Flexible self-adaptive acquisition ultrasonic detection system and method based on data fusion

Abstract

The application discloses a flexible self-adaptive acquisition ultrasonic detection system and method based on data fusion, which is composed of an intelligent ultrasonic excitation module, a distributed acquisition unit, a flexible self-adaptive registration platform, a data fusion and optimization module, an imaging module and a display terminal. The intelligent ultrasonic excitation module excites ultrasonic signals to a workpiece, the distributed acquisition unit collects reflected signals from multiple perspectives, the flexible self-adaptive registration platform adaptively adjusts the posture and adhesion of the distributed acquisition unit, the data fusion and optimization module fuses the multi-perspective ultrasonic data based on a graph neural network, and provides an optimization scheme of excitation parameters and ultrasonic sensor subarray actions based on the fused features, the imaging module completes imaging through an energy inversion method, and finally, the display terminal is used for display. The application realizes adaptive fusion and intelligent imaging of distributed ultrasonic data through the graph neural network, and effectively improves the detection precision and imaging quality of complex curved workpieces.

Description

Technical Field

[0001] This invention belongs to the field of nondestructive testing technology, and particularly relates to a flexible adaptive acquisition ultrasonic detection system and method based on data fusion. Background Technology

[0002] Traditional ultrasonic testing techniques typically rely on single-channel probes, manual point-by-point scanning, or fixedly arranged rigid array probes. Their testing capabilities are limited in several ways: for workpieces with complex geometries, rough surfaces, or significant local curvature changes, rigid arrays struggle to achieve stable acoustic coupling, leading to inconsistent echo signal quality, low signal-to-noise ratio, and incomplete defect information. In large-area or irregular structure testing, traditional scanning methods are inefficient and prone to producing beam blind spots in areas with abrupt structural changes or curvature shifts. Furthermore, existing ultrasonic testing systems generally lack real-time analysis and intelligent adjustment of sensor attitude, coupling state, and acquisition quality. Their testing effectiveness is highly dependent on operator experience. When faced with workpieces exhibiting heterogeneous material properties, complex geometries, or local deformations, traditional systems struggle to dynamically compensate based on changes in sound field distribution or imaging quality feedback, failing to guarantee the consistency of distributed acquisition data and the stability of global imaging. On the other hand, existing ultrasonic 3D imaging technologies often rely on single-array acquisition, making it difficult to cover large workpieces; multiple acquisitions from different arrays or locations still face technical bottlenecks in spatial registration, feature consistency, and fusion imaging.

[0003] In summary, there is an urgent need for an ultrasonic testing system that can achieve flexible adaptive fitting, highly consistent distributed acquisition, and intelligent multi-source data fusion. Summary of the Invention

[0004] The purpose of this invention is to provide a flexible adaptive acquisition ultrasonic detection system and method based on data fusion, so as to solve the problems and defects of existing ultrasonic nondestructive testing technology in large-size, complex curved workpieces, such as poor coupling, acquisition blind zone and insufficient imaging accuracy.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] The flexible adaptive acquisition ultrasonic detection system based on data fusion mainly includes the following modules:

[0007] The intelligent ultrasonic excitation module is used to generate ultrasonic signals to the workpiece under test and supports adaptive adjustment of ultrasonic excitation parameters.

[0008] The distributed acquisition unit consists of several independent ultrasonic sensor subarrays, used to acquire reflected signals inside the workpiece from multiple perspectives;

[0009] A flexible adaptive registration platform includes a flexible substrate, several micro actuators, and several high-precision positioning sensors. The micro actuators and high-precision positioning sensors are integrated inside the flexible substrate. The flexible substrate is used to be placed on the surface of the workpiece as a carrier for the distributed acquisition unit. The micro actuators are used to adaptively adjust the local attitude, tilt angle, and surface coupling fit of the ultrasonic sensor subarray. The high-precision positioning sensors are used to monitor the global three-dimensional coordinates and six-degree-of-freedom attitude information of the ultrasonic sensor subarray.

[0010] The data fusion and optimization module is used to adaptively fuse multi-view ultrasound data acquired by different ultrasound sensor subarrays based on a graph neural network, evaluate the acquisition quality and imaging reliability based on the fused ultrasound data, and generate ultrasound excitation parameter optimization schemes and ultrasound sensor subarray motion optimization schemes. The graph neural network uses the ultrasound sensor subarray as nodes, and its spatial topology, attitude information, coupling quality, and signal similarity relationships as edges. The generated ultrasound excitation parameter optimization scheme guides the intelligent ultrasound excitation module to complete the adaptive adjustment of ultrasound excitation parameters, and the ultrasound sensor subarray motion optimization scheme guides the flexible adaptive registration platform to complete the adjustment of the local attitude, tilt angle, and surface coupling fit of the ultrasound sensor subarray.

[0011] The imaging module is used to construct a multi-scale energy response model based on the fused ultrasonic data, generate the three-dimensional energy distribution volume inside the workpiece through multi-scale energy inversion, and reconstruct the internal structure of the workpiece with the energy peak region.

[0012] The display terminal is used to display the three-dimensional imaging results inside the workpiece.

[0013] Preferably, the flexible adaptive registration platform supports manual adjustment of the local attitude, tilt angle, and surface coupling fit of the ultrasonic sensor subarray.

[0014] The present invention further provides an ultrasonic detection method based on the aforementioned data fusion-based flexible adaptive acquisition ultrasonic detection system, comprising the following steps:

[0015] S1. A flexible adaptive registration platform and a distributed acquisition unit are deployed on the surface of the workpiece to be tested, and an intelligent ultrasonic excitation module excites ultrasonic signals of multiple frequencies, multiple angles, and multiple waveforms.

[0016] S2. The distributed acquisition unit acquires the reflected signals inside the workpiece, and the high-precision positioning sensor on the flexible adaptive registration platform records the global three-dimensional coordinates and six-degree-of-freedom attitude information of each ultrasonic sensor subarray in real time.

[0017] S3, the data fusion and optimization module adaptively fuses multi-view ultrasound data acquired by different ultrasound sensor subarrays;

[0018] S4. The imaging module constructs a multi-scale energy spectrum based on the fused ultrasonic data, and reconstructs the three-dimensional energy volume through an energy inversion algorithm, using the energy peak region to reconstruct the internal structure of the workpiece.

[0019] S5. The data fusion and optimization module evaluates and judges the acquisition quality and imaging reliability. If the acquisition quality and imaging reliability do not meet the set requirements, it generates an ultrasonic sensor subarray motion optimization scheme and an ultrasonic excitation parameter optimization scheme. This guides the intelligent ultrasonic excitation module to adaptively adjust the ultrasonic excitation parameters and guides the micro-driver in the flexible adaptive registration platform to adaptively adjust the local attitude, tilt angle and surface coupling fit of the ultrasonic sensor subarray until the acquisition quality and imaging reliability meet the set requirements.

[0020] S6. Output the image of the internal structure of the workpiece through the display terminal.

[0021] Preferably, step S3 specifically includes the following steps: the data fusion and optimization module performs denoising, filtering, and envelope extraction processing on the acquired reflection signal, and combines deep learning algorithms to identify and extract multi-scale structural features, energy features, and coupling quality indicators inside the workpiece; using the ultrasonic sensor subarray as nodes, and its spatial topological relationship, attitude information, coupling quality, and signal similarity relationship as graph edge features to construct an ultrasonic acquisition map, and achieving adaptive fusion of multi-view ultrasonic data through graph convolution or graph attention mechanisms.

[0022] Preferably, in step S5, the ultrasonic sensor subarray motion optimization scheme is generated by constructing a graph-constrained attitude minimum perturbation optimization model, and the output characteristics of the graph-constrained attitude minimum perturbation optimization model are:

[0023] ;

[0024] in, For the first The six-degree-of-freedom attitude adjustment of an ultrasonic sensor subarray Let be the drive control increment vector corresponding to the ith ultrasonic sensor subarray in one optimization iteration, used to characterize the displacement, rotation, or pressure adjustment commands generated by the micro-actuator for the ultrasonic sensor subarray. and For graph neural networks to the first The and the first Fusion characteristics of the outputs of each ultrasonic sensor subarray For the first Local reflection energy of an ultrasonic sensor subarray Let E be the mean energy of the neighborhood, and E be the edge set of the ultrasound acquisition map. The edge weights represent the spatial distance or coupling quality between arrays. These are weighting coefficients used to balance the relative importance of the attitude adjustment range constraint and the energy boost term in the objective function. The learning rate parameter is used to control the step size of each iteration during the pose optimization iteration process.

[0025] Compared with the prior art, the present invention has the following advantages:

[0026] (1) The present invention adopts a flexible adaptive registration platform, which performs local six-degree-of-freedom attitude adjustment on each ultrasonic sensor subarray through an integrated micro-driver. It can maintain stable acoustic coupling on complex curved surfaces, rough workpiece surfaces and local abrupt regions, effectively solving the problems of poor coupling, low signal-to-noise ratio and acquisition blind zone of traditional rigid arrays in curved surface detection, thereby significantly improving the stability of echo signal and imaging reliability.

[0027] (2) The present invention uses a distributed acquisition unit to realize large-scale signal acquisition from multiple perspectives and multiple paths, which not only has higher detection coverage and acquisition efficiency, but also provides rich data support for three-dimensional fusion imaging, enabling the system to be applicable to the high-precision non-destructive testing requirements of large-size and complex structure workpieces;

[0028] (3) This invention proposes a multi-source ultrasound data fusion method based on data fusion. The signal characteristics, three-dimensional spatial topology, and attitude information of the ultrasound sensor subarray are constructed into an ultrasound acquisition map, and graph convolution or graph attention mechanism is used to achieve deep fusion and consistency compensation of cross-array signals. This method effectively eliminates the signal inconsistency problem caused by coupling differences, attitude differences, and path differences between different arrays, and significantly enhances the expressive power and robustness of ultrasound data. At the same time, this invention uses a minimum perturbation attitude optimization algorithm with graph structure constraints to solve the optimal attitude adjustment amount by using the node features and local energy distribution output by the graph neural network. This enables the array to achieve minimum amplitude adjustment while maintaining overall stability, maximizing the improvement of fitting quality and reducing the impact of environmental disturbances, thereby ensuring the continuity of the detection process and the consistency of imaging quality.

[0029] (4) This invention employs a three-dimensional fusion imaging method based on multi-scale energy inversion. An energy response model is established based on the fused distributed ultrasonic data. The three-dimensional energy distribution volume inside the workpiece is constructed through multi-scale energy inversion, effectively avoiding dependence on wave velocity models or reverse time migration algorithms. This makes the imaging process more adaptable to material heterogeneity, signal attenuation differences, and complex path structures. Compared to traditional wavefield imaging methods, the three-dimensional imaging results generated by this invention are more stable and robust, clearly revealing cracks, delamination, voids, and other internal structural features. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the flexible adaptive acquisition ultrasonic detection system based on data fusion according to the present invention;

[0031] Figure 2 This is a schematic diagram showing the layout of the flexible adaptive registration platform and the distributed acquisition unit in this invention;

[0032] Figure 3 This is a flowchart of the detection method of the flexible adaptive acquisition ultrasonic detection system based on data fusion according to the present invention;

[0033] Reference numerals: 1. Intelligent ultrasonic excitation module; 2. Distributed acquisition unit; 21. Ultrasonic sensor subarray; 3. Flexible adaptive registration platform; 31. Flexible substrate; 32. Micro-actuator; 33. High-precision positioning sensor; 4. Data fusion and optimization module; 5. Imaging module; 6. Display terminal; 7. Workpiece under test; 8. Ultrasonic signal; 9. Reflected signal. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0035] This invention provides a flexible adaptive acquisition ultrasonic detection system based on data fusion, such as... Figure 1 As shown, the system consists of an intelligent ultrasound excitation module 1, a distributed acquisition unit 2, a flexible adaptive registration platform 3, a data fusion and optimization module 4, an imaging module 5, and a display terminal 6.

[0036] 1. Intelligent ultrasonic excitation module

[0037] The intelligent ultrasonic excitation module 1 is used to excite ultrasonic signals 8 into the interior of the workpiece 7 under test. It is a programmable sweep frequency or pulse generator. The intelligent ultrasonic excitation module 1 supports dynamic adjustment of excitation signal parameters (such as ultrasonic frequency, pulse width, and transmission power) to ensure that the signal parameters can adaptively select the optimal penetration and resolution parameters according to the workpiece material and real-time imaging feedback. The dynamic adjustment capability of the intelligent ultrasonic excitation module 1 is controlled by the data fusion and optimization module 4, which provides the intelligent ultrasonic excitation module 1 with an optimized excitation parameter scheme.

[0038] 2. Distributed acquisition unit

[0039] The distributed acquisition unit 2 is used to acquire the internal reflection signals 9 of the workpiece 7 under test. The distributed acquisition unit 2 consists of several independent ultrasonic sensor subarrays 21, such as high-density phased array probes or full-matrix acquisition arrays. These ultrasonic sensor subarrays are all integrated onto the flexible adaptive registration platform 3, and can independently or in parallel acquire the reflection signals 9 of different local areas. This distributed structure is the basis for realizing large-area, high-coverage detection, and solves the limitations of traditional single arrays or fixed arrays.

[0040] 3. Flexible Adaptive Registration Platform

[0041] The flexible adaptive registration platform 3 mainly includes a flexible substrate 31, several micro-actuators 32, and several high-precision positioning sensors 33. The micro-actuators 32, high-precision positioning sensors 33, and distributed acquisition units are integrated on the flexible substrate 31. Each ultrasonic sensor subarray 21 is equipped with one micro-actuator 32 and one high-precision positioning sensor 33. The flexible substrate 31, serving as the carrier for the ultrasonic sensor subarray 21, can be bent and deformed, ensuring close contact between the ultrasonic sensor subarray 21 and the surface curvature or unevenness of the workpiece 7 under test. The micro-actuators 32, employing piezoelectric ceramic actuators or micro-airbags, receive instructions from the data fusion and optimization module 4 to perform fine adaptive adjustments to the local attitude, tilt angle, and surface coupling fit of the ultrasonic sensor subarray 21 to achieve optimal acoustic coupling. The high-precision positioning sensors 33, employing integrated inertial measurement units (IMUs) or laser / vision positioning units, record the global three-dimensional coordinates and six-degree-of-freedom attitude information of each ultrasonic sensor subarray 21 in real time, providing precise geometric input for subsequent spatial registration and fusion technology.

[0042] 4. Data Fusion and Optimization Module

[0043] The data fusion and optimization module 4 is responsible for receiving the reflected signal 9 collected by the distributed acquisition unit 2 and performing preprocessing such as denoising, filtering, and envelope extraction. Then, it combines deep learning algorithms to quickly identify and extract the internal structural feature information of the workpiece 7 under test from the preprocessed reflected signal 9, such as cracks, voids, delamination, or the location of reinforcing bars. It also constructs a graph neural network to adaptively fuse multi-view ultrasonic data collected by different ultrasonic sensor subarrays 21, and evaluates the acquisition quality and imaging reliability based on the fused ultrasonic data. If the acquisition quality and imaging reliability do not meet the set requirements, it generates an ultrasonic sensor subarray motion optimization scheme and an ultrasonic excitation parameter optimization scheme to guide the intelligent ultrasonic excitation module 1 to adaptively adjust the ultrasonic excitation parameters, and guides the micro-actuator 32 in the flexible adaptive registration platform 3 to adaptively adjust the local attitude, tilt angle, and surface coupling fit of the ultrasonic sensor subarray 21 until the acquisition quality and imaging reliability meet the set requirements.

[0044] (1) Adaptive fusion mechanism of graph neural network (GNN)

[0045] Using the ultrasonic sensor subarray 21 as nodes, a graph neural network (GNN) is constructed with its spatial topology, attitude information, coupling quality, and signal similarity as edges. The GNN aggregates the feature information of adjacent ultrasonic sensor subarrays 21 through a message passing mechanism to generate robust fused features.

[0046] ;

[0047] in, and They are nodes In the Layer and first The feature vector of the layer; It is a non-linear activation function; The weight matrix for the node's own information update terms. The weight matrix for fusing information from neighboring nodes. and All can be learned; It is the set of adjacent nodes; Adjacent nodes In the The feature vector of the layer; The attention coefficients for adaptive learning represent the neighboring nodes. For nodes Weight of contribution.

[0048] (2) Minimum perturbation attitude optimization model with graph structure constraints

[0049] The data fusion and optimization module 4 couples the fused features output by the GNN, local reflection energy, and driving energy consumption constraints to construct a minimum perturbation attitude optimization model with graph structure constraints. This model is used to evaluate the current fitting state of the flexible adaptive registration platform 3 and solves for the optimal six-degree-of-freedom attitude adjustment amount for each ultrasonic sensor subarray 21 by minimizing the objective function.

[0050] ;

[0051] in, For the first The six-degree-of-freedom attitude adjustment of an ultrasonic sensor subarray Let be the drive control increment vector corresponding to the ith ultrasonic sensor subarray in one optimization iteration, used to characterize the displacement, rotation, or pressure adjustment commands generated by the micro-actuator for the ultrasonic sensor subarray. and For graph neural networks to the first The and the first Fusion characteristics of the outputs of each ultrasonic sensor subarray For the first Local reflection energy of an ultrasonic sensor subarray Let E be the mean energy of the neighborhood, and E be the edge set of the ultrasound acquisition map. The edge weights represent the spatial distance or coupling quality between arrays. These are weighting coefficients used to balance the relative importance of the attitude adjustment range constraint and the energy boost term in the objective function. The learning rate parameter is used to control the step size of each iteration during the pose optimization iteration process.

[0052] (3) Iterative optimization process of attitude adjustment

[0053] The system iteratively solves the above objective function through data fusion and optimization module 4 to optimize the attitude adjustment strategy of the flexible adaptive registration platform 3 in real time. This optimization problem is solved using a gradient-based optimization algorithm, and the iterative update formula for the attitude adjustment amount can be expressed as:

[0054] ;

[0055] in, and The first Second and third The attitude adjustment amount calculated in the next iteration. For the learning rate step size, For the objective function Regarding attitude adjustment amount The gradient.

[0056] (4) Human decision support mechanism

[0057] When the system does not automatically trigger optimization, users can choose to intervene manually to determine whether further optimization of the acquisition parameters is needed based on the imaging data and system feedback results.

[0058] 5. Imaging module

[0059] Imaging module 5 performs multi-scale energy inversion based on the fused ultrasonic data to obtain the spatial energy distribution of internal defects in the workpiece, and then performs three-dimensional volume reconstruction based on this energy distribution. The three-dimensional volume reconstruction generates a high-resolution three-dimensional structural image of the workpiece's interior by voxelizing the inverted energy field and performing geometric correction.

[0060] 6. Display terminal

[0061] The display terminal 6 is connected to the data fusion and optimization module 4 and the imaging module 5 respectively, and is used to display the internal structural feature information of the workpiece, the three-dimensional imaging results, and the ultrasonic excitation parameter optimization scheme and ultrasonic sensor subarray action optimization scheme provided by the data fusion and optimization module 4 in real time.

[0062] The workflow of the flexible adaptive acquisition ultrasonic detection system based on data fusion of this invention is as follows:

[0063] S1. A flexible adaptive registration platform and a distributed acquisition unit are deployed on the surface of the workpiece to be tested, and an intelligent ultrasonic excitation module excites ultrasonic signals of multiple frequencies, multiple angles, and multiple waveforms.

[0064] S2. The distributed acquisition unit acquires the reflected signals inside the workpiece, and the high-precision positioning sensor on the flexible adaptive registration platform records the global three-dimensional coordinates and six-degree-of-freedom attitude information of each ultrasonic sensor subarray in real time.

[0065] The S3 data fusion and optimization module performs denoising, filtering, and envelope extraction on the acquired reflection signals. It also combines deep learning algorithms to identify and extract multi-scale structural features, energy features, and coupling quality indicators inside the workpiece. Using the ultrasonic sensor subarray as nodes, it constructs an ultrasonic acquisition map with its spatial topology, attitude information, coupling quality, and signal similarity as graph edge features. The adaptive fusion of multi-view ultrasonic data is achieved through graph convolution or graph attention mechanisms.

[0066] S4. The imaging module constructs a multi-scale energy spectrum based on the fused ultrasonic data, and reconstructs the three-dimensional energy volume through an energy inversion algorithm, using the energy peak region to reconstruct the internal structure of the workpiece.

[0067] S5. The data fusion and optimization module evaluates and judges the acquisition quality and imaging reliability. If the acquisition quality and imaging reliability do not meet the set requirements, it generates an ultrasonic sensor subarray motion optimization scheme and an ultrasonic excitation parameter optimization scheme. This guides the intelligent ultrasonic excitation module to adaptively adjust the ultrasonic excitation parameters and guides the micro-driver in the flexible adaptive registration platform to adaptively adjust the local attitude, tilt angle and surface coupling fit of the ultrasonic sensor subarray until the acquisition quality and imaging reliability meet the set requirements.

[0068] S6. Output the image of the internal structure of the workpiece through the display terminal.

[0069] In summary, the flexible adaptive acquisition ultrasonic detection system based on data fusion provided by this invention delivers high-precision and high-efficiency detection results in complex curved surfaces and heterogeneous media environments through adaptive data fusion and minimum disturbance posture optimization using graph neural networks. The system's intelligent optimization and feedback mechanism, combined with human intervention, ensures the system's intelligence and flexibility, adapting to the detection needs of different workpiece structures and material conditions, and greatly improving the accuracy and reliability of the detection.

[0070] The above description is only a preferred embodiment of the present invention and is 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 within the protection scope of the present invention.

Claims

1. A flexible adaptive acquisition ultrasonic detection system based on data fusion, characterized in that, Includes the following modules: The intelligent ultrasonic excitation module is used to generate ultrasonic signals to the workpiece under test and supports adaptive adjustment of ultrasonic excitation parameters. The distributed acquisition unit consists of several independent ultrasonic sensor subarrays, used to acquire reflected signals inside the workpiece from multiple perspectives; A flexible adaptive registration platform includes a flexible substrate, several micro actuators, and several high-precision positioning sensors. The micro actuators and high-precision positioning sensors are integrated inside the flexible substrate. The flexible substrate is used to be placed on the surface of the workpiece as a carrier for the distributed acquisition unit. The micro actuators are used to adaptively adjust the local attitude, tilt angle, and surface coupling fit of the ultrasonic sensor subarray. The high-precision positioning sensors are used to monitor the global three-dimensional coordinates and six-degree-of-freedom attitude information of the ultrasonic sensor subarray. The data fusion and optimization module is used to adaptively fuse multi-view ultrasound data acquired by different ultrasound sensor subarrays based on a graph neural network, evaluate the acquisition quality and imaging reliability based on the fused ultrasound data, and generate ultrasound excitation parameter optimization schemes and ultrasound sensor subarray motion optimization schemes. The graph neural network uses the ultrasound sensor subarray as nodes, and its spatial topology, attitude information, coupling quality, and signal similarity relationships as edges. The generated ultrasound excitation parameter optimization scheme guides the intelligent ultrasound excitation module to complete the adaptive adjustment of ultrasound excitation parameters, and the ultrasound sensor subarray motion optimization scheme guides the flexible adaptive registration platform to complete the adjustment of the local attitude, tilt angle, and surface coupling fit of the ultrasound sensor subarray. The imaging module is used to construct a multi-scale energy response model based on the fused ultrasonic data, generate the three-dimensional energy distribution volume inside the workpiece through multi-scale energy inversion, and reconstruct the internal structure of the workpiece with the energy peak region. The display terminal is used to display the three-dimensional imaging results inside the workpiece.

2. The flexible adaptive acquisition ultrasonic detection system based on data fusion according to claim 1, characterized in that, The flexible adaptive registration platform supports manual adjustment of the local attitude, tilt angle, and surface coupling fit of the ultrasonic sensor subarray.

3. An ultrasonic detection method based on the system described in claim 1 or 2, characterized in that, Includes the following steps: S1. A flexible adaptive registration platform and a distributed acquisition unit are deployed on the surface of the workpiece to be tested, and an intelligent ultrasonic excitation module excites ultrasonic signals of multiple frequencies, multiple angles, and multiple waveforms. S2. The distributed acquisition unit acquires the reflected signals inside the workpiece, and the high-precision positioning sensor on the flexible adaptive registration platform records the global three-dimensional coordinates and six-degree-of-freedom attitude information of each ultrasonic sensor subarray in real time. S3, the data fusion and optimization module adaptively fuses multi-view ultrasound data acquired by different ultrasound sensor subarrays; S4. The imaging module constructs a multi-scale energy spectrum based on the fused ultrasonic data, and reconstructs the three-dimensional energy volume through an energy inversion algorithm, using the energy peak region to reconstruct the internal structure of the workpiece. S5. The data fusion and optimization module evaluates and judges the acquisition quality and imaging reliability. If the acquisition quality and imaging reliability do not meet the set requirements, it generates an ultrasonic sensor subarray motion optimization scheme and an ultrasonic excitation parameter optimization scheme. This guides the intelligent ultrasonic excitation module to adaptively adjust the ultrasonic excitation parameters and guides the micro-driver in the flexible adaptive registration platform to adaptively adjust the local attitude, tilt angle and surface coupling fit of the ultrasonic sensor subarray until the acquisition quality and imaging reliability meet the set requirements. S6. Output the image of the internal structure of the workpiece through the display terminal.

4. The ultrasonic detection method according to claim 3, characterized in that, Step S3 specifically includes the following steps: The data fusion and optimization module performs denoising, filtering, and envelope extraction processing on the acquired reflection signals, and combines deep learning algorithms to identify and extract multi-scale structural features, energy features, and coupling quality indicators inside the workpiece; Using the ultrasonic sensor subarray as nodes, the ultrasonic acquisition map is constructed with its spatial topological relationship, attitude information, coupling quality, and signal similarity relationship as graph edge features, and adaptive fusion of multi-view ultrasonic data is achieved through graph convolution or graph attention mechanisms.

5. The ultrasonic detection method according to claim 4, characterized in that, In step S5, the ultrasonic sensor subarray motion optimization scheme is generated by constructing a graph-constrained attitude minimum perturbation optimization model. The output characteristics of the graph-constrained attitude minimum perturbation optimization model are: ； in, For the first The six-degree-of-freedom attitude adjustment of an ultrasonic sensor subarray Let be the drive control increment vector corresponding to the ith ultrasonic sensor subarray in one optimization iteration, used to characterize the displacement, rotation, or pressure adjustment commands generated by the micro-actuator for the ultrasonic sensor subarray. and For graph neural networks to the first The and the first Fusion characteristics of the outputs of each ultrasonic sensor subarray For the first Local reflection energy of an ultrasonic sensor subarray Let E be the mean energy of the neighborhood, and E be the edge set of the ultrasound acquisition map. The edge weights represent the spatial distance or coupling quality between arrays. These are weighting coefficients used to balance the relative importance of the attitude adjustment range constraint and the energy boost term in the objective function. The learning rate parameter is used to control the step size of each iteration during the pose optimization iteration process.

Images