A probe parameter optimization system and method for an eddy current sensor

Abstract

The application provides a kind of probe parameter optimization system and method of eddy current sensor, including time delay processing module, difference module, three-channel gray processing module and convolutional neural network;Time delay processing module is used to receive sinusoidal output signal, and sinusoidal output signal is time-delayed, and sinusoidal delay signal is generated;Difference module is used to receive sinusoidal delay signal and cosine output signal, and difference signal is generated;Three-channel gray processing module is used to receive sinusoidal output signal, cosine output signal, difference signal, and three-channel gray image is generated after splicing and gray processing;Convolutional neural network is used to receive three-channel gray image, and nine kinds of optimization labels are generated after feature extraction and classification of three-channel gray image.The application can accurately identify the sinusoidal coil, cosine coil and excitation coil in the probe, improve the parameter optimization efficiency and optimization accuracy.

Description

Technical Field

[0001] This application relates to the field of sensor technology, and in particular to a method and system for optimizing probe parameters of an eddy current sensor. Background Technology

[0002] Eddy current sensors generate a high-frequency alternating magnetic field through an excitation coil. Eddy currents are induced in the metal target rotor within this magnetic field. Sine and cosine coils respectively induce and output sine and cosine signals that vary with the rotor's angular position. After signal processing, the angle value is calculated. Due to their non-contact and high reliability characteristics, this type of sensor is widely used in industrial servo control, robot joints, and other fields.

[0003] The angle measurement accuracy of eddy current sensors is closely related to the parameter configuration of the sine coil, cosine coil, and excitation coil in the probe. Existing probe parameter optimization methods typically rely on finite element simulation combined with manual parameter tuning. This requires adjusting parameters such as the number of turns, width, pitch, and excitation frequency of each coil individually in the simulation model, followed by verification through actual measurements. However, there are complex high-dimensional nonlinear coupling relationships between the parameters of the sine coil, cosine coil, and excitation coil. A change in a single parameter often causes a chain reaction of changes in the output characteristics of other coils, making existing probe parameter optimization methods inefficient and difficult to guarantee the angle measurement accuracy of the obtained parameter configuration. Therefore, there is an urgent need for a probe parameter optimization method for eddy current sensors that can accurately identify the sine coil, cosine coil, and excitation coil in the probe, thereby improving the efficiency and accuracy of parameter optimization. Summary of the Invention

[0004] In view of the above problems, this application is proposed to provide a probe parameter optimization system and method for an eddy current sensor. The difference signal between the sine and cosine output signals is introduced as an independent third channel. The difference signal carries the amplitude inconsistency and nonlinear error information between the sine and cosine output signals. Together with the sine and cosine output signals, it forms a three-channel input, providing richer error features to the convolutional neural network. This enables accurate identification of the sine coil, cosine coil, and excitation coil in the probe, thereby improving the efficiency and accuracy of parameter optimization.

[0005] To achieve the above objectives, in a first aspect, this application provides a probe parameter optimization system for an eddy current sensor, including a time delay processing module, a difference module, a three-channel grayscale processing module, and a convolutional neural network;

[0006] The time delay processing module is used to receive the sinusoidal output signal, perform time delay processing on the sinusoidal output signal, and generate a sinusoidal delayed signal.

[0007] The difference module is used to receive the sinusoidal delayed signal and the cosine output signal, perform difference processing on the sinusoidal delayed signal and the cosine output signal, and generate a difference signal.

[0008] The three-channel grayscale processing module is used to receive sine output signals, cosine output signals, and difference signals, and generate a three-channel grayscale image after splicing and grayscale processing.

[0009] A convolutional neural network is used to receive a three-channel grayscale image, extract features from the three-channel grayscale image, classify it, and generate nine optimized labels.

[0010] Furthermore, the nine optimization tags include optimization of the number of turns of the sine coil, optimization of the width of the sine coil, optimization of the pitch of the sine coil, optimization of the number of turns of the cosine coil, optimization of the width of the cosine coil, optimization of the pitch of the cosine coil, optimization of the excitation frequency, optimization of the amplitude of the excitation current, and optimization of the number of turns of the excitation coil.

[0011] Furthermore, the convolutional neural network includes a first convolutional layer, a second convolutional layer, a first Transformer module, a second Transformer module, a third Transformer module, a feature fusion module, and an accuracy improvement module;

[0012] The first convolutional feature extraction module is used to receive a three-channel grayscale image, perform shallow feature extraction operations, and generate a first feature map.

[0013] The second convolutional feature extraction module is used to receive the first feature map, perform shallow feature extraction operations, and generate the second feature map.

[0014] The first Transformer module is used to receive the second feature map, perform deep feature extraction operations, and generate the third feature map.

[0015] The second Transformer module is used to receive the third feature map, perform deep feature extraction operations, and generate the fourth feature map.

[0016] The third Transformer module is used to receive the fourth feature map, perform deep feature extraction, and generate the fifth feature map.

[0017] The feature fusion module is used to receive the first feature map, the second feature map, the third feature map, the fourth feature map, and the fifth feature map to generate a fused feature map;

[0018] The accuracy improvement module is used to receive the fused feature map, enhance detailed features and waveform boundary information, and generate the final feature map.

[0019] The classification module receives the final feature map, classifies it, and generates 9 optimized labels.

[0020] Furthermore, the feature fusion module includes a cascaded extended fusion module and a main fusion module;

[0021] The expansion fusion module is used to receive the third feature map, the fourth feature map and the fifth feature map, and perform multi-scale expansion fusion.

[0022] The main fusion module is used to receive the output of the extended fusion module, the first feature map, and the second feature map, and to fuse the output of the extended fusion module, the first feature map, and the second feature map to generate a fused feature map.

[0023] Furthermore, the extended fusion module includes a first 1×1 convolutional layer, a first extended convolutional block, a first upsampling, a first stitching module, a second 1×1 convolutional layer, a second extended convolutional block, a second upsampling, a second stitching module, a third 1×1 convolutional layer, and a third extended convolutional block;

[0024] The first 1×1 convolutional layer is used to receive the fifth feature map and perform convolution operations on the fifth feature map;

[0025] The first dilated convolutional block is used to receive the output of the first 1×1 convolutional layer and perform dilated convolution operation on the output of the first 1×1 convolutional layer.

[0026] The first upsampling module is used to receive the output of the first dilated convolutional block and perform an upsampling operation on the output of the first dilated convolutional block.

[0027] The first stitching module is used to receive the output of the first upsampling module and the fourth feature map, and to perform a stitching operation on the output of the first upsampling module and the fourth feature map.

[0028] The second 1×1 convolutional layer is used to receive the output of the first splicing module and perform convolution operation on the output of the first splicing module;

[0029] The second dilated convolution block is used to receive the output of the second 1×1 convolutional layer and perform dilated convolution operation on the output of the second 1×1 convolutional layer.

[0030] The second upsampling module is used to receive the output of the second dilated convolutional block and perform upsampling operation on the output of the second dilated convolutional block.

[0031] The second stitching module is used to receive the output of the second upsampling module and the third feature map, and to perform a stitching operation on the output of the second upsampling module and the third feature map.

[0032] The third 1×1 convolutional layer is used to receive the output of the second splicing module and perform convolution operation on the output of the second splicing module;

[0033] The third dilated convolution block is used to receive the output of the third 1×1 convolutional layer and perform dilated convolution operations on the output of the third 1×1 convolutional layer.

[0034] Furthermore, the first dilated convolutional block, the second dilated convolutional block, and the third dilated convolutional block each include a cascaded 3×3 convolutional layer, a dilated convolutional layer, a batch normalization layer, and an activation layer.

[0035] Furthermore, the expansion rates of the first, second, and third expanded convolutional blocks gradually increase.

[0036] Furthermore, the main fusion module includes a third stitching module, a first 3×3 convolutional layer, a third upsampling module, a fourth stitching module, a second 3×3 convolutional layer, a fourth upsampling module, and a third 3×3 convolutional layer;

[0037] The third splicing module is used to receive the output of the third dilated convolutional block and the second feature map, and to splice the output of the third dilated convolutional block and the second feature map.

[0038] The first 3×3 convolutional layer is used to receive the output of the third splicing module and perform convolution operations on the output of the third splicing module;

[0039] The third upsampling module is used to receive the output of the first 3×3 convolutional layer and perform upsampling operation on the output of the first 3×3 convolutional layer;

[0040] The fourth stitching module is used to receive the output of the third upsampling module and the first feature map, and to stitch the output of the third upsampling module and the first feature map together.

[0041] The second 3×3 convolutional layer is used to receive the output of the fourth splicing module and perform convolution operations on the output of the fourth splicing module;

[0042] The fourth upsampling module is used to receive the output of the second 3×3 convolutional layer and perform upsampling operation on the output of the second 3×3 convolutional layer;

[0043] The third 3×3 convolutional layer is used to receive the output of the fourth upsampling module, perform convolution operations on the output of the fourth upsampling module, and generate a fused feature map.

[0044] Furthermore, the accuracy improvement module includes a 1×3 convolutional layer, a 3×1 convolutional layer, a 1×5 convolutional layer, a 5×1 convolutional layer, a fifth stitching module, and a fourth 1×1 convolutional layer;

[0045] The 1×3 convolutional layer, 3×1 convolutional layer, 1×5 convolutional layer, and 5×1 convolutional layer receive the fused feature map, perform convolution processing, and then input it into the fifth stitching module.

[0046] The fifth stitching module receives the output feature maps of 1×3 convolutional layer, 3×1 convolutional layer, 1×5 convolutional layer and 5×1 convolutional layer respectively, and performs stitching operation;

[0047] The fourth 1×1 convolutional layer is used to receive the output feature map from the fifth stitching module and generate the final feature map.

[0048] Secondly, this application also provides a method for optimizing the probe parameters of an eddy current sensor, wherein the probe parameter optimization method uses the aforementioned probe parameter optimization system to optimize the probe parameters.

[0049] The beneficial effects of this application are:

[0050] (1) The probe parameter optimization system and method of the eddy current sensor adopts a convolutional neural network to perform three-channel comprehensive identification of the sine output signal, cosine output signal, and the difference signal between the sine output signal and the cosine output signal. It can not only perceive the distortion characteristics of the sine output signal and the cosine output signal themselves, but also perceive the amplitude inconsistency, phase quadrature offset and nonlinear error characteristics between the sine and cosine signals. In this way, a precise mapping relationship between the probe coil parameters and the output signal quality is established, and the optimization direction of the sine coil, cosine coil and excitation coil in the probe is accurately identified, thereby improving the parameter optimization efficiency and optimization accuracy.

[0051] (2) In the probe parameter optimization system of the eddy current sensor, the convolutional neural network sets up two cascaded convolutional feature extraction modules to enhance the extraction capability of shallow spatial features, expand the channel dimension, and avoid the loss of shallow texture and edge information; three cascaded Transformer modules are set up to progressively expand the channel depth, which enhances the extraction capability of amplitude inconsistency, phase orthogonal offset and nonlinear error features between sine and cosine signals; a fusion module is set up to gradually combine semantic context information with high-resolution spatial information, which can accurately extract the distortion features of sine output signal and cosine output signal itself, amplitude inconsistency, phase orthogonal offset and nonlinear error features between sine and cosine signals, and improve the accuracy of probe parameter optimization; finally, an accuracy improvement module is added to further enhance the detailed features and waveform boundary information, and improve the ability to identify amplitude and phase offset between sine and cosine signals.

[0052] (3) In the feature fusion module, the expansion fusion module aggregates multi-level features from all stages of the three-layer Transformer module in a top-down manner, gradually fusing earlier, more refined shallow spatial details from the deepest detailed features to obtain complete context information. This allows for the accurate aggregation of amplitude inconsistencies, phase orthogonal shifts, and nonlinear error features between sine and cosine signals, avoiding feature loss during progressive feature extraction and improving the accuracy of subsequent probe parameter optimization. Simultaneously, the main fusion module fuses context information and shallow spatial features to accurately aggregate the distortion features of the sine and cosine output signals themselves, amplitude inconsistencies, phase orthogonal shifts, and nonlinear error features between sine and cosine signals, further improving the accuracy of subsequent probe parameter optimization.

[0053] (4) In the accuracy improvement module, four asymmetric convolutional layers are used to enhance detailed features and waveform boundary information, thereby improving the ability to identify amplitude and phase shifts between sine and cosine signals. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A structural block diagram of the probe parameter optimization system provided in this application;

[0056] Figure 2 A structural block diagram of the convolutional neural network provided in this application;

[0057] Figure 3 A structural block diagram of the feature fusion module provided in this application;

[0058] Figure 4 A structural block diagram of the accuracy improvement module provided in this application. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0060] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the application. The singular forms “a,” “said,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. “Multiple” generally includes at least two, but does not exclude the inclusion of at least one.

[0061] This application provides a probe parameter optimization system and method for an eddy current sensor. It employs a convolutional neural network to perform three-channel comprehensive identification of the sine, cosine, and difference signals of the sine and cosine output signals. This not only detects the distortion characteristics of the sine and cosine output signals themselves, but also detects the amplitude inconsistency, phase quadrature shift, and nonlinear error characteristics between the sine and cosine signals. This establishes a precise mapping relationship between the probe coil parameters and the output signal quality, thereby accurately identifying the optimization direction of the sine coil, cosine coil, and excitation coil in the probe, improving the efficiency and accuracy of parameter optimization.

[0062] The present application will be further described below with reference to the accompanying drawings and specific embodiments.

[0063] In some embodiments of this application, such as Figure 1 As shown, this application provides a probe parameter optimization system for an eddy current sensor. This probe parameter optimization system includes a time delay processing module, a difference module, a three-channel grayscale processing module, and a convolutional neural network.

[0064] The time delay processing module is used to receive the sinusoidal output signal, perform time delay processing on the sinusoidal output signal, and generate a sinusoidal delayed signal.

[0065] For example, in an eddy current sensor, the phase difference between the sinusoidal output signal and the cosine output signal is π / 2, forming an orthogonal signal to calculate the rotor angle. Therefore, the time delay processing module performs a π / 2 phase time delay processing on the sinusoidal output signal to generate a sinusoidal delayed signal, which is aligned with the cosine output signal to facilitate the subsequent acquisition of the difference between the sinusoidal and cosine output signals.

[0066] The difference module is used to receive the sinusoidal delayed signal and the cosine output signal, perform subtraction processing on the sinusoidal delayed signal and the cosine output signal to generate a difference signal, which is used to represent the difference between the sinusoidal output signal and the cosine output signal. This difference signal carries the amplitude inconsistency and nonlinear error information between the sinusoidal output signal and the cosine output signal.

[0067] The three-channel grayscale processing module is used to receive the sine output signal, the cosine output signal, and the difference signal between the sine output signal and the cosine output signal. It performs splicing and grayscale processing on the sine output signal, the cosine output signal, and the difference signal to generate a three-channel grayscale image.

[0068] For example, the sine output signal is mapped to the pixel value of the first channel grayscale image, the cosine output signal is mapped to the pixel value of the second channel grayscale image, and the difference signal is mapped to the pixel value of the third channel grayscale image. The first channel grayscale image, the second channel grayscale image, and the third channel grayscale image are stitched together and grayscale processed to obtain a three-channel grayscale image.

[0069] A convolutional neural network is used to receive three-channel grayscale images. After feature extraction and classification of the three-channel grayscale images, nine types of optimization labels are generated to accurately identify the optimization directions of the sine coil, cosine coil and excitation coil in the probe, thereby improving the efficiency and accuracy of parameter optimization.

[0070] For example, the nine optimization tags include sine coil turns optimization, sine coil width optimization, sine coil pitch optimization, cosine coil turns optimization, cosine coil width optimization, cosine coil pitch optimization, excitation frequency optimization, excitation current amplitude optimization, and excitation coil turns optimization.

[0071] In this embodiment of the application, the probe parameter optimization system of the eddy current sensor adopts a convolutional neural network to perform three-channel comprehensive identification of the sine output signal, cosine output signal, and the difference signal between the sine and cosine output signals. It can not only perceive the distortion characteristics of the sine and cosine output signals themselves, but also perceive the amplitude inconsistency, phase quadrature offset, and nonlinear error characteristics between the sine and cosine signals. This establishes a precise mapping relationship between the probe coil parameters and the output signal quality, thereby accurately identifying the optimization direction of the sine coil, cosine coil, and excitation coil in the probe, and improving the parameter optimization efficiency and accuracy.

[0072] In some embodiments of this application, in order to accurately extract the distortion features of the sine and cosine signals themselves, the amplitude inconsistency between sine and cosine signals, the phase orthogonality shift, and the nonlinear error features, thereby improving the accuracy of probe parameter optimization, this application also discloses a convolutional neural network. For example... Figure 2 As shown, the convolutional neural network includes a first convolutional layer, a second convolutional layer, a first Transformer module, a second Transformer module, a third Transformer module, a feature fusion module, and an accuracy improvement module.

[0073] The first convolutional feature extraction module receives a three-channel grayscale image, performs shallow feature extraction, and generates a first feature map F1. This module employs one 3×3 convolutional layer and one pooling layer.

[0074] The second convolutional feature extraction module receives the first feature map F1, performs shallow feature extraction, and generates the second feature map F2. This second convolutional feature extraction module uses one 3×3 convolutional layer and one pooling layer.

[0075] The first and second convolutional feature extraction modules improve the extraction of shallow spatial features through 3×3 convolutional layers, expanding the channel dimension while reducing resolution, thus avoiding the loss of shallow texture and edge information.

[0076] The first Transformer module receives the second feature map F2, performs deep feature extraction, and generates the third feature map F3. The first Transformer module can use any known Transformer, such as VisionTransformer.

[0077] The second Transformer module receives the third feature map F3, performs deep feature extraction, and generates the fourth feature map F4. The second Transformer module can use any existing known Transformer, such as VisionTransformer.

[0078] The third Transformer module receives the fourth feature map F4, performs depth feature extraction, and generates the fifth feature map F5. The third Transformer module can use any existing known Transformer, such as VisionTransformer.

[0079] The first, second, and third Transformer modules, based on a self-attention mechanism, focus on detailed features while preserving global contextual information. Furthermore, by progressively expanding the channel depth through these three Transformer modules, the extraction capabilities for amplitude inconsistencies, phase quadrature shifts, and nonlinear error features between sine and cosine signals are further enhanced.

[0080] The feature fusion module is used to receive the first feature map F1, the second feature map F2, the third feature map F3, the fourth feature map F4, and the fifth feature map F5 to generate a fused feature map F6.

[0081] The accuracy improvement module receives the fused feature map F6, further enhances the detailed features and waveform boundary information, and generates the final feature map F7.

[0082] The classification module receives the final feature map F7, classifies the final feature map F7, and generates 9 optimized labels.

[0083] In this embodiment of the application, two cascaded convolutional feature extraction modules are set to enhance the extraction capability of shallow spatial features, expand the channel dimension, and avoid the loss of shallow texture and edge information; three cascaded Transformer modules are set to progressively expand the channel depth, enhancing the extraction capability of amplitude inconsistency, phase orthogonal shift, and nonlinear error features between sine and cosine signals; a fusion module is set to gradually combine semantic context information with high-resolution spatial information, which can accurately extract the distortion features of sine and cosine output signals, amplitude inconsistency, phase orthogonal shift, and nonlinear error features between sine and cosine signals, improving the accuracy of probe parameter optimization; finally, an accuracy improvement module is added to further enhance detailed features and waveform boundary information, improving the ability to identify amplitude and phase shifts between sine and cosine signals.

[0084] In some embodiments of this application, such as Figure 3 As shown, this application also discloses a feature fusion module, which includes a cascaded extended fusion module and a main fusion module.

[0085] The expansion fusion module is used to receive the third feature map F3, the fourth feature map F4, and the fifth feature map F5, and perform multi-scale expansion fusion.

[0086] The main fusion module is used to receive the output of the extended fusion module, the first feature map F1, and the second feature map F2, and fuse the output of the extended fusion module, the first feature map F1, and the second feature map F2 to generate a fused feature map F6.

[0087] Furthermore, the extended fusion module includes a first 1×1 convolutional layer, a first extended convolutional block, a first upsampling, a first stitching module, a second 1×1 convolutional layer, a second extended convolutional block, a second upsampling, a second stitching module, a third 1×1 convolutional layer, and a third extended convolutional block, wherein the first 1×1 convolutional layer, the second 1×1 convolutional layer, and the third 1×1 convolutional layer use 1×1 convolutional kernels.

[0088] The first 1×1 convolutional layer is used to receive the fifth feature map F5 and perform convolution operations on the fifth feature map F5.

[0089] The first dilated convolutional block is used to receive the output of the first 1×1 convolutional layer and perform dilated convolution operation on the output of the first 1×1 convolutional layer.

[0090] The first upsampling module is used to receive the output of the first dilated convolutional block and perform an upsampling operation on the output of the first dilated convolutional block.

[0091] The first splicing module is used to receive the output of the first upsampling module and the fourth feature map F4, and to splice the output of the first upsampling module and the fourth feature map F4.

[0092] The second 1×1 convolutional layer is used to receive the output of the first splicing module and perform convolution operation on the output of the first splicing module;

[0093] The second dilated convolution block is used to receive the output of the second 1×1 convolutional layer and perform dilated convolution operation on the output of the second 1×1 convolutional layer.

[0094] The second upsampling module is used to receive the output of the second dilated convolutional block and perform upsampling operation on the output of the second dilated convolutional block.

[0095] The second splicing module is used to receive the output of the second upsampling module and the third feature map F3, and to splice the output of the second upsampling module and the third feature map F3.

[0096] The third 1×1 convolutional layer is used to receive the output of the second splicing module and perform convolution operation on the output of the second splicing module;

[0097] The third dilated convolution block is used to receive the output of the third 1×1 convolutional layer and perform dilated convolution operations on the output of the third 1×1 convolutional layer.

[0098] Furthermore, each of the first, second, and third dilated convolutional blocks includes a cascaded 3×3 convolutional layer, a dilated convolutional layer, a batch normalization layer, and an activation layer.

[0099] Furthermore, the expansion rates of the first, second, and third expanded convolutional blocks gradually increase to enlarge the receptive field. Preferably, the expansion rates are 2, 4, and 6.

[0100] Specifically, in each fusion process of the expansion fusion module, the input feature map is first projected to a uniform channel dimension through a 1×1 convolutional layer, and then input into the expansion convolutional block to expand the receptive field while preserving the resolution to identify deep detail features. After that, it is input into the upsampling module to perform a stitching operation to match the spatial size of the input feature map.

[0101] To avoid contextual information loss during progressive feature extraction, the dilated fusion module aggregates multi-level features from all stages of the three Transformer modules in a top-down manner. Starting from the fifth feature map F5 corresponding to the deepest detailed features, it gradually fuses finer, shallower spatial details from earlier stages, such as the fourth feature map F4 and the third feature map F3, to obtain complete contextual information. This allows for the accurate aggregation of amplitude inconsistencies, phase orthogonality shifts, and nonlinear error features between sine and cosine signals, preventing feature loss during progressive feature extraction and improving the accuracy of subsequent probe parameter optimization. Furthermore, in each fusion process, a combination of non-dilated convolution and dilated convolution is used to accurately aggregate multi-scale features from each Transformer module while capturing global and local contextual information.

[0102] Furthermore, the main fusion module includes a third stitching module, a first 3×3 convolutional layer, a third upsampling module, a fourth stitching module, a second 3×3 convolutional layer, a fourth upsampling module, and a third 3×3 convolutional layer.

[0103] The third splicing module is used to receive the output of the third dilated convolutional block and the second feature map F2, and to splice the output of the third dilated convolutional block and the second feature map F2.

[0104] The first 3×3 convolutional layer is used to receive the output of the third splicing module and perform convolution operations on the output of the third splicing module;

[0105] The third upsampling module is used to receive the output of the first 3×3 convolutional layer and perform upsampling operation on the output of the first 3×3 convolutional layer;

[0106] The fourth splicing module is used to receive the output of the third upsampling module and the first feature map F1, and to splice the output of the third upsampling module and the first feature map F1.

[0107] The second 3×3 convolutional layer is used to receive the output of the fourth splicing module and perform convolution operations on the output of the fourth splicing module;

[0108] The fourth upsampling module is used to receive the output of the second 3×3 convolutional layer and perform upsampling operation on the output of the second 3×3 convolutional layer;

[0109] The third 3×3 convolutional layer is used to receive the output of the fourth upsampling module, perform convolution operations on the output of the fourth upsampling module, and generate the fused feature map F6.

[0110] To incorporate shallow spatial features and improve the accuracy of subsequent probe parameter optimization, the main fusion module integrates contextual information with shallow spatial features. Specifically, after each layer of fusion in the main fusion module, only 3×3 convolutional layers and upsampling modules are used for feature fusion to accurately aggregate the distortion features of the sine and cosine output signals themselves, the amplitude inconsistencies between sine and cosine signals, the phase orthogonality shift, and the nonlinear error features, thereby improving the accuracy of subsequent probe parameter optimization.

[0111] In some embodiments of this application, such as Figure 4 As shown, this application also discloses an accuracy improvement module, which includes a 1×3 convolutional layer, a 3×1 convolutional layer, a 1×5 convolutional layer, a 5×1 convolutional layer, a fifth splicing module, and a fourth 1×1 convolutional layer.

[0112] The 1×3 convolutional layer, 3×1 convolutional layer, 1×5 convolutional layer, and 5×1 convolutional layer each receive the fused feature map F6, perform convolution processing, and then input it into the fifth stitching module.

[0113] The fifth stitching module receives the output feature maps of the 1×3 convolutional layer, the 3×1 convolutional layer, the 1×5 convolutional layer, and the 5×1 convolutional layer respectively, and performs stitching operations.

[0114] The fourth 1×1 convolutional layer is used to receive the output feature map of the fifth stitching module and generate the final feature map F7.

[0115] In this embodiment of the application, the accuracy improvement module employs four asymmetric convolutional layers to enhance detailed features and waveform boundary information, thereby improving the ability to identify amplitude and phase shifts between sine and cosine signals. Specifically, the 1×3, 3×1, 1×5, and 5×1 convolutional layers capture the horizontal (1×3), vertical (3×1), and shift details (1×5, 5×1) of the sine and cosine signals, respectively. After adding detailed features and waveform boundary information, the layers are spliced ​​along the channel dimension for output, and the fourth 1×1 convolutional layer is used for dimensionality reduction to improve feature fusion efficiency.

[0116] In another embodiment of this application, a method for optimizing the probe parameters of an eddy current sensor is also provided. This method utilizes the aforementioned probe parameter optimization system to optimize the probe parameters.

[0117] The foregoing description illustrates and describes several preferred embodiments of this application. However, as previously stated, it should be understood that this application is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of this application should be within the protection scope of the appended claims.

Claims

1. A probe parameter optimization system for an eddy current sensor, characterized in that, It includes a latency processing module, an interpolation module, a three-channel grayscale processing module, and a convolutional neural network; The time delay processing module is used to receive the sinusoidal output signal, perform time delay processing on the sinusoidal output signal, and generate a sinusoidal delayed signal. The difference module is used to receive the sinusoidal delayed signal and the cosine output signal, perform difference processing on the sinusoidal delayed signal and the cosine output signal, and generate a difference signal. The three-channel grayscale processing module is used to receive sine output signals, cosine output signals, and difference signals, and generate a three-channel grayscale image after splicing and grayscale processing. A convolutional neural network is used to receive a three-channel grayscale image, extract features from the three-channel grayscale image, classify it, and generate nine optimized labels.

2. The probe parameter optimization system according to claim 1, characterized in that, The nine optimization categories include optimization of sine coil turns, sine coil width, sine coil pitch, cosine coil turns, cosine coil width, cosine coil pitch, excitation frequency, excitation current amplitude, and excitation coil turns.

3. The probe parameter optimization system according to claim 1, characterized in that, The convolutional neural network includes a first convolutional layer, a second convolutional layer, a first Transformer module, a second Transformer module, a third Transformer module, a feature fusion module, and an accuracy improvement module; The first convolutional feature extraction module is used to receive a three-channel grayscale image, perform shallow feature extraction operations, and generate a first feature map. The second convolutional feature extraction module is used to receive the first feature map, perform shallow feature extraction operations, and generate the second feature map. The first Transformer module is used to receive the second feature map, perform deep feature extraction operations, and generate the third feature map. The second Transformer module is used to receive the third feature map, perform deep feature extraction operations, and generate the fourth feature map. The third Transformer module is used to receive the fourth feature map, perform deep feature extraction, and generate the fifth feature map. The feature fusion module is used to receive the first feature map, the second feature map, the third feature map, the fourth feature map, and the fifth feature map to generate a fused feature map; The accuracy improvement module is used to receive the fused feature map, enhance detailed features and waveform boundary information, and generate the final feature map. The classification module receives the final feature map, classifies it, and generates 9 optimized labels.

4. The probe parameter optimization system according to claim 3, characterized in that, The feature fusion module includes a cascaded extended fusion module and a main fusion module; The expansion fusion module is used to receive the third feature map, the fourth feature map and the fifth feature map, and perform multi-scale expansion fusion. The main fusion module is used to receive the output of the extended fusion module, the first feature map, and the second feature map, and to fuse the output of the extended fusion module, the first feature map, and the second feature map to generate a fused feature map.

5. The probe parameter optimization system according to claim 4, characterized in that, The expanded fusion module includes a first 1×1 convolutional layer, a first expanded convolutional block, a first upsampling, a first stitching module, a second 1×1 convolutional layer, a second expanded convolutional block, a second upsampling, a second stitching module, a third 1×1 convolutional layer, and a third expanded convolutional block; The first 1×1 convolutional layer is used to receive the fifth feature map and perform convolution operations on the fifth feature map; The first dilated convolutional block is used to receive the output of the first 1×1 convolutional layer and perform dilated convolution operation on the output of the first 1×1 convolutional layer; The first upsampling module is used to receive the output of the first dilated convolutional block and perform an upsampling operation on the output of the first dilated convolutional block. The first stitching module is used to receive the output of the first upsampling module and the fourth feature map, and to perform a stitching operation on the output of the first upsampling module and the fourth feature map. The second 1×1 convolutional layer is used to receive the output of the first splicing module and perform convolution operation on the output of the first splicing module; The second dilated convolution block is used to receive the output of the second 1×1 convolutional layer and perform dilated convolution operation on the output of the second 1×1 convolutional layer. The second upsampling module is used to receive the output of the second dilated convolutional block and perform upsampling operation on the output of the second dilated convolutional block. The second stitching module is used to receive the output of the second upsampling module and the third feature map, and to perform a stitching operation on the output of the second upsampling module and the third feature map. The third 1×1 convolutional layer is used to receive the output of the second splicing module and perform convolution operation on the output of the second splicing module; The third dilated convolution block is used to receive the output of the third 1×1 convolutional layer and perform dilated convolution operations on the output of the third 1×1 convolutional layer.

6. The probe parameter optimization system according to claim 5, characterized in that, The first dilated convolutional block, the second dilated convolutional block, and the third dilated convolutional block each include a cascaded 3×3 convolutional layer, a dilated convolutional layer, a batch normalization layer, and an activation layer.

7. The probe parameter optimization system according to claim 6, characterized in that, The expansion rates of the first, second, and third expanded convolutional blocks gradually increase.

8. The probe parameter optimization system according to claim 7, characterized in that, The main fusion module includes a third stitching module, a first 3×3 convolutional layer, a third upsampling module, a fourth stitching module, a second 3×3 convolutional layer, a fourth upsampling module, and a third 3×3 convolutional layer; The third splicing module is used to receive the output of the third dilated convolutional block and the second feature map, and to splice the output of the third dilated convolutional block and the second feature map. The first 3×3 convolutional layer is used to receive the output of the third splicing module and perform convolution operations on the output of the third splicing module; The third upsampling module is used to receive the output of the first 3×3 convolutional layer and perform upsampling operation on the output of the first 3×3 convolutional layer; The fourth stitching module is used to receive the output of the third upsampling module and the first feature map, and to stitch the output of the third upsampling module and the first feature map together. The second 3×3 convolutional layer is used to receive the output of the fourth splicing module and perform convolution operations on the output of the fourth splicing module; The fourth upsampling module is used to receive the output of the second 3×3 convolutional layer and perform upsampling operation on the output of the second 3×3 convolutional layer; The third 3×3 convolutional layer is used to receive the output of the fourth upsampling module, perform convolution operations on the output of the fourth upsampling module, and generate a fused feature map.

9. The probe parameter optimization system according to any one of claims 2-8, characterized in that, The accuracy improvement module includes a 1×3 convolutional layer, a 3×1 convolutional layer, a 1×5 convolutional layer, a 5×1 convolutional layer, a fifth splicing module, and a fourth 1×1 convolutional layer; The 1×3 convolutional layer, 3×1 convolutional layer, 1×5 convolutional layer and 5×1 convolutional layer receive the fused feature map, perform convolution processing and input it into the fifth stitching module; The fifth stitching module receives the output feature maps of 1×3 convolutional layer, 3×1 convolutional layer, 1×5 convolutional layer and 5×1 convolutional layer respectively, and performs stitching operation; The fourth 1×1 convolutional layer is used to receive the output feature map from the fifth stitching module and generate the final feature map.

10. A method for optimizing probe parameters of an eddy current sensor, characterized in that, The probe parameter optimization method uses the probe parameter optimization system described in any one of claims 1-9 to optimize the probe parameters.

Images