The physical parameters are input into a pre-trained surface wave velocity prediction model, and the output is the surface wave velocity of the titanium alloy aerospace component affected only by residual stress. The surface wave velocity prediction model is obtained by training a pre-defined neural network using a known sample set, which includes a target input sample set and a target output sample set. The target input sample set is obtained by sequentially performing K-fold cross-validation and normalization on the original input sample set, and the target output sample set is obtained by sequentially performing K-fold cross-validation and normalization on the original output sample set. The original input sample set consists of the physical parameters of the titanium alloy aerospace component obtained using different shot peening processes, and the original output sample set is obtained through simulation using a pre-defined laser ultrasonic testing finite element model.

[0008] The residual stress of the titanium alloy aerospace component is calculated based on the surface wave velocity affected only by residual stress.

[0009] Secondly, embodiments of the present invention also provide a neural network-based residual stress measurement device for shot peening, comprising: an acquisition unit for acquiring physical parameters of titanium alloy aerospace components obtained by target shot peening process; wherein the physical parameters include surface wave velocity, surface roughness, surface hardening layer depth and average grain size;

[0010] The prediction unit is used to input the physical parameters into a pre-trained surface wave velocity prediction model and output the surface wave velocity of the titanium alloy aerospace component affected only by residual stress. The surface wave velocity prediction model is obtained by training a pre-set neural network using a known sample set. The known sample set includes a target input sample set and a target output sample set. The target input sample set is obtained by sequentially performing K-fold cross-validation and normalization on the original input sample set. The target output sample set is obtained by sequentially performing K-fold cross-validation and normalization on the original output sample set. The original input sample set consists of the physical parameters of the titanium alloy aerospace component obtained using different shot peening processes. The original output sample set is obtained through simulation using a pre-set laser ultrasonic testing finite element model.

[0011] The calculation unit is used to calculate the residual stress of the titanium alloy aerospace component based on the surface wave velocity of the titanium alloy aerospace component, which is only affected by residual stress.

[0012] Thirdly, embodiments of the present invention also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method described in any embodiment of this specification.

[0013] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the methods described in any embodiment of this specification.

[0014] This invention provides a method and apparatus for measuring residual stress in shot peening based on a neural network. The method uses a non-contact laser ultrasonic testing system to measure ultrasonic surface wave velocities under different shot peening process parameters. Simultaneously, the surface roughness, surface hardening layer depth, and average grain size of the sample are measured as input parameters to the neural network. A finite element model for shot peening and a finite element model for laser ultrasonic testing incorporating the residual stress distribution from shot peening are established. The ultrasonic surface wave velocity influenced only by residual stress is obtained and used as the output label of the neural network. Based on the obtained input parameters and output labels, an original sample dataset is established and normalized. The neural network model is trained using the normalized sample dataset. The neural network model yields predicted values ​​of surface wave velocities influenced only by residual stress. These predicted values ​​are combined with the acoustoelastic coefficient to obtain an accurate measurement of the residual stress. By employing laser ultrasonic non-destructive testing technology and combining it with an arithmetic optimization algorithm to optimize the neural network, accurate prediction of ultrasonic surface wave velocities influenced only by residual stress is achieved, thereby improving the accuracy and precision of measuring residual stress using the acoustoelastic coefficient. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of a method for measuring residual stress in shot peening reinforcement based on neural networks, provided in an embodiment of the present invention.

[0017] Figure 2 This is a structural diagram of a shot peening reinforcement residual stress measurement device based on a neural network according to an embodiment of the present invention;

[0018] Figure 3 This is a schematic diagram of a non-contact laser ultrasonic testing system for residual stress provided in an embodiment of the present invention;

[0019] Figure 4 This is a schematic diagram illustrating the selection of training and validation sets using K-fold cross-validation according to an embodiment of the present invention;

[0020] Figure 5 This is a schematic diagram of the surface wave velocity prediction model structure provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] The specific implementation of the above concept is described below.

[0023] Please refer to Figure 1 This invention provides a method for measuring residual stress in shot peening based on neural networks. The method includes:

[0024] Step 100: Obtain the physical parameters of the titanium alloy aerospace component obtained by the target shot peening strengthening process; wherein, the physical parameters include surface wave velocity, surface roughness, surface hardening layer depth and average grain size;

[0025] Step 102: Input the physical parameters into a pre-trained surface wave velocity prediction model, and output the surface wave velocity of the titanium alloy aerospace component affected only by residual stress; wherein, the surface wave velocity prediction model is obtained by training a preset neural network with a known sample set, the known sample set including a target input sample set and a target output sample set, the target input sample set is obtained by sequentially performing K-fold cross-validation and normalization on the original input sample set, the target output sample set is obtained by sequentially performing K-fold cross-validation and normalization on the original output sample set, the original input sample set is the physical parameters of the titanium alloy aerospace component obtained by using different shot peening strengthening processes, and the original output sample set is obtained by simulation using a preset laser ultrasonic testing finite element model;

[0026] Step 103: Calculate the residual stress of the titanium alloy aerospace component based on the surface wave velocity of the titanium alloy aerospace component affected only by residual stress.

[0027] This invention provides a method and apparatus for measuring residual stress in shot peening based on a neural network. The method uses a non-contact laser ultrasonic testing system to measure ultrasonic surface wave velocities under different shot peening process parameters. Simultaneously, the surface roughness, surface hardening layer depth, and average grain size of the sample are measured as input parameters to the neural network. A finite element model for shot peening and a finite element model for laser ultrasonic testing incorporating the residual stress distribution from shot peening are established. The ultrasonic surface wave velocity influenced only by residual stress is obtained and used as the output label of the neural network. Based on the obtained input parameters and output labels, an original sample dataset is established and normalized. The neural network model is trained using the normalized sample dataset. The predicted surface wave velocity influenced only by residual stress is obtained from the neural network model. The accurate measurement value of residual stress is obtained by combining the predicted value with the acoustoelastic coefficient. By employing laser ultrasonic non-destructive testing technology and combining it with an arithmetic optimization algorithm to optimize the neural network, accurate prediction of ultrasonic surface wave velocity influenced only by residual stress is achieved, thereby improving the accuracy and precision of measuring residual stress using the acoustoelastic coefficient.

[0028] It is worth noting that existing technologies typically use piezoelectric ultrasonic critical refraction longitudinal waves to detect residual stress after shot peening. However, this process requires fixing a rigid piezoelectric probe to the surface of the object being tested and applying a certain thickness of coupling agent, which can affect the stability of the received signal. Although surface waves generated by laser ultrasonic technology can also achieve non-contact detection of residual stress, the surface condition of the material, including surface roughness, the depth of the surface hardened layer after shot peening, and the average grain size, all affect the velocity of the surface waves to varying degrees, leading to significant errors in residual stress measurements based on the acoustoelastic coefficient. The neural network-based residual stress measurement method provided in this embodiment, however, can obtain the surface wave velocity affected only by residual stress, achieving accurate measurement of residual stress after shot peening. This method has high academic value and significant practical application.

[0029] The following description Figure 1 The execution method of each step is shown.

[0030] For step 100:

[0031] In this embodiment of the invention, by measuring the titanium alloy aerospace component obtained by shot peening, parameters such as the surface roughness, the depth of the hardened layer in the cross section, and the average grain size of the cross section can be obtained relatively easily. However, the surface wave velocity parameter of the titanium alloy aerospace component needs to be measured by an additional laser ultrasonic testing system.

[0032] like Figure 3The diagram shown illustrates a non-contact laser ultrasonic testing system built according to this embodiment. The system comprises a data acquisition card, a computer, a motion controller, a scanning platform, a pulsed laser, and a dual-wavelength hybrid interferometer. First, a pulsed laser generates ultrasonic surface waves propagating along the surface of the component under test, and a dual-wavelength hybrid interferometer receives the ultrasonic signals. Next, the ultrasonic signals received by the dual-wavelength hybrid interferometer are digitized and recorded using the data acquisition card. This yields a time-domain ultrasonic signal covering the time from the generation of the pulsed laser to the reception of the surface waves by the interferometer.

[0033] Based on the obtained time-domain ultrasonic signal, the surface wave velocity at a fixed distance between the ultrasonic emission point and the receiving point is calculated using the following formula:

[0034]

[0035] In the formula, L is the fixed distance between the ultrasonic emission point and the receiving point, T is the surface wave propagation time, and V' is the laser ultrasonic surface wave velocity. This surface wave velocity is the measured value of the laser ultrasonic surface wave velocity, that is, the actual measured velocity under the influence of the surface roughness, surface hardening layer depth, average grain size, and residual stress of the measured component.

[0036] It is worth noting that although the aforementioned non-contact laser ultrasonic testing system can achieve non-destructive measurement of residual stress, when using laser ultrasonic surface wave detection to measure the residual stress of the tested component after shot peening, the surface wave velocity is affected by the surface roughness, surface hardening layer depth, and average grain size of the tested component, leading to inaccurate results for residual stress measured based on the acoustoelastic coefficient. Therefore, the inventors creatively devised a neural network model to establish a complex nonlinear mapping relationship between the aforementioned physical parameters, including the surface roughness, surface hardening layer depth, and average grain size of the tested component, and the surface wave velocity affected only by residual stress. This model predicts the ultrasonic surface wave velocity affected only by residual stress, thereby improving the accuracy and precision of measuring residual stress using the acoustoelastic coefficient.

[0037] Regarding step 102:

[0038] Since neural networks are a type of machine learning algorithm, they can effectively establish complex and nonlinear intrinsic relationships between various physical quantities by processing or analyzing large amounts of raw data, thereby achieving accurate prediction of the target problem. This gives them significant advantages and good practicality in data mining. Therefore, after obtaining the original input sample set of the neural network using the methods described above, it is also necessary to calculate the original output sample set of the neural network.

[0039] In this embodiment of the invention, by establishing a shot peening strengthening finite element model and a laser ultrasonic testing finite element model that includes the residual stress distribution of shot peening strengthening, a relatively accurate surface wave velocity of titanium alloy aerospace components affected only by residual stress can be obtained.

[0040] Specifically, the residual stress field is first obtained using a shot peening finite element model. Then, the residual stress field generated by shot peening is extracted and used as a boundary condition, which is applied to the upper boundary of a laser ultrasonic testing finite element model containing the residual stress distribution of shot peening. This is used to simulate the laser ultrasonic surface wave signal containing the residual stress field. It is worth noting that the method of obtaining the residual stress field using a shot peening finite element model is well known to those skilled in the art and will not be elaborated here.

[0041] The following details the process of establishing a laser ultrasonic testing finite element model that includes the distribution of residual stress from shot peening.

[0042] Neglecting thermal radiation and convection losses, the thermal diffusion equation within a solid irradiated by a laser is expressed as:

[0043]

[0044] In the formula, T(x,y,z,t) represents the temperature field distribution within the solid material, Q(x,y,z,t) represents the heat generated by the heat source per unit volume and per unit time within the solid material, k=K / (ρC) represents the thermal diffusivity of the solid material, K represents the thermal diffusivity of the solid material, ρ represents the density of the material, and C represents the specific heat capacity of the material.

[0045] The initial conditions for the temperature field of a solid material irradiated by laser are:

[0046] T(x,y,z,0)=T0

[0047] In the formula, T0 represents a function of the uniformly distributed temperature field within the material, i.e., it is in a uniform temperature field before laser irradiation.

[0048] The heat exchange boundary conditions for laser irradiation of a solid surface are as follows:

[0049] T(x,y,z,0)=T0

[0050] The thermal effect of a laser beam is as follows:

[0051] Q(x,y,z,t)=I m ·A(γ·e -γz )·f(x,y,z)·q(t)

[0052] In the formula, I mLet f(x,y,z) represent the peak power density of the incident pulsed laser, A represent the laser absorptivity of the material, γ represent the laser absorption coefficient of the material, f(x,y,z) represent the spatial distribution function of the pulsed laser beam, and q(t) represent the temporal distribution function of the pulsed laser beam.

[0053] The laser beam emitted by a pulsed laser follows a Gaussian distribution in both space and time; therefore, f(x,y,z) and q(t) are expressed as follows:

[0054]

[0055]

[0056] In the formula, r0 represents the half-width of the pulsed laser beam, and τ represents the rise time of the pulsed laser beam.

[0057] When a solid material surface is irradiated by a laser, and other surfaces of the solid material are subjected to thermal insulation treatment, the following thermal boundary conditions must be satisfied:

[0058]

[0059] In the formula, n represents the normal direction vector of the surface other than the material surface irradiated by the laser.

[0060] The thermoelastic coupling governing equations of the temperature and displacement fields within a solid under laser irradiation are expressed as follows:

[0061]

[0062] In the formula, u is the displacement vector field, β represents the thermal modulus, β=(3λ+2μ), λ and μ represent the Lamé constant, and α T This represents the coefficient of linear thermal expansion.

[0063] Set the time step: The time step Δt is set as follows:

[0064]

[0065] In the formula, f max This represents the maximum frequency of the ultrasonic surface wave in the simulation calculation.

[0066] Set the grid size: The grid size should be at least 1 / 20 of the smallest wavelength being analyzed, and can be obtained from the following formula:

[0067]

[0068] In the formula, λ min This represents the minimum wavelength of an ultrasonic surface wave.

[0069] It is worth noting that, in order to reduce the computational burden of the finite element model, the inventors creatively simplified the three-dimensional model into a two-dimensional model for research. In stress analysis, the bottom and sides of the two-dimensional model are set as low-reflection boundaries to suppress the influence of reflected echoes on the measurement of the received surface wave propagation time. The residual stress field is obtained using the shot peening finite element model. Then, the residual stress field σ(x,y) generated by shot peening is extracted and used as a boundary condition, which is applied to the upper boundary of the laser ultrasonic testing finite element model containing the residual stress distribution of shot peening, to simulate the laser ultrasonic surface wave signal containing the residual stress field. Finally, based on the laser ultrasonic testing finite element model containing the residual stress distribution of shot peening established in the above process, the time-domain displacement signal of the laser thermoelastic excited ultrasound can be obtained after solving the simulation software. Thus, the ultrasonic surface wave velocity affected only by residual stress can be calculated, which is the original output sample set mentioned above. The above process of solving the problem using simulation software is well known to those skilled in the art and will not be described in detail here.

[0070] In this embodiment of the invention, after obtaining the original input sample set and the original output sample set for training the neural network, in order to alleviate the overfitting phenomenon of the neural network model and further improve the generalization ability of the neural network model, the inventors creatively conceived of performing K-fold cross-validation on the original sample dataset (please refer to...). Figure 4 First, the sample dataset is divided into a training set and a test set. Then, the training set is evenly divided into K groups, namely D1, D2, ... D2. K For each set of data, training and validation are performed sequentially. One set is designated as the validation set, and the remaining K-1 sets are designated as the training set. Generally, the value of K is between 5 and 10. The training is repeated K times, and the optimal value of the evaluation metric from the K iterations is used as the final evaluation metric. Finally, the model is retrained on the entire training set using the optimal hyperparameters, and its performance is evaluated using the test set.

[0071] Furthermore, the original sample set is linearly mapped to the range [0,1] using the min-max normalization method, and each data point in the original sample set is normalized to obtain the target input sample set and target output sample set used to train the neural network model. The normalization formula is shown below:

[0072]

[0073] In the formula, x is the original data sample set, x min-value It is the minimum value in the original dataset, x max-value It is the maximum value in the original dataset, x k The data is normalized. The output predicted by the neural network is inversely normalized by reversing the normalization process of the input data.

[0074] A backpropagation (BP) neural network is a type of feedforward neural network that uses the backpropagation algorithm to adjust network weights and achieve model training and prediction. However, BP neural networks also have some limitations when predicting surface wave velocities affected only by residual stress. For example, they have long training times, require extensive data preprocessing, and are prone to getting trapped in local optima, which can lead to significant prediction biases and thus fail to accurately predict surface wave velocities affected only by residual stress.

[0075] Furthermore, the inventors' ingenious discovery allows for the optimization of the weights and thresholds of the neural network using the advantages of global and local search in arithmetic optimization algorithms, accurately predicting the surface wave velocity affected only by residual stress. Therefore, by constructing an improved neural network (i.e., a pre-defined neural network), the measurement of residual stress in shot peening based on neural networks has high academic value and significant practical application.

[0076] like Figure 5 The diagram shown illustrates the structure of a neural network model provided in this embodiment. During the forward propagation of the neural network, the input layer has four nodes: surface wave velocity, surface roughness, surface hardening layer, and average grain size. The output layer predicts the surface wave velocity, which is only affected by residual stress. According to empirical rules, the number of neurons in the hidden layer can be determined by the following formula:

[0077] B≤2A+1

[0078] In the formula, B is the number of hidden layer neurons, and A is the number of input layer neurons. A range for the number of hidden layer neurons is determined, and within this range, different numbers of hidden layer neurons are selected for prediction. The predicted values ​​are compared with the actual values, and finally, the number of hidden layer neurons is adjusted based on the error.

[0079] Furthermore, considering that neural networks are prone to getting stuck in local optima and have long training times, an arithmetic optimization algorithm is used to optimize their structure and parameters. Multiplication and division operations are used for global search, while addition and subtraction operations are used for local search. This can quickly find the global optimum and improve the efficiency of neural networks in predicting surface wave velocities.

[0080] The training process of the surface wave velocity prediction model is described below.

[0081] In one embodiment of the present invention, the surface wave velocity prediction model is specifically trained in the following manner:

[0082] An initial population with a uniform distribution is randomly generated based on an arithmetic optimization algorithm; wherein, the initial population includes the weights and thresholds of the neural network, and the weights and thresholds are individuals in the initial population;

[0083] The optimization strategy is determined based on the acceleration function of the mathematical optimizer; the optimization strategy includes a global search strategy and a local exploitation strategy.

[0084] The position of each individual in the initial population is initialized, and the fitness value of each individual in the initial population is calculated.

[0085] The position of each individual in the initial population is updated, and the fitness value of each individual in the initial population is calculated.

[0086] The fitness value of each individual is calculated based on a preset fitness function to obtain the next generation population. The fitness value calculation for each individual is performed iteratively until the fitness value meets the preset standard or the maximum number of iterations is completed, and the globally optimal individual with the highest fitness value is output. The globally optimal individual includes the optimal weight and the optimal threshold.

[0087] The optimal weights and optimal thresholds are assigned to the neural network to obtain the preset neural network;

[0088] The surface wave velocity prediction model is obtained by training the preset neural network using a known sample set.

[0089] Specifically, the process of optimizing the weights and thresholds of a neural network using arithmetic optimization algorithms includes the following steps:

[0090] First, initialize the parameters and number of iterations of the arithmetic optimization algorithm, setting the maximum number of iterations to 100.

[0091] According to the acceleration function of the Math Optimizer (MOA), as shown in the formula below, when r1 > MOA, AOA enters the global search phase; when r2 < MOA, AOA enters the local development phase.

[0092]

[0093] In the formula, r1 is a random number between [0,1], MOA(t) is the current value of the acceleration function, and Min and Max are the minimum and maximum values ​​of the acceleration function, respectively, taking values ​​of 0.2 and 1. T is the maximum number of iterations, and t is the current number of iterations.

[0094] Global search phase

[0095] When r² ≥ 0.5, a multiplication search strategy is executed; when r² < 0.5, a division search strategy is executed. The position update formula is as follows:

[0096] X (t+1) =X b(t)×MOP×((Ub-Lb)×μ+Lb),r2≥0.5

[0097] X (t+1) =X b (t)÷(MOP+ε)×((Ub-Lb)×μ+Lb),r2<0.5

[0098] In the formula, r2 is a random number in the range [0, 1], and X... (t+1) For the position of the next generation of particles, X b (t) represents the position of the particle with the best current fitness, μ is the search process control parameter with a value of 0.499, ε is the minimum value, and MOP is the Math Optimizer Probability (MOP), which is calculated as follows:

[0099]

[0100] In the formula, MOP(t) is the current mathematical optimizer probability, and α is the iteration sensitivity coefficient, which takes a value of 5.

[0101] Local search phase

[0102] When r3 ≥ 0.5, an additive search strategy is executed; when r3 < 0.5, a subtractive search strategy is executed. The position update formula is as follows:

[0103] X (t+1) =X b (t)+MOP×((Ub-Lb)×μ+Lb),r3≥0.5

[0104] X (t+1) =X b (t)-MOP×((Ub-Lb)×μ+Lb),r3<0.5

[0105] In the formula, r3 is a random number in the range [0, 1].

[0106] It is worth noting that each individual in the initial population performs global optimization and position update based on the distribution characteristics of arithmetic operators. Randomly generated individuals are assigned to the neural network, and the error function is used as the fitness function to check whether the obtained individuals meet the optimal criteria. Individuals that do not meet the criteria are checked for fitness again. If they meet the criteria, the optimal individual is used as the weight and threshold of the BP neural network, and the training is repeated a number of times until the neural network parameters are adjusted to the set accuracy range or until the training number is satisfied.

[0107] In this embodiment of the invention, the surface wave velocity prediction model trained after optimization by the arithmetic optimization algorithm is as follows:

[0108]

[0109]

[0110]

[0111]

[0112] In the formula, g(x) is the sigmoid growth function; w ij w represents the weights from the input layer to the hidden layer. jk a represents the weights from the hidden layer to the output layer. j b is the threshold from the input layer to the hidden layer. k H is the threshold from the hidden layer to the output layer. j x is the output of the j-th node in the hidden layer; i This is the input of the i-th node in the input layer; For the predicted surface wave velocity of titanium alloy aerospace components affected only by residual stress; y i Y represents the surface wave velocity of the titanium alloy aerospace component affected only by residual stress; Y is the error function; and N is the number of data points in the known sample set.

[0113] In this embodiment of the invention, the residual stress of the titanium alloy aerospace component is calculated using the following formula:

[0114] Δv=v-v0=Kσ

[0115] In the formula, σ represents residual stress; K represents the acoustoelastic coefficient; v represents the surface wave velocity of the titanium alloy aerospace component affected only by residual stress; and v0 represents the surface wave velocity without stress. It should be noted that the acoustoelastic coefficient mentioned in this embodiment needs to be calibrated experimentally beforehand. For example, the calibration experiment requires no residual stress in the workpiece, therefore, the calibration component needs to undergo stress-relief annealing. After stress-relief annealing to eliminate residual stress, the calibration workpiece is subjected to graded loading tensile testing using a tensile testing machine. This induces tensile stress within the workpiece. The test consists of eight loading levels, each at 30 MPa, held for 5 minutes. During continuous loading, the surface wave velocity at a fixed propagation distance is measured using laser ultrasonic equipment. By fitting the stress experienced by the titanium alloy aerospace component with the measured surface wave velocity, the corresponding acoustoelastic coefficient can be obtained.

[0116] Understandably, by combining the surface wave velocity value predicted by the surface wave velocity prediction model, which is only affected by residual stress, with the acoustoelastic coefficient K, the accurate residual stress value of the tested component can be calculated. The residual stress measurement results are then compared with the average residual stress measured by X-ray diffraction, thus verifying the effectiveness of the proposed method. This verification process is readily available to those skilled in the art based on actual measurement results and will not be elaborated upon here.

[0117] The foregoing has described specific embodiments of the invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than those in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0118] According to another embodiment, the present invention provides a neural network-based device for measuring residual stress in shot peening reinforcement. Figure 2 A structural diagram of a neural network-based residual stress measurement device for shot peening reinforcement, according to one embodiment, is shown. It will be understood that this device can be implemented using any apparatus, device, platform, or cluster of devices with computing and processing capabilities. Figure 2 As shown, the device includes: an acquisition unit 200, a prediction unit 202, and a calculation unit 204. The main functions of each component unit are as follows:

[0119] The acquisition unit 200 is used to acquire the physical parameters of the titanium alloy aerospace component obtained by the target shot peening process; wherein, the physical parameters include surface wave velocity, surface roughness of the shot-peened sample, depth of surface hardened layer and average grain size;

[0120] The prediction unit 202 is used to input physical parameters into a pre-trained surface wave velocity prediction model and output the surface wave velocity of the titanium alloy aerospace component affected only by residual stress. The surface wave velocity prediction model is obtained by training a preset neural network with a known sample set. The known sample set includes a target input sample set and a target output sample set. The target input sample set is obtained by sequentially performing K-fold cross-validation and normalization on the original input sample set. The target output sample set is obtained by sequentially performing K-fold cross-validation and normalization on the original output sample set. The original input sample set consists of the physical parameters of the titanium alloy aerospace component obtained by different shot peening strengthening processes. The original output sample set is obtained by simulation using a preset laser ultrasonic testing finite element model.

[0121] The calculation unit 204 is used to calculate the residual stress of the titanium alloy aerospace component based on the surface wave velocity of the titanium alloy aerospace component affected only by residual stress.

[0122] In one embodiment of the present invention, the surface wave velocity prediction model is specifically trained in the following manner:

[0123] An initial population with a uniform distribution is randomly generated based on an arithmetic optimization algorithm; wherein, the initial population includes the weights and thresholds of the neural network, and the weights and thresholds are individuals in the initial population;

[0124] The optimization strategy is determined based on the acceleration function of the mathematical optimizer; the optimization strategy includes a global search strategy and a local exploitation strategy.

[0125] The position of each individual in the initial population is initialized, and the fitness value of each individual in the initial population is calculated.

[0126] The position of each individual in the initial population is updated, and the fitness value of each individual in the initial population is calculated.

[0127] The fitness value of each individual is calculated based on a preset fitness function to obtain the next generation population. The fitness value calculation for each individual is performed iteratively until the fitness value meets the preset standard or the maximum number of iterations is completed, and the globally optimal individual with the highest fitness value is output. The globally optimal individual includes the optimal weight and the optimal threshold.

[0128] The optimal weights and optimal thresholds are assigned to the neural network to obtain the preset neural network;

[0129] A surface wave velocity prediction model is obtained by training a pre-defined neural network using a known sample set.

[0130] In this embodiment of the invention, the surface wave velocity prediction model is as follows:

[0131]

[0132]

[0133]

[0134]

[0135] In the formula, g(x) is the sigmoid growth function; w ij w represents the weights from the input layer to the hidden layer. jk a represents the weights from the hidden layer to the output layer. j b is the threshold from the input layer to the hidden layer. k H is the threshold from the hidden layer to the output layer. j x is the output of the j-th node in the hidden layer; i This is the input of the i-th node in the input layer; For the predicted surface wave velocity of titanium alloy aerospace components affected only by residual stress; y i Y represents the surface wave velocity of the titanium alloy aerospace component affected only by residual stress; Y is the error function; and N is the number of data points in the known sample set.

[0136] In this embodiment of the invention, the residual stress is calculated using the following formula:

[0137] Δv=v-v0=Kσ

[0138] In the formula, σ is the residual stress; K is the acoustoelastic coefficient; v is the surface wave velocity of the titanium alloy aerospace component affected only by the residual stress; and v0 is the surface wave velocity without stress. According to another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed in a computer, causes the computer to perform a combination of... Figure 1 The method described.

[0139] According to another embodiment, an electronic device is also provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, it implements a combination... Figure 1 The method described.

[0140] The various embodiments of the present invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0141] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for measuring residual stress in shot peening based on neural networks, characterized in that, include: The physical parameters of titanium alloy aerospace components obtained by target shot peening strengthening process are obtained; wherein, the physical parameters include surface wave velocity, surface roughness of shot-peened sample, surface hardening layer depth and average grain size; The physical parameters are input into a pre-trained surface wave velocity prediction model, and the output is the surface wave velocity of the titanium alloy aerospace component affected only by residual stress. The surface wave velocity prediction model is obtained by training a pre-set neural network using a known sample set, which includes a target input sample set and a target output sample set. The target input sample set is obtained by sequentially performing K-fold cross-validation and normalization on the original input sample set, and the target output sample set is obtained by sequentially performing K-fold cross-validation and normalization on the original output sample set. The original input sample set consists of the physical parameters of the titanium alloy aerospace component obtained using different shot peening processes. The original output sample set is obtained through simulation using a pre-set laser ultrasonic testing finite element model. First, the residual stress field is obtained using the shot peening finite element model. Then, the residual stress field generated by shot peening is extracted and used as a boundary condition, which is applied to the upper boundary of the laser ultrasonic testing finite element model containing the residual stress distribution of shot peening, to simulate the laser ultrasonic surface wave signal containing the residual stress field. The residual stress of the titanium alloy aerospace component is calculated based on the surface wave velocity affected only by residual stress. The surface wave velocity prediction model is specifically trained in the following manner: An initial population with a uniform distribution is randomly generated based on an arithmetic optimization algorithm; wherein, the initial population includes the weights and thresholds of the neural network, and the weights and thresholds are individuals of the initial population; The optimization strategy is determined based on the acceleration function of the mathematical optimizer; wherein the optimization strategy includes a global search strategy and a local exploitation strategy; The position of each individual in the initial population is initialized, and the fitness value of each individual in the initial population is calculated; The position of each individual in the initial population is updated, and the fitness value of each individual in the initial population is calculated. The fitness value of each individual is calculated based on a preset fitness function to obtain the next generation population. The fitness value calculation for each individual is performed iteratively until the fitness value meets the preset standard or the maximum number of iterations is completed, and the globally optimal individual with the highest fitness value is output. The globally optimal individual includes the optimal weight and the optimal threshold. The optimal weights and the optimal thresholds are assigned to the neural network to obtain the preset neural network; The surface wave velocity prediction model is obtained by training the preset neural network using a known sample set. The surface wave velocity prediction model is as follows: In the formula, It is an S-shaped growth function; The weights from the input layer to the hidden layer. The weights from the hidden layer to the output layer. The threshold from the input layer to the hidden layer. The threshold value from the hidden layer to the output layer; For the hidden layer The output of each node; For the input layer Input of each node; The predicted surface wave velocity of the titanium alloy aerospace component is affected only by residual stress; The surface wave velocity of the titanium alloy aerospace component is affected only by residual stress; It is the error function; The number of data points in the known sample set.

2. The method according to claim 1, characterized in that, The residual stress is calculated using the following formula: In the formula, This is residual stress; K The acoustic elastic coefficient; The surface wave velocity of the titanium alloy aerospace component is affected only by residual stress. It is the surface wave velocity under no stress.

3. A device for measuring residual stress in shot peening based on neural networks, characterized in that, include: The acquisition unit is used to acquire the physical parameters of titanium alloy aerospace components obtained by target shot peening strengthening process; wherein, the physical parameters include surface wave velocity, surface roughness, surface hardening layer depth and average grain size; The prediction unit is used to input the physical parameters into a pre-trained surface wave velocity prediction model and output the surface wave velocity of the titanium alloy aerospace component affected only by residual stress. The surface wave velocity prediction model is obtained by training a pre-set neural network using a known sample set, which includes a target input sample set and a target output sample set. The target input sample set is obtained by sequentially performing K-fold cross-validation and normalization on the original input sample set, and the target output sample set is obtained by sequentially performing K-fold cross-validation and normalization on the original output sample set. The original input sample set consists of the physical parameters of the titanium alloy aerospace component obtained using different shot peening processes. The original output sample set is obtained through simulation using a pre-set laser ultrasonic testing finite element model. First, the residual stress field is obtained using the shot peening finite element model. Then, the residual stress field generated by shot peening is extracted and used as a boundary condition, which is applied to the upper boundary of the laser ultrasonic testing finite element model containing the residual stress distribution of shot peening, to simulate the laser ultrasonic surface wave signal containing the residual stress field. A calculation unit is used to calculate the residual stress of the titanium alloy aerospace component based on the surface wave velocity of the titanium alloy aerospace component, which is only affected by residual stress. The surface wave velocity prediction model is specifically trained in the following manner: An initial population with a uniform distribution is randomly generated based on an arithmetic optimization algorithm; wherein, the initial population includes the weights and thresholds of the neural network, and the weights and thresholds are individuals of the initial population; The optimization strategy is determined based on the acceleration function of the mathematical optimizer; wherein the optimization strategy includes a global search strategy and a local exploitation strategy; The position of each individual in the initial population is initialized, and the fitness value of each individual in the initial population is calculated; The position of each individual in the initial population is updated, and the fitness value of each individual in the initial population is calculated. The fitness value of each individual is calculated based on a preset fitness function to obtain the next generation population. The fitness value calculation for each individual is performed iteratively until the fitness value meets the preset standard or the maximum number of iterations is completed, and the globally optimal individual with the highest fitness value is output. The globally optimal individual includes the optimal weight and the optimal threshold. The optimal weights and the optimal thresholds are assigned to the neural network to obtain the preset neural network; The surface wave velocity prediction model is obtained by training the preset neural network using a known sample set. The surface wave velocity prediction model is as follows: In the formula, It is an S-shaped growth function; The weights from the input layer to the hidden layer. The weights from the hidden layer to the output layer. The threshold from the input layer to the hidden layer. The threshold value from the hidden layer to the output layer; For the hidden layer The output of each node; For the input layer Input of each node; The predicted surface wave velocity of the titanium alloy aerospace component is affected only by residual stress; The surface wave velocity of the titanium alloy aerospace component is affected only by residual stress; It is the error function; The number of data points in the known sample set.

4. The apparatus according to claim 3, characterized in that, The residual stress is calculated using the following formula: In the formula, This is residual stress; K The acoustic elastic coefficient; The surface wave velocity of the titanium alloy aerospace component is affected only by residual stress. It is the surface wave velocity under no stress.

5. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1 or 2.

6. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed in a computer, causes the computer to perform the method of any one of claims 1 or 2.

Citation Information

Patent Citations

  • Shot peening strengthening surface integrity prediction method based on BP neural network

    CN111401623A