A reverse design method and device for high-sensitivity impact test sensor

Abstract

The present invention discloses a method and device for reverse design of a high-sensitivity impact test sensor, the method comprising: determining the structural geometric parameters of a unit cell of the impact test sensor, including ligament length, ligament thickness, intermediate circular node radius and horseshoe radius, and determining the corresponding structural equivalent strain; arraying different unit cells to obtain different impact test sensors, and determining the performance parameters of the sensors, including sensitivity, acceleration and peak pressure; constructing a sensor deep learning model, taking the structural parameters of different impact test sensors as input and the corresponding performance parameters as output; the structural parameters include structural geometric parameters and structural equivalent strain; setting the target performance parameters of the impact test sensor, combining the sensor deep learning model, and using the whale optimization algorithm to reversely design the impact test sensor to obtain the optimal structural parameters. The present invention combines deep learning and the whale optimization algorithm to reversely design the impact test sensor.

Description

Technical Field

[0001] The present invention relates to the technical field of impact test sensors, and in particular to a reverse design method and device for a high-sensitivity impact test sensor. Background Art

[0002] In the field of high-precision testing and measurement, impact test sensors are one of the key components, especially in situations where instantaneous force changes need to be accurately measured, such as automobile collision testing, aviation safety assessment, ship structure damage, and material science research. These sensors can capture the magnitude and duration of impact force and acceleration in a very short time, providing engineers and researchers with key data required for accurate analysis. With the development of technology, the design of impact sensors has become increasingly complex, requiring sensors to have not only high sensitivity, but also good stability and reliability. Traditional sensor design methods often rely on experience and trial and error, which is not only inefficient, but also difficult to achieve optimal design.

[0003] In recent years, deep learning (DL) has been widely used in various fields due to its powerful capabilities in pattern recognition, data analysis and predictive modeling. In sensor design, deep learning can process and analyze a large number of complex nonlinear relationships between design parameters and performance. However, deep learning usually requires a large amount of data to train, and may fall into local optimal solution problems when optimizing design parameters. The Whale Optimization Algorithm (WOA) is a new optimization algorithm inspired by the predation behavior of whales in nature. It simulates the social behavior and unique hunting strategy of whales, showing excellent global search capabilities and efficient convergence performance, but lacks adaptability and detail processing capabilities for specific problems when used alone. Summary of the invention

[0004] In view of the problems of low design efficiency and low design accuracy of impact test sensors in the prior art, the present invention provides a reverse design method and device for a high-sensitivity impact test sensor, which uses deep learning to accurately predict the sensor performance. At the same time, the whale optimization algorithm is used to globally optimize the sensor design parameters to ensure that the sensor performance reaches the optimal state. By combining deep learning and the whale optimization algorithm, the design efficiency of the impact test sensor is improved and the design cycle is shortened. In addition, the problem that traditional design methods are prone to fall into local optimal solutions is solved, the global optimal ability and adaptability of the design are enhanced, and the reliability of the design results is improved.

[0005] To achieve the above object, according to a first aspect of the present invention, a method for reverse designing a high-sensitivity impact test sensor is provided, the method comprising:

[0006] Determine the structural geometric parameters of the impact test sensor unit cell, including ligament length, ligament thickness, middle circular node radius and horseshoe radius, and determine the corresponding structural equivalent strain;

[0007] Different unit cells are arrayed to obtain different impact test sensors, and the performance parameters of the sensors, including sensitivity, acceleration and peak pressure, are determined;

[0008] Constructing a sensor deep learning model, wherein the sensor deep learning model takes structural parameters of different impact test sensors as input and takes corresponding performance parameters as output; wherein the structural parameters include structural geometric parameters and their corresponding structural equivalent strains;

[0009] The target performance parameters of the impact test sensor are set, and the sensor deep learning model is combined with the whale optimization algorithm to reversely design the impact test sensor to obtain the optimal structural parameters corresponding to the set target performance parameters.

[0010] Furthermore, target performance parameters of the impact test sensor are set, and the impact test sensor is reversely designed using the whale optimization algorithm in combination with the sensor deep learning model to obtain the optimal structural parameters corresponding to the set target performance parameters, including:

[0011] Setting target performance parameters for impact test sensors;

[0012] Initialize the whale population, and each individual whale corresponds to the structural parameters of the impact test sensor;

[0013] The sensor deep learning model is used to obtain the performance parameters of each individual whale in the population, and the fitness value of each individual whale is calculated in combination with the set target performance parameters;

[0014] Select the best individual and update the position of whale individuals in the population, i.e. structural parameters, by surrounding prey, attacking with net bubbles and searching for food;

[0015] The fitness value is recalculated until the optimal structural parameters corresponding to the target performance parameters set for the impact test sensor are obtained.

[0016] Further, the fitness value is recalculated until the optimal structural parameters corresponding to the target performance parameters set for the impact test sensor are obtained, including:

[0017] When the current number of iterations t is less than the maximum number of iterations Max_iter, continue to calculate the fitness value of each whale individual in the population;

[0018] Determine whether it is better than the previous generation of individuals. If so, select the optimal individual; otherwise, use the previous generation of individuals for replacement; until t is greater than the maximum number of iterations, output the optimal structural parameters corresponding to the target performance parameters set by the impact test sensor.

[0019] Furthermore, the fitness of each individual whale is calculated in combination with the set target performance parameters, including:

[0020]

[0021] In the formula, F1 is the fitness, k is the number of target performance parameters, x is the position of individual whales in the population, that is, the structural parameter, and d i (x) is the target performance parameter, is the performance parameter predicted by the sensor deep learning model, and i represents the i-th parameter in the performance parameter.

[0022] Furthermore, the method further comprises:

[0023] The structural geometric parameters are normalized using the ligament length L: normalized ligament thickness α = t / L, normalized circular node radius γ = r / L, normalized horseshoe radius β = r0 / L; where t, r and r0 are the ligament thickness, the middle circular node radius and the horseshoe radius, respectively;

[0024] The calculation formula of structural equivalent strain is:

[0025]

[0026] Where ε is the equivalent strain of the structure, Δl is the compression displacement of the structure, and L is the ligament length;

[0027] The normalized ligament thickness α, normalized circular node radius γ, normalized horseshoe radius β and structural equivalent strain ε are taken as structural parameters of the impact test sensor and used as input of the sensor deep learning model.

[0028] Furthermore, the mapping relationship between the structural parameters and performance parameters of the impact test sensor is constructed through the multi-layer perceptron back propagation algorithm as follows:

[0029] y(GF,a,F m )=f[x(α,β,γ,ε)]

[0030] In the formula, GF is the sensitivity of the sensor, a is the acceleration, and F m is the peak pressure, x, y and f represent the mapping relationship between them, which is an implicit function expression.

[0031] Furthermore, the sensitivity GF, acceleration a and peak pressure F of the sensor are obtained through impact testing and finite element simulation.m ; The sensitivity calculation formula is:

[0032]

[0033] Wherein, GF is the sensitivity of the shock test sensor, ΔR0 is the resistance change during the shock test sensor process, and R0 is the initial resistance of the shock test sensor.

[0034] Furthermore, the method also includes:

[0035] Train and validate the sensor deep learning model; during training, automatically adjust the learning rate of each parameter:

[0036]

[0037] Where W new Represented as the next step weight, W old is the current weight, η is the learning rate, is the gradient of the loss function with respect to the weight;

[0038] Regularization parameter:

[0039]

[0040] In the formula, J0(W) is the original loss function, J(W) is the current loss function, λ is the regularization coefficient, and w i is the weight value;

[0041] The mean square error (MSE) is used as the loss function to evaluate the accuracy of the sensor deep learning model. The mean square error (MSE) of the loss function is defined as:

[0042]

[0043] In the formula, l is the loss, N is the amount of data, and y i is a true value, and is the corresponding predicted value, i represents the i-th data amount. Further, the whale population is initialized based on the cubic chaos map as follows:

[0044] x(n+1)=4x(n) 3 -3x(n)

[0045] -1≤x(n)≤1,n=1,2,3...

[0046] The chaos initialization steps are as follows:

[0047] (1) Search and randomly generate a d-dimensional individual x(n) in the space [-1, 1];

[0048] (2) Iterate the first individual generated according to the above formula to generate other individual chaotic sequences;

[0049] (3) After all whale individuals have been iterated, they are mapped to the original search space according to the following formula:

[0050]

[0051] Where: m d and l d is the upper and lower limits of each dimension of each individual x, X d is the chaotic variable mapped to the solution space.

[0052] According to a second aspect of the present invention, there is provided a high-sensitivity impact test sensor reverse design device for implementing any one of the above-mentioned high-sensitivity impact test sensor reverse design methods, the device comprising:

[0053] The structural parameter unit is used to determine the structural geometric parameters of the impact test sensor unit cell, including ligament length, ligament thickness, intermediate circular node radius and horseshoe radius, and determine the corresponding structural equivalent strain;

[0054] A performance parameter unit is used to array different unit cells to obtain different impact test sensors and determine the performance parameters of the sensors, including sensitivity, acceleration and peak pressure;

[0055] A deep learning unit, used to construct a sensor deep learning model, wherein the sensor deep learning model takes structural parameters of different impact test sensors as input and takes corresponding performance parameters as output; wherein the structural parameters include structural geometric parameters and their corresponding structural equivalent strains;

[0056] The reverse design unit is used to set the target performance parameters of the impact test sensor, and in combination with the sensor deep learning model, the impact test sensor is reversely designed using the whale optimization algorithm to obtain the optimal structural parameters corresponding to the set target performance parameters.

[0057] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0058] The present invention proposes a reverse design method and device for a high-sensitivity impact test sensor based on a deep learning-whale optimization algorithm, and combines deep learning and the whale optimization algorithm to perform reverse design of the impact test sensor: first, through the powerful data processing capability of deep learning, the relationship between sensor performance (acceleration, peak pressure, and sensitivity) and structural design parameters is accurately captured and simulated to achieve accurate prediction; then, combined with the whale optimization method, the selection of sensor structural design parameters is optimized to achieve the search for the global optimal solution, thereby enhancing the accuracy of the reverse design sensor.

[0059] In addition, this design method is universal and adaptable. It is not limited to this type of sensor structure design and can be widely used in various impact test sensor designs, providing a new idea for sensor design. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A flow chart of a reverse design method for a high-sensitivity impact test sensor provided by an embodiment of the present invention;

[0061] Figure 2 A sample curve diagram of representative acceleration, sensitivity, and peak pressure of an impact sensor measured based on experiments and finite element methods provided in an embodiment of the present invention;

[0062] Figure 3 A mapping relationship diagram between structural parameters of an impact test sensor provided by an embodiment of the present invention and structural sensitivity, acceleration, and peak pressure;

[0063] Figure 4 A graph showing the relationship between the loss of sensitivity, acceleration, peak pressure and the number of iterations during training and testing provided by an embodiment of the present invention;

[0064] Figure 5 A flow chart of reverse design of an impact test sensor based on a whale optimization algorithm provided by an embodiment of the present invention;

[0065] Figure 6 A graph of the loss function and the number of iterations of a reverse-designed impact test sensor based on a whale optimization algorithm in training and testing provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0067] The present invention provides an intelligent reverse design method and device for a high-sensitivity impact test sensor, which combines deep learning and a whale optimization algorithm to perform reverse design of the impact test sensor, aiming to utilize the powerful data processing and pattern recognition capabilities of deep learning and the excellent global search capabilities of WOA to achieve efficient optimization of the impact test sensor design, improve design efficiency and shorten the design cycle.

[0068] Embodiment 1

[0069] This embodiment provides a reverse design method for an impact test sensor based on a deep learning-whale optimization algorithm. Figure 1 The flow chart of the reverse design method of the shock test sensor based on the deep learning-whale optimization algorithm is given. It mainly includes determining the representative cell geometric parameters of the shock test sensor, arraying them into a shock test sensor, and then, Figure 2 As shown in the figure, a large sample finite element or experimental impact test is performed on it to obtain the relationship between its impact response (sensitivity, acceleration, peak pressure) and structural parameters (normalized thickness α = t / L, normalized horseshoe radius β = r0 / L, normalized circular node radius γ = r / L, structural equivalent strain ε). Based on this sample, the training set and test set are divided to construct a deep learning network model. Its deep learning network model is shown in the figure. Figure 3 As shown in the figure, it includes the characteristic parameter input layer, hidden layer and impact response regression output layer. The hidden layer gives the mapping relationship between the impact response and structural parameters of the model. The relationship between the loss of sensitivity, acceleration, peak pressure and the number of iterations during training and testing is shown in the figure. Figure 4 As shown in the figure, the prediction ability of the deep learning model is demonstrated. The model is given the target impact response, and the impact test sensor is reversely designed using the whale optimization algorithm to obtain its structural geometric size parameters. The whale optimization algorithm process is as follows Figure 5 As shown in the figure, when the iteration satisfies a certain relationship, the whale optimization algorithm uses its local convergence stability to obtain the optimal geometric parameters of the impact test sensor. Figure 6 The curves of loss function and iteration number of the impact test sensor reversely designed based on the whale optimization algorithm during training and testing are given, showing that the impact test sensor reversely designed based on the whale optimization algorithm has strong stability and robustness.

[0070] The specific steps of this method are as follows:

[0071] S1. Determine the geometric parameters of a representative unit cell structure of the impact test sensor, wherein the geometric parameters of the structure include: ligament length L, ligament thickness t, middle circular node radius r, and horseshoe radius r0.

[0072] In order to facilitate the input of parameters in the deep learning model, they are normalized with the ligament layer L: normalized thickness α = t / L, normalized horseshoe radius β = r0 / L, normalized circular node radius γ = r / L, and structural equivalent strain ε.

[0073] The equivalent strain calculation formula of the impact test sensor structure is:

[0074]

[0075] Where ε is the equivalent strain of the structure, Δl is the compressive displacement of the structure, and L is the ligament length of the impact test sensor.

[0076] S2. According to the geometric parameters of the representative unit cell structure of the selected impact test sensor, the structural unit cells are arrayed to obtain the impact test sensor, copper is used as the substrate of the sensor, and the experimental sample is printed based on the metal 3D printing technology. Then, the impact test and finite element simulation of the large sample sensors with different geometric dimensions are carried out. The impact test adopts the drop hammer test, and the finite element simulation adopts ABAQUS software. The finite element calculation steps include material parameter setting (copper is used as the substrate, its elastic model is 60GPa, and Poisson's ratio is 0.3), load setting (impact velocity is 10m / s), mesh division (structural mesh size is 5mm) and other steps, and the sensitivity GF, acceleration a, and peak pressure F of the impact test sensor under different geometric size parameters are obtained. m size, thereby building a deep learning database.

[0077] In this step, some representative unit cell structure geometric parameters are selected, and then the impact test sensor is obtained by array, and then the performance parameters of the sensor are obtained through impact testing and finite element simulation to obtain a sample set.

[0078] The sensitivity calculation formula of the impact test sensor is:

[0079]

[0080] Wherein, GF is the sensitivity of the sensor during the shock test, ΔR0 is the resistance change during the shock test of the sensor, and R0 is the initial resistance of the sensor during the shock test.

[0081] S3. According to the sample set obtained from the experiment and finite element simulation process, the sample set is divided into a training set and a test set. 80% of the sample set is used for training and 20% of the sample set is used for testing. The structural parameters of the sensor (normalized thickness α = t / L, normalized horseshoe radius β = r0 / L, normalized circular node radius γ = r / L, strain ε) are used as input, and the sensitivity GF, acceleration a and peak pressure F of the sensor are used as input. mThe size is used as the output to train the deep learning model, and the multi-layer perceptron back propagation algorithm is used to simulate the response of neurons to stimulation using the Sigmoid continuous function to construct the impact test sensor structure parameters and sensor sensitivity GF, acceleration a, peak pressure F in the training set. m The mapping relationship between sizes.

[0082] The mapping relationship constructed in the multi-layer perceptron back propagation algorithm is:

[0083] y(GF,a,F m )=f[x(α,β,γ,ε)] (3)

[0084] In the formula, GF is the sensitivity of the sensor, a is the acceleration, and F m is the peak pressure. Among them, x, y, and f represent the mapping relationship between them, which is an implicit function expression.

[0085] The Sigmoid function is defined as:

[0086]

[0087] S4. Based on the sensor deep learning model relationship of the impact test constructed in the training set, the data in the test set is used for verification. First, determine the initial hyperparameter configuration, such as learning rate, batch size, network structure parameters, regularization parameters and optimizer selection, where the learning rate is set to 0.001, the amount of data per batch is 20,000, and the maximum number of iterations is 200. In order to automatically adjust the learning rate of each parameter during the training process, the Adam optimizer is used. The mean square error (MSE) is used as the loss function to evaluate the accuracy of the deep learning model. When the loss function in the training set reaches the minimum value, the impact response data in the test set is called to further evaluate the training model's ability to predict the sensor sensitivity, acceleration and peak pressure between the impact test sensor structure.

[0088] The learning rate is calculated as:

[0089]

[0090] Where W new Represented as the next step weight, W old is the current weight, η is the learning rate, is the gradient of the loss function with respect to the weights.

[0091] The regularization parameter is calculated as:

[0092]

[0093] In the formula, J0(W) is the original loss function, J(W) is the current loss function, λ is the regularization coefficient, and w i is the weight value.

[0094] The mean square error (MSE) of the loss function is defined as:

[0095]

[0096] Where N is the amount of data in the database, y i is a true value, and is the corresponding predicted value.

[0097] S5. According to the constructed deep learning model, for a given target response function of the impact test sensor (i.e., the relationship between the structural geometric parameters and the sensor sensitivity, acceleration, peak pressure, etc.), the whale optimization algorithm is implemented through steps such as surrounding prey, attacking with a bubble, searching for food, etc.

[0098] When the current iteration number t is less than the maximum iteration number Max_iter, continue to calculate the fitness value of the individuals in the population. On this basis, determine whether it is better than the previous generation of individuals. If so, select the best individual, otherwise, use the previous generation of individuals to replace it. Until t is greater than the maximum iteration number, output the optimal impact test sensor structure parameters, thereby building a model of the whale optimization algorithm.

[0099] The implementation process of the whale optimization algorithm includes the processes of surrounding prey, bubble attack, and searching for food. The process is as follows:

[0100] (1) Encirclement and predation:

[0101] In the determination process, once the best performing individual in the population, i.e. the optimal solution, has been identified, the population will converge towards the optimal solution according to formula (8):

[0102] X(t+1)=X*(t)-A*D (8)

[0103] D=|C*X*(t)-X(t)| (9)

[0104] in is the position vector of the prey, t is the current iteration number, X is the position vector of the individual, where A and C are coefficient vectors, which are determined by the following equations (10) and (11):

[0105] A=2ar1-a (10)

[0106] C=2r2 (11)

[0107]

[0108] Where: r1 and r2 are random numbers between [0, 1], a is a convergence factor that decreases linearly from 2 to 0, and Max_iter represents the maximum number of iterations.

[0109] (2) Internet Bubble Attack:

[0110] Contraction and encirclement: This is achieved by reducing the convergence factor a in equation (12). The value of a decreases with the increase of the number of iterations, so that the individual whales continuously update their position coefficients to approach the prey and complete the encirclement.

[0111] Spiral update: When searching for prey, individual whales use a spiral upward strategy to adjust their position relative to the prey. Its mathematical model is shown in formula (13):

[0112] X(t+1)=D'*ebl*cos(2πl)+X*(t) (13)

[0113] D ′ =|X*(t)-X(t)| (14)

[0114] Where: D' represents the distance between the prey and the individual, b is a constant that defines the spiral search shape, and l is a random number in [-1, 1]. When |A| ≤ 1, the bubble net attack is used, and the probability of executing shrinkage and spiral update is 50% each. When p < 0.5, shrinkage is executed, otherwise spiral update is executed. Its mathematical model is shown in formula (15):

[0115]

[0116] (3) Searching for food:

[0117] In contrast to the shrinking and encircling stage, when |A|>1, an individual is randomly selected from the population to find the best solution, and no longer follows the reference whale. The mathematical model is as follows:

[0118]

[0119] Where: Xrand is the position vector of a randomly selected individual.

[0120] Among them, the fitness value of each individual whale is calculated as follows:

[0121]

[0122] Wherein, F1 is fitness, k is the number of target performance parameters, and in this embodiment, k is 3. x is the position vector of individual whales in the population, i.e., structural parameters, including normalized thickness α=t / L, normalized horseshoe radius β=r0 / L, normalized circular node radius γ=r / L, strain ε. di(x) is the target performance parameter, including the sensitivity GF, acceleration a, and peak pressure Fm of the sensor. is the performance parameter predicted by the sensor deep learning model, and i represents the i-th parameter in the performance parameter.

[0123] S6. Based on the constructed model of deep learning-whale optimization algorithm, by initializing the population based on cubic chaos mapping, completing the reverse construction between the target response and the structural geometric parameters, verifying the data not tested in the experiment and simulation, adjusting the initialization of the whale population and fitness, and further optimizing the parameters of the reverse design model. Iteratively obtain the optimal solution between structural sensitivity, acceleration and peak pressure, and optimize the sensor structure design parameters.

[0124] Among them, the reverse design of the sensor structure based on the whale optimization algorithm involves initializing the population based on cubic chaos mapping: the position of the initial solution in the algorithm has a direct impact on the optimization effect and convergence speed. In the initialization stage of the algorithm, pseudo-random numbers are usually used to generate individual positions. Although this can ensure the wide distribution of the population in the solution space, it will inevitably lead to uneven distribution of the population, and some areas may be too dense. In order to improve this situation, chaos theory is introduced, and its pseudo-randomness, non-periodicity and global traversal properties are used to enable the population to achieve a more uniform distribution in the search space. The population is initialized using the cubic chaos mapping method, as shown in formula (17):

[0125]

[0126] The chaos initialization steps are as follows:

[0127] (1) A d-dimensional individual of [-1, 1] is randomly generated in the search space; in this embodiment, it is a 4-dimensional individual, which is normalized thickness α=t / L, normalized horseshoe radius β=r0 / L, normalized circular node radius γ=r / L, strain ε;

[0128] (2) Iterate the first individual generated according to formula (17) to generate other individual chaotic sequences;

[0129] (3) After all whale individuals have been iterated, they are mapped to the original search space according to formula (18).

[0130]

[0131] Where: u d , ld is the upper and lower limits of each dimension for each individual, X d is the chaotic variable mapped to the solution space.

[0132] The present invention proposes a reverse design method for impact test sensors based on deep learning-whale optimization algorithm. Through the powerful data processing capability of deep learning, it can accurately capture and simulate the relationship between sensor performance (acceleration, peak pressure and sensitivity) and structural design parameters, and achieve accurate prediction (its prediction accuracy is about 98%). Combined with the whale optimization method, the present invention optimizes the selection of sensor structural design parameters, realizes the search for the global optimal solution, and enhances the accuracy of the reverse design sensor (its reverse prediction accuracy reaches about 92%).

[0133] Embodiment 2

[0134] This embodiment provides a high-sensitivity impact test sensor reverse design device for implementing any one of the above-mentioned high-sensitivity impact test sensor reverse design methods, the device comprising:

[0135] The structural parameter unit is used to determine the structural geometric parameters of the impact test sensor unit cell, including ligament length, ligament thickness, intermediate circular node radius and horseshoe radius, and determine the corresponding structural equivalent strain;

[0136] A performance parameter unit is used to array different unit cells to obtain different impact test sensors and determine the performance parameters of the sensors, including sensitivity, acceleration and peak pressure;

[0137] A deep learning unit, used to construct a sensor deep learning model, wherein the sensor deep learning model takes structural parameters of different impact test sensors as input and takes corresponding performance parameters as output; wherein the structural parameters include structural geometric parameters and their corresponding structural equivalent strains;

[0138] The reverse design unit is used to set the target performance parameters of the impact test sensor, and in combination with the sensor deep learning model, the impact test sensor is reversely designed using the whale optimization algorithm to obtain the optimal structural parameters corresponding to the set target performance parameters.

[0139] In summary, the present invention aims to solve the problems of low design efficiency and low design accuracy of existing impact test sensors. By combining deep learning and the whale optimization algorithm, the design process is automated and intelligent, the design efficiency is improved and the design cycle is shortened. The present invention uses deep learning to accurately predict sensor performance. At the same time, the whale optimization algorithm is used to globally optimize the design parameters to ensure that the sensor performance reaches the optimal state. In addition, the present invention solves the problem that traditional design methods are prone to fall into local optimal solutions, enhances the global optimality and adaptability of the design, and improves the reliability of the design results.

[0140] It should be noted that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0141] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0142] It will be easily understood by those skilled in the art that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A reverse design method for a high-sensitivity impact test sensor, characterized in that: The method includes: Determine the structural geometric parameters of the impact test sensor unit cell, including ligament length, ligament thickness, middle circular node radius and horseshoe radius, and determine the corresponding structural equivalent strain; Different unit cells are arrayed to obtain different impact test sensors, and the performance parameters of the sensors, including sensitivity, acceleration and peak pressure, are determined; Constructing a sensor deep learning model, wherein the sensor deep learning model takes structural parameters of different impact test sensors as input and takes corresponding performance parameters as output; wherein the structural parameters include structural geometric parameters and their corresponding structural equivalent strains; Setting target performance parameters of the impact test sensor, combining the sensor deep learning model, and using the whale optimization algorithm to reversely design the impact test sensor to obtain optimal structural parameters corresponding to the set target performance parameters; Among them, the target performance parameters of the impact test sensor are set, and the impact test sensor is reversely designed using the whale optimization algorithm in combination with the sensor deep learning model to obtain the optimal structural parameters corresponding to the set target performance parameters, including: Setting target performance parameters for impact test sensors; Initialize the whale population, and each individual whale corresponds to the structural parameters of the impact test sensor; The sensor deep learning model is used to obtain the performance parameters of each individual whale in the population, and the fitness value of each individual whale is calculated in combination with the set target performance parameters; Select the best individual and update the position of whale individuals in the population, i.e. structural parameters, by surrounding prey, attacking with net bubbles and searching for food; Recalculate the fitness value until the optimal structural parameters corresponding to the target performance parameters set by the impact test sensor are obtained; The method further includes: The structural geometric parameters are normalized using the ligament length L: normalized ligament thickness α = t / L, normalized circular node radius γ = r / L, normalized horseshoe radius β = r0 / L; where t, r and r0 are the ligament thickness, the middle circular node radius and the horseshoe radius, respectively; The calculation formula of structural equivalent strain is: Where ε is the equivalent strain of the structure, Δl is the compression displacement of the structure, and L is the ligament length; The normalized ligament thickness α, the normalized circular node radius γ, the normalized horseshoe radius β and the structural equivalent strain ε are used as the structural parameters of the impact test sensor and as the input of the sensor deep learning model; The mapping relationship between the structural parameters and performance parameters of the impact test sensor is constructed through the multi-layer perceptron back propagation algorithm as follows: y(GF,a,F m )=f[x(α,β,γ,ε)] Where GF is the sensitivity of the impact test sensor, a is the acceleration, and F m is the peak pressure, x, y and f represent the mapping relationship between them, which is an implicit function expression.

2. The reverse design method of a high-sensitivity impact test sensor according to claim 1, characterized in that: Recalculate the fitness value until the optimal structural parameters corresponding to the target performance parameters set by the impact test sensor are obtained, including: When the current number of iterations t is less than the maximum number of iterations Max_iter, continue to calculate the fitness value of each whale individual in the population; Determine whether it is better than the previous generation of individuals. If so, select the optimal individual; otherwise, use the previous generation of individuals for replacement; until t is greater than the maximum number of iterations, output the optimal structural parameters corresponding to the target performance parameters set by the impact test sensor.

3. The reverse design method of a high-sensitivity impact test sensor according to claim 1, characterized in that: The fitness of each individual whale is calculated based on the target performance parameters, including: In the formula, F1 is the fitness, k is the number of target performance parameters, x is the position of individual whales in the population, that is, the structural parameter, and d i (x) is the target performance parameter, is the performance parameter predicted by the sensor deep learning model, and i represents the i-th parameter in the performance parameter.

4. The reverse design method of a high-sensitivity impact test sensor according to claim 1, characterized in that: The sensitivity GF, acceleration a and peak pressure F of the impact test sensor are obtained through impact test and finite element simulation. m ; The sensitivity calculation formula is: Wherein, GF is the sensitivity of the shock test sensor, ΔR0 is the resistance change during the shock test sensor, and R0 is the initial resistance of the shock test sensor.

5. The reverse design method of a high-sensitivity impact test sensor according to claim 1, characterized in that: The method further includes: Train and validate the sensor deep learning model; during training, automatically adjust the learning rate of each parameter: Where W new Represented as the next step weight, W old is the current weight, η is the learning rate, is the gradient of the loss function with respect to the weight; Regularization parameter: In the formula, J0(W) is the original loss function, J(W) is the current loss function, λ is the regularization coefficient, and w i is the weight value; The mean square error (MSE) is used as the loss function to evaluate the accuracy of the sensor deep learning model. The mean square error (MSE) of the loss function is defined as: In the formula, l is the loss, N is the amount of data, and y i is a true value, and is the corresponding predicted value, and i represents the i-th data amount.

6. The reverse design method of a high-sensitivity impact test sensor according to claim 1, characterized in that: Initialize the whale population based on the cubic chaos map as follows: x(n+1)=4x(n) 3 -3x(n) -1≤x(n)≤1,n=1,2,3... The chaos initialization steps are as follows: (1) Search and randomly generate a d-dimensional individual x(n) in the space [-1, 1]; (2) Iterate the first individual generated according to the above formula to generate other individual chaotic sequences; (3) After all whale individuals have been iterated, they are mapped to the original search space according to the following formula: Where: m d and l d is the upper and lower limits of each dimension of each individual x, X d is the chaotic variable mapped to the solution space.

7. A high-sensitivity impact test sensor reverse design device for implementing the high-sensitivity impact test sensor reverse design method according to any one of claims 1 to 6, characterized in that: The device includes: The structural parameter unit is used to determine the structural geometric parameters of the impact test sensor unit cell, including ligament length, ligament thickness, intermediate circular node radius and horseshoe radius, and determine the corresponding structural equivalent strain; A performance parameter unit is used to array different unit cells to obtain different impact test sensors and determine the performance parameters of the sensors, including sensitivity, acceleration and peak pressure; A deep learning unit, used to construct a sensor deep learning model, wherein the sensor deep learning model takes structural parameters of different impact test sensors as input and takes corresponding performance parameters as output; wherein the structural parameters include structural geometric parameters and their corresponding structural equivalent strains; The reverse design unit is used to set the target performance parameters of the impact test sensor, and in combination with the sensor deep learning model, the impact test sensor is reversely designed using the whale optimization algorithm to obtain the optimal structural parameters corresponding to the set target performance parameters.

Images