Three-dimensional space acoustic detection method based on COMSOL simulation

Abstract

The invention relates to the technical field of acoustic detection, and discloses a three-dimensional space acoustic detection method based on COMSOL simulation, and the method comprises the steps: S1, building a physical data factory, generating a three-dimensional sound pressure field data set through a physical constraint generative adversarial network, and enabling the network to comprise a verification layer for executing the verification of a sound wave propagation physical rule; s2, model training is detected in real time, and the mapping relation between acoustic signals and defect parameters is based on the three-dimensional sound pressure field data set; s3, three-dimensional space signal acquisition; s4, identifying defect parameters; s5, performing physical auxiliary verification; and S6, generating a three-dimensional detection report. According to the method, an acoustic physical verification mechanism is deeply integrated through the generative adversarial network, a Helmholtz equation constraint and boundary energy absorption criterion is forced to be executed in the data generation process, an acoustic data set with millimeter-level spatial resolution is constructed, a high-fidelity training basis strictly following a wave equation is provided for three-dimensional defect recognition, and the defect recognition accuracy is improved. And the accuracy of acoustic detection and recognition is improved.

Description

Technical Field

[0001] This invention relates to the field of acoustic testing technology, and in particular to a three-dimensional spatial acoustic testing method based on COMSOL simulation. Background Technology

[0002] With the increasing demands for precision in internal defect detection in high-precision equipment manufacturing, especially in the production of aerospace and energy equipment, it is necessary to achieve three-dimensional spatial positioning and quantitative analysis of micro-defects (such as microcracks and pores) inside complex components. In the existing acoustic detection system, how to establish a deterministic mapping relationship between the physical laws of sound wave propagation in three-dimensional space and defect characteristic parameters, and ensure millimeter-level depth positioning accuracy in industrial scenarios, has become a core bottleneck restricting the reliability of high-value-added components. However, with the help of the high-fidelity numerical COMSOL simulation platform, a three-dimensional sound field physical model containing material acoustic parameters (sound velocity, density, attenuation coefficient) can be constructed to accurately simulate the scattering field distribution of the interaction between sound waves and defects. Traditional technologies mostly rely on full-size COMSOL simulation.

[0003] However, current technology simplifies acoustic model generation training data and ignores the Helmholtz equation and boundary absorption effect, resulting in problems such as blurred depth localization, deviation from the physical laws of data, and large errors in micro-defect identification. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a three-dimensional spatial acoustic detection method based on COMSOL simulation, which aims to improve the problems of fuzzy depth positioning, deviation of data physical laws, and large errors in micro-defect identification.

[0005] In a first aspect, the present invention provides the following technical solution: a three-dimensional spatial acoustic detection method based on COMSOL simulation, comprising:

[0006] S1. Physical data factory construction: A three-dimensional sound pressure field dataset is generated through a physical constraint generative adversarial network. The network includes a verification layer that performs verification of the physical rules for sound wave propagation.

[0007] S2. Real-time detection model training: Based on the mapping relationship between acoustic signals and defect parameters in the three-dimensional sound pressure field dataset, a three-dimensional convolutional network model is trained.

[0008] S3, Three-dimensional spatial signal acquisition, which acquires the acoustic signals of the object under test through a spatially distributed multi-channel acoustic sensor array;

[0009] S4. Defect parameter identification: Input the collected signals into the training model and output three-dimensional spatial defect parameters.

[0010] S5. Physical auxiliary verification: COMSOL is called to reconstruct the sound field of the defect area in the output.

[0011] S6. Generate a 3D inspection report, which maps defect parameters to a 3D spatial coordinate system and outputs the location results and a material integrity report.

[0012] By adopting the above technical solution, a physical rule-driven generative adversarial network is used to construct a high-fidelity three-dimensional sound pressure field dataset. A lightweight three-dimensional convolutional network model is trained to establish the mapping relationship between acoustic signals and spatial defect parameters. Real-time sound wave signals are collected based on a multi-channel sensor array, and the three-dimensional defect parameters are directly output by inputting them into the training model. Then, the COMSOL physics engine is called to perform local sound field reconstruction in the prediction area. The dynamic calibration results are verified by dual verification of sound pressure peak value and main frequency offset. Finally, the parameters are mapped to the CAD coordinate system to generate a three-dimensional spherical solid model and output a depth risk quantification report.

[0013] Preferably, step S1 specifically includes the following steps:

[0014] S101, Defect parameter input: Input a vector containing three-dimensional coordinates of the location, defect size and material code into the generator, and output three-dimensional sound pressure field data (X, Y, Z). The material code includes digital identifiers for steel, aluminum and composite materials.

[0015] S102, Physical rule verification: Perform Helmholtz equation residual verification and boundary absorption verification on the output sound pressure field;

[0016] S103, Data generation control: Reinitialize the generation process when the verification does not meet the threshold conditions.

[0017] Preferably, the residual verification of the Helmholtz equation in S102 This is used to verify whether the sound pressure field data p satisfies the physical constraints of the wave equation, where, For the Laplace operator of the sound pressure field, k 2 p is the product of the square of the wave number and the sound pressure level;

[0018] The boundary absorption verification requirement is: the sound pressure amplitude in the perfectly matched layer ≤ 10% × max(p 全局 ), where p 全局 The sound pressure level in the entire computational domain.

[0019] Preferably, step S2 specifically includes the following steps:

[0020] S201, Loading 3D Sound Pressure Field Data: Using the verified 3D sound pressure field data p generated in S1. gen ∈R t×X×Y×Z As the training input, t is the number of time or frequency sampling points, X and Y are the number of spatial grids in the horizontal direction, and Z is the number of spatial grids in the depth direction;

[0021] S202, Spatial coordinate mapping training, establishing a regression mapping from the sound pressure field to the coordinates of the defect center:

[0022] (x, y, z) = F(p) gen ); where F is the mapping function for a lightweight 3D convolutional network;

[0023] Training objective: Minimize the spatial location error ||(x, y, z) - (x 真 y 真 , z 真 )||2, where (x 真 y 真 , z 真 () represents the three-dimensional coordinates of the defect location input by S101;

[0024] S203, Depth and accuracy compliance verification, when the model satisfies the following on the validation set: δ z =|zz 真 Training will be terminated if the thickness is ≤0.5mm.

[0025] Preferably, step S4 specifically includes the following steps:

[0026] S401. Convert the multi-channel acoustic signal into a three-dimensional sound pressure field distribution p(x, y, z, t), which is consistent with the coordinate system of the defect parameters in S101.

[0027] S402: Input the 3D convolutional network trained from p to S2, and output the coordinates (x, y, z) and size of the defect center;

[0028] S403, 3D positioning optimization, calculation of depth reliability coefficient:

[0029] Among them, z 训练 Let δ be the median depth of the S2 training model on the validation set. 训练 The depth error threshold (0.5mm) is defined for S203;

[0030] When R z If the value is ≥0.9, the positioning result is confirmed to be valid, and the coordinate parameters of the defect center are output.

[0031] Preferably, step S5 specifically includes the following steps:

[0032] S501. Based on the defect center coordinates and dimensions output by S4, construct a cubic computational domain centered at (x,y,z).

[0033] S502. Call COMSOL within the computational domain to perform transient acoustic simulation. Set the boundary conditions to be the surface free displacement and the outer layer perfectly matched. Input the incident sound wave signal collected by the S3 acoustic sensor and output the three-dimensional position coordinates, equivalent size and type classification results of the defect.

[0034] Preferably, step S6 specifically includes the following steps:

[0035] S601. Transform the output defect coordinates (x, y, z) to the coordinate system of the CAD model of the object under test, and output the precise position through spatial transformation relationship;

[0036] S602. Using the defect coordinates as the center, generate a spherical solid model through the dimension parameter d output by S4, and label the defect type with color classification based on S4.

[0037] S603, Calculate the depth risk coefficient: the percentage of the maximum material thickness at depth;

[0038] Size risk factor: The percentage of size to the critical defect size of the material, outputting a comprehensive risk index.

[0039] Secondly, the present invention provides the following technical solution: a three-dimensional spatial acoustic detection system based on COMSOL simulation, the system comprising:

[0040] The Physical Data Factory module generates a 3D acoustic pressure field dataset through a physical constraint generative adversarial network and automatically verifies the Helmholtz equation residuals and boundary absorption rates.

[0041] The real-time detection model training module trains a three-dimensional convolutional network based on sound pressure field data to establish a mapping relationship between acoustic signals and spatial defect parameters;

[0042] A three-dimensional signal acquisition module is used to synchronously acquire multi-channel acoustic signals through a spatially distributed array of acoustic sensors;

[0043] The defect parameter identification module is used to input the collected signals into a pre-trained 3D convolutional network model and output the 3D coordinates, size, and type classification of the defects.

[0044] The COMSOL physical verification module is used to call the COMSOL engine to reconstruct the sound field in the predicted defect area and perform sound pressure peak matching and main frequency offset verification.

[0045] The 3D report generation module is used to map defect parameters to the CAD coordinate system, generate a 3D sphere model, and output a risk quantification report.

[0046] Thirdly, the invention provides the following technical solution: a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned three-dimensional spatial acoustic detection method based on COMSOL simulation.

[0047] Fourthly, the present invention provides the following technical solution: a readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the above-mentioned three-dimensional spatial acoustic detection method based on COMSOL simulation.

[0048] The present invention has the following beneficial effects:

[0049] 1. In this invention, a deep integrated acoustic physics verification mechanism using generative adversarial networks is employed to enforce Helmholtz equation constraints and boundary energy absorption criteria during data generation. This constructs an acoustic dataset with millimeter-level spatial resolution, providing a high-fidelity training foundation that strictly follows the wave equation for three-dimensional defect identification and improving the accuracy of acoustic detection and identification.

[0050] 2. In this invention, based on the construction of a local computational domain and driven by measured acoustic signals, COMSOL physical field reconstruction is performed in the predicted defect area. The dynamic verification of coordinates and dimensions is achieved through cross-dimensional matching of sound pressure peak positioning and frequency shift characteristics. A three-dimensional spatial parameter self-calibration system is established, and the positioning drift is eliminated by utilizing the physical nature of sound wave scattering.

[0051] 3. In this invention, a dyed three-dimensional solid model is generated by mapping the defect coordinates to the CAD design coordinate system. The comprehensive risk index is output by integrating the depth ratio and the critical size factor, thereby realizing millimeter-level positioning of industrial-grade three-dimensional defects, closed-loop verification of physical rules, and visualized risk decision-making. Attached Figure Description

[0052] Figure 1 This is a flowchart of the three-dimensional spatial acoustic detection method based on COMSOL simulation proposed in this invention;

[0053] Figure 2 This is an architecture diagram of the three-dimensional spatial acoustic detection system based on COMSOL simulation proposed in this invention. Detailed Implementation

[0054] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Example 1

[0056] Reference Figure 1 In the first embodiment of the present invention, the present invention provides a three-dimensional spatial acoustic detection method based on COMSOL simulation, comprising:

[0057] S1. Physical data factory construction: A three-dimensional sound pressure field dataset is generated through a physical constraint generative adversarial network. The network includes a verification layer that performs verification of the physical rules for sound wave propagation.

[0058] S2. Real-time detection model training: Based on the mapping relationship between acoustic signals and defect parameters in the three-dimensional sound pressure field dataset, a three-dimensional convolutional network model is trained.

[0059] S3, Three-dimensional spatial signal acquisition, which acquires the acoustic signals of the object under test through a spatially distributed multi-channel acoustic sensor array;

[0060] S4. Defect parameter identification: Input the collected signals into the training model and output three-dimensional spatial defect parameters.

[0061] S5. Physical auxiliary verification: COMSOL is called to reconstruct the sound field of the defect area in the output.

[0062] S6. Generate a 3D inspection report, which maps defect parameters to a 3D spatial coordinate system and outputs the location results and a material integrity report.

[0063] Specifically, S1 constructs a data factory through a physically constrained generative adversarial network. Its generator receives an input vector containing the three-dimensional location, size, and material properties of defects and outputs three-dimensional sound pressure field data. The discriminator integrates a physical verification layer to perform sound wave propagation equation verification and boundary absorption verification, ensuring that the output data strictly follows the physical laws of sound wave propagation. The generated three-dimensional sound pressure field has a spatial resolution of 0.1 mm, providing a physically realistic training basis for high-precision three-dimensional acoustic detection.

[0064] S2 establishes a direct mapping between sound pressure field data and spatial defect parameters through a three-dimensional convolutional network model. It takes a multi-dimensional sound pressure tensor (containing time / frequency and three-dimensional spatial grid information) as input, extracts the spatiotemporal joint features of sound wave propagation through a three-dimensional convolutional kernel, and outputs the three-dimensional coordinates of the defect center, equivalent diameter and type classification simultaneously. Furthermore, it optimizes the spatial positioning accuracy and classification accuracy through a composite loss function. After training, the model achieves high-precision end-to-end conversion of acoustic signals to three-dimensional defect parameters, providing a real-time inference core for online three-dimensional spatial acoustic detection.

[0065] S3 acquires the three-dimensional sound field signal of the object under test through a spatially distributed multi-channel acoustic sensor array. The piezoelectric ceramic sensor array arranged on the spherical surface synchronously receives the sound wave signal with precise spatial spacing. The signal is converted into a time-domain sound pressure sequence by the multi-channel synchronous acquisition system and reconstructed into a three-dimensional sound pressure tensor. This tensor completely preserves the propagation characteristics and spatiotemporal correlation of the sound wave in three-dimensional space, providing a sound field input with spatial dimension for subsequent defect parameter identification.

[0066] S4 inputs the collected multi-channel sound pressure tensor into a pre-trained three-dimensional convolutional network model. By extracting the feature correlation of acoustic signals in the spatial and temporal dimensions, it directly outputs spatial defect parameters containing the three-dimensional coordinates of the defect center, equivalent diameter, and type classification (crack / pore / corrosion), realizing the intelligent conversion from the original acoustic signal to three-dimensional quantified defect information.

[0067] S5 calls the COMSOL simulation engine to construct a local three-dimensional computational domain based on the output defect spatial coordinates and dimensions, performs high-fidelity sound field reconstruction and verifies the physical consistency of the predicted parameters. Through sound pressure peak position matching, main frequency offset feature comparison and sound energy focusing intensity analysis, it realizes secondary verification and closed-loop correction of spatial defect parameters to ensure that the three-dimensional detection results conform to the physical laws of sound wave propagation.

[0068] S6 maps the identified and verified defect parameters to the three-dimensional design coordinate system of the object under test, achieving precise spatial alignment. Based on the defect center coordinates and equivalent diameter, it constructs a three-dimensional spherical solid model, adds color markings based on the defect type, and finally outputs a material integrity analysis report containing a three-dimensional defect location image, depth risk coefficient, and size risk coefficient, providing a visual basis for engineering decisions.

[0069] S1 specifically includes the following steps:

[0070] S101, Defect parameter input: Input a vector containing the three-dimensional coordinates of the location, the defect size and the material code into the generator, and output three-dimensional sound pressure field data (X, Y, Z). The material code includes the digital identifier of steel, aluminum and composite materials.

[0071] S102, Physical rule verification: Perform Helmholtz equation residual verification and boundary absorption verification on the output sound pressure field;

[0072] S103, Data generation control: Reinitialize the generation process when the verification does not meet the threshold conditions.

[0073] Specifically, S101 inputs a digitally encoded vector containing three-dimensional spatial coordinates, defect geometry, and material properties to the generator, accurately defining the acoustic simulation scene. Based on the input parameters, it automatically outputs three-dimensional sound pressure field data with spatial dimensions (grid distribution in the X, Y, and Z directions). The material encoding system (steel / aluminum / composite materials) ensures accurate mapping of the physical properties of the medium.

[0074] S102 performs Helmholtz equation propagation law verification and boundary absorption characteristic verification on the generated three-dimensional sound pressure field, forces the sound wave behavior to follow the essential law of the wave equation and suppresses the false reflection effect. Through the dual-core mechanism, it ensures the physical rationality of the sound pressure field data, eliminates invalid data that violates the basic principles of acoustics, and makes the output dataset physically complete enough to be directly used in high-precision detection models.

[0075] When physical verification fails to meet the standards, the S103 automatically triggers a data regeneration process. It intelligently adjusts network parameters or input vectors according to the residual anomaly type, establishes a closed-loop quality control system, and ensures from the source that each three-dimensional sound pressure field meets the constraints of the sound wave propagation equation and the boundary absorption requirements. Finally, it outputs a high-fidelity dataset with a spatial resolution of millimeters, providing zero-error training samples for three-dimensional acoustic detection.

[0076] Helmholtz equation residual verification in S102 This is used to verify whether the sound pressure field data p satisfies the physical constraints of the wave equation, where, For the Laplace operator of the sound pressure field, k 2 p is the product of the square of the wave number and the sound pressure level;

[0077] Boundary absorption verification requirement: Sound pressure amplitude in the perfectly matched layer ≤ 10% × max(p 全局 ), where p 全局 The sound pressure level in the entire computational domain.

[0078] Specifically, by verifying whether the generated three-dimensional sound pressure field strictly satisfies the physical essence of the sound wave equation, the propagation behavior of sound waves is forced to conform to the laws of energy conservation and phase propagation in the medium, eliminating non-physical interpretations (such as spontaneous sound energy abrupt changes or phase velocity exceeding limits), providing a physically realistic training basis for the defect identification model. Boundary absorption verification suppresses false reflected waves by quantifying the boundary sound energy absorption rate, eliminates sound wave reflection artifacts at the boundary of the computational domain, ensures that the attenuation behavior of the three-dimensional sound field data at the spatial edge is consistent with that of the open space, and improves the positioning accuracy in the depth direction.

[0079] S2 specifically includes the following steps:

[0080] S201, Loading 3D Sound Pressure Field Data: Using the verified 3D sound pressure field data p generated in S1. gen ∈R t×X×Y×Z As the training input, t is the number of time or frequency sampling points, X and Y are the number of spatial grids in the horizontal direction, and Z is the number of spatial grids in the depth direction;

[0081] S202, Spatial coordinate mapping training, establishing a regression mapping from the sound pressure field to the coordinates of the defect center:

[0082] (x, y, z) = F(p) gen ); where F is the mapping function for a lightweight 3D convolutional network;

[0083] Training objective: Minimize the spatial location error ||(x, y, z) - (x 真 y 真 ,z 真 )||2, where (x 真y 真 ,z 真 () represents the three-dimensional coordinates of the defect location input by S101;

[0084] S203, Depth and accuracy compliance verification, when the model satisfies the following on the validation set: δ z =|zz 真 Training will be terminated if the thickness is ≤0.5mm.

[0085] Specifically, S201 loads physically verified three-dimensional sound pressure field data, transforms the propagation characteristics of sound waves in three-dimensional space into structured input, and constructs a training sample set with spatial resolution capabilities. The number of grids Z in the depth direction explicitly represents the sound wave penetration characteristics, providing a key physical carrier for depth coordinate regression and supporting the model in learning the defect scattering laws in three-dimensional space.

[0086] S202 establishes an end-to-end mapping from the sound pressure field to the coordinates (x, y, z) of the defect center through a lightweight 3D convolutional network F. It takes minimizing the Euclidean distance between the predicted coordinates and the true input value of S101 as the training objective to achieve sub-millimeter-level spatial positioning and directly output 3D coordinates instead of traditional 2D projection.

[0087] S203 uses the absolute error in the depth direction as the convergence condition, dynamically monitors the depth positioning accuracy on the validation set, and forces the model to prioritize the optimization of depth axis prediction capabilities.

[0088] S4 specifically includes the following steps:

[0089] S401. Convert the multi-channel acoustic signal into a three-dimensional sound pressure field distribution p(x, y, z, t), which is consistent with the coordinate system of the defect parameters in S101.

[0090] S402: Input the 3D convolutional network trained from p to S2, and output the coordinates (x, y, z) and size of the defect center;

[0091] S403, 3D positioning optimization, calculation of depth reliability coefficient:

[0092] Among them, z 训练 Let δ be the median depth of the S2 training model on the validation set. 训练 The depth error threshold (0.5mm) is defined for S203;

[0093] When R z If the value is ≥0.9, the positioning result is confirmed to be valid, and the coordinate parameters of the defect center are output.

[0094] Specifically, S401 converts multi-channel acoustic signals into a three-dimensional sound pressure field spatial distribution, which is consistent with the defect parameter coordinate system in S101. This ensures that the reconstructed three-dimensional sound pressure field is precisely aligned with the simulation coordinate system (including grid origin, direction, and resolution) of the S1 physical data factory, establishing a unified spatial reference. By reconstructing the three-dimensional topology of the signal and keeping it consistent with the simulation reference, sensor array deployment errors are eliminated, providing standardized input data with a spatial resolution of 0.1 mm for defect identification.

[0095] The S402 uses a 3D convolutional network trained by S2 to map sound pressure field data end-to-end into the 3D coordinates and size parameters of the defect center, realizing the intelligent conversion from the original acoustic signal to spatial defect parameters; its core effect is to directly output millimeter-level positioning results containing depth information.

[0096] S403 quantifies the deviation between the predicted value and the empirical distribution of the training set by using the depth reliability coefficient (based on the median depth of the training set and an error threshold of 0.5 mm). It dynamically intercepts depth anomalies caused by differences in sound velocity or material heterogeneity. When the reliability coefficient is ≥0.9, the result is confirmed to be valid, ensuring the depth stability of aerospace-grade detection.

[0097] S5 specifically includes the following steps:

[0098] S501. Based on the defect center coordinates and dimensions output by S4, construct a cubic computational domain centered at (x,y,z).

[0099] S502. Call COMSOL within the computational domain to perform transient acoustic simulation. Set the boundary conditions to be the surface free displacement and the outer layer perfectly matched. Input the incident sound wave signal collected by the S3 acoustic sensor and output the three-dimensional position coordinates, equivalent size and type classification results of the defect.

[0100] Specifically, based on the defect center coordinates (x, y, z) and size d output by S4, S501 constructs a cubic computational domain, automatically adapts to the defect size, and generates a high-resolution sound field reconstruction area, achieving efficient physical verification while ensuring the accuracy of three-dimensional spatial calculation.

[0101] S502 calls COMSOL within the computational domain to perform transient acoustic simulation, sets free displacement boundaries to accurately simulate the acoustic impedance characteristics of the material surface, uses an outer perfectly matched layer to suppress boundary reflections, and inputs real incident sound wave signals collected by S3 to drive the simulation, reconstructs a physical field that is strictly synchronized with the measured signal, and outputs calibrated three-dimensional defect coordinates, equivalent dimensions, and type classification. The verification results can directly replace destructive testing.

[0102] The S601 precisely maps the three-dimensional coordinates (x, y, z) of the defect to the CAD design coordinate system of the object under test through a spatial transformation matrix, achieving millimeter-level alignment between the inspection data and the engineering model (error ≤ 0.01 mm). This eliminates the systematic deviation between the equipment coordinate system and the design datum, ensuring that the defect location is directly related to the structural features of the part (such as the interface between stiffeners or weld seams), and providing a zero-conversion-loss spatial reference for maintenance positioning.

[0103] S602 generates a 3D sphere model centered on the mapped coordinates based on the dimension d output by S4, and labels it with colors (crack red / porosity yellow / corrosion blue) according to the S4 defect type classification results. It transforms abstract parameters into interactive engineering entities (such as STEP format 3D models), and intuitively labels the defect morphology and location in the CAD environment, guiding engineers to accurately locate micron-level internal damage.

[0104] S603 calculates the depth risk coefficient (the percentage of defect depth to the total material thickness) and the size risk coefficient (the percentage of defect size to the critical defect size), and combines the two types of coefficients to output a 0-1 standardized comprehensive risk index. When the index is ≥0.8, a high-risk warning is automatically triggered. Its effect is to quantify the probability of structural failure based on spatial location and geometric parameters.

[0105] Example 2:

[0106] Reference Figure 2 In a second embodiment of the present invention, the present invention provides a three-dimensional spatial acoustic detection system based on COMSOL simulation, the system comprising:

[0107] The Physical Data Factory module generates a 3D acoustic pressure field dataset through a physical constraint generative adversarial network and automatically verifies the Helmholtz equation residuals and boundary absorption rates.

[0108] The real-time detection model training module trains a three-dimensional convolutional network based on sound pressure field data to establish a mapping relationship between acoustic signals and spatial defect parameters;

[0109] A three-dimensional signal acquisition module is used to synchronously acquire multi-channel acoustic signals through a spatially distributed array of acoustic sensors;

[0110] The defect parameter identification module is used to input the collected signals into a pre-trained 3D convolutional network model and output the 3D coordinates, size, and type classification of the defects.

[0111] The COMSOL physical verification module is used to call the COMSOL engine to reconstruct the sound field in the predicted defect area and perform sound pressure peak matching and main frequency offset verification.

[0112] The 3D report generation module is used to map defect parameters to the CAD coordinate system, generate a 3D sphere model, and output a risk quantification report.

[0113] Specifically, the physical data factory module generates a 3D sound pressure field dataset for physical rule verification; the model training module constructs a mapping relationship between acoustic signals and spatial defect parameters; the signal acquisition module synchronously acquires multi-channel sound wave signals; the identification module outputs millimeter-level defect coordinates, size, and type containing depth information; the physical verification module verifies the authenticity of parameters based on COMSOL local sound field reconstruction; and the report generation module maps the results to the CAD coordinate system and generates a risk quantification 3D model, realizing millimeter-level localization of industrial-grade 3D defects, closed-loop verification of physical rules, and visualized risk decision-making.

[0114] Example 3

[0115] In the third embodiment of the present invention, based on the same inventive concept, the present invention proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the three-dimensional spatial acoustic detection method based on COMSOL simulation described in the above embodiments.

[0116] Example 4

[0117] In the fourth embodiment of the present invention, based on the same inventive concept, the present invention proposes a computer, which includes: a processor and a memory; the processor and the memory communicate with each other; the memory is used to store instructions; the processor is used to execute the instructions in the memory and execute the three-dimensional spatial acoustic detection method based on COMSOL simulation of the above embodiment.

[0118] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0119] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A three-dimensional spatial acoustic detection method based on COMSOL simulation, characterized in that, include: S1. Physical data factory construction: A three-dimensional sound pressure field dataset is generated through a physical constraint generative adversarial network. The network includes a verification layer that performs verification of the physical rules for sound wave propagation. S2. Real-time detection model training: Based on the mapping relationship between acoustic signals and defect parameters in the three-dimensional sound pressure field dataset, a three-dimensional convolutional network model is trained. S3, Three-dimensional spatial signal acquisition, which acquires the acoustic signals of the object under test through a spatially distributed multi-channel acoustic sensor array; S4. Defect parameter identification: Input the collected signals into the training model and output three-dimensional spatial defect parameters. S5. Physical auxiliary verification: COMSOL is called to reconstruct the sound field of the defect area in the output. S6. Generate a 3D inspection report, which maps defect parameters to a 3D spatial coordinate system and outputs the location results and a material integrity report.

2. The three-dimensional spatial acoustic detection method based on COMSOL simulation according to claim 1, characterized in that, S1 specifically includes the following steps: S101, Defect parameter input: Input a vector containing three-dimensional coordinates of the location, defect size and material code into the generator, and output three-dimensional sound pressure field data (X, Y, Z). The material code includes digital identifiers for steel, aluminum and composite materials. S102, Physical rule verification: Perform Helmholtz equation residual verification and boundary absorption verification on the output sound pressure field; S103, Data generation control: Reinitialize the generation process when the verification does not meet the threshold conditions.

3. The three-dimensional spatial acoustic detection method based on COMSOL simulation according to claim 2, characterized in that, Helmholtz equation residual verification in S102 This is used to verify whether the sound pressure field data p satisfies the physical constraints of the wave equation, where, For the Laplace operator of the sound pressure field, k 2 p is the product of the square of the wave number and the sound pressure level; The boundary absorption verification requirement is: the sound pressure amplitude in the perfectly matched layer ≤ 10% × max(p 全局 ), where p 全局 The sound pressure level in the entire computational domain.

4. The three-dimensional spatial acoustic detection method based on COMSOL simulation according to claim 1, characterized in that, S2 specifically includes the following steps: S201, Loading 3D Sound Pressure Field Data: Using the verified 3D sound pressure field data p generated in S1. gen ∈R t×X×Y×Z As the training input, t is the number of time or frequency sampling points, X and Y are the number of spatial grids in the horizontal direction, and Z is the number of spatial grids in the depth direction; S202, Spatial coordinate mapping training, establishing a regression mapping from the sound pressure field to the coordinates of the defect center: (xyz)=F(p gen ); where F is the mapping function for a lightweight 3D convolutional network; Training objective: Minimize the spatial location error ||(x, y, z) - (x 真 y 真 , z 真 )||2, where (x 真 y 真 , z 真 () represents the three-dimensional coordinates of the defect location input by S101; S203, Depth and accuracy compliance verification, when the model satisfies the following on the validation set: δ z =|zz 真 Training will be terminated if the thickness is ≤0.5mm.

5. The three-dimensional spatial acoustic detection method based on COMSOL simulation according to claim 1, characterized in that, S4 specifically includes the following steps: S401. Convert the multi-channel acoustic signal into a three-dimensional sound pressure field distribution p(x, y, z, t), which is consistent with the coordinate system of the defect parameters in S101. S402: Input the 3D convolutional network trained from p to S2, and output the coordinates (x, y, z) and size of the defect center; S403, 3D positioning optimization, calculation of depth reliability coefficient: Among them, z 训练 Let δ be the median depth of the S2 training model on the validation set. 训练 The depth error threshold (0.5mm) is defined for S203; When R z If the value is ≥0.9, the positioning result is confirmed to be valid, and the coordinate parameters of the defect center are output.

6. The three-dimensional spatial acoustic detection method based on COMSOL simulation according to claim 1, characterized in that, S5 specifically includes the following steps: S501. Based on the defect center coordinates and dimensions output by S4, construct a cubic computational domain centered at (x,y,z). S502. Call COMSOL within the computational domain to perform transient acoustic simulation. Set the boundary conditions to be the surface free displacement and the outer layer perfectly matched. Input the incident sound wave signal collected by the S3 acoustic sensor and output the three-dimensional position coordinates, equivalent size and type classification results of the defect.

7. The three-dimensional spatial acoustic detection method based on COMSOL simulation according to claim 1, characterized in that, S6 specifically includes the following steps: S601. Transform the output defect coordinates (x, y, z) to the coordinate system of the CAD model of the object under test, and output the precise position through spatial transformation relationship; S602. Using the defect coordinates as the center, generate a spherical solid model through the dimension parameter d output by S4, and label the defect type with color classification based on S4. S603, Calculate the depth risk coefficient: the percentage of the maximum material thickness at depth; Size risk factor: The percentage of size to the critical defect size of the material, outputting a comprehensive risk index.

8. A three-dimensional spatial acoustic detection system based on COMSOL simulation, characterized in that, The system for the three-dimensional spatial acoustic detection method based on COMSOL simulation according to any one of claims 1-7 comprises: The Physical Data Factory module generates a 3D acoustic pressure field dataset through a physical constraint generative adversarial network and automatically verifies the Helmholtz equation residuals and boundary absorption rates. The real-time detection model training module trains a three-dimensional convolutional network based on sound pressure field data to establish a mapping relationship between acoustic signals and spatial defect parameters; A three-dimensional signal acquisition module is used to synchronously acquire multi-channel acoustic signals through a spatially distributed array of acoustic sensors; The defect parameter identification module is used to input the collected signals into a pre-trained 3D convolutional network model and output the 3D coordinates, size, and type classification of the defects. The COMSOL physical verification module is used to call the COMSOL engine to reconstruct the sound field in the predicted defect area and perform sound pressure peak matching and main frequency offset verification. The 3D report generation module is used to map defect parameters to the CAD coordinate system, generate a 3D sphere model, and output a risk quantification report.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the three-dimensional spatial acoustic detection method based on COMSOL simulation as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program, which, when executed by a processor, implements the three-dimensional spatial acoustic detection method based on COMSOL simulation as described in any one of claims 1 to 7.

Images