Piezoelectric composite system vibration response image analysis and dynamic processing system

Abstract

The invention discloses a piezoelectric composite system vibration response image analysis and dynamic processing system, and relates to the technical field of image analysis and processing, and the system comprises a data acquisition module which is used for obtaining a vibration response image sequence of a piezoelectric composite material under mechanical vibration excitation, the full-field displacement data and the vibration response image sequence are synchronously collected, and the frequency, the waveform and the amplitude of the mechanical vibration excitation are all controllable; the neural network processing module is used for inputting the vibration response image sequence and the full-field displacement data into a preset single encoder-single decoder parallel multi-task head structure, and jointly outputting an enhanced vibration characteristic pattern and a sub-pixel-level displacement field in an equivalent displacement mode length single channel form; the method has the beneficial effects that the electron-hole separation efficiency is represented in real time with high precision, and the parameters of mechanical vibration excitation are automatically adjusted in real time according to the current real force-electricity coupling state of the material.

Description

Technical Field

[0001] This invention relates to the field of image analysis and processing technology, and in particular to a system for image analysis and dynamic processing of vibration response of piezoelectric composite systems. Background Technology

[0002] Piezoelectric composite materials, due to their unique force-electric coupling properties, have shown broad application prospects in fields such as self-cleaning of building exterior walls, photo / piezoelectric catalytic degradation of pollutants, energy harvesting, and intelligent vibration energy conversion. In recent years, with the advancement of green building policies, self-powered environmental catalytic systems based on the piezoelectric effect have received widespread attention. These systems can generate instantaneous polarized electric fields under environmental vibration or artificially applied mechanical excitation, driving the directional separation of photogenerated or catalytically generated electron-hole pairs, thereby achieving all-weather, power-free degradation of organic pollutants and surface self-cleaning functions. In practical applications, to optimize the electron-hole separation efficiency of piezoelectric composite materials and achieve efficient and stable catalytic performance, it is usually necessary to apply controllable mechanical vibration excitation to the material and monitor its vibration response characteristics in real time using image analysis and processing systems.

[0003] However, existing vibration response image analysis and dynamic processing systems typically employ a serial processing architecture, relying solely on traditional digital image correlation methods to independently calculate the displacement field and obtain strain through simple spatial differentiation. This fails to deeply integrate the semantic features of vibration images with the spatiotemporal dynamic characteristics of displacement information. Consequently, when processing continuous vibration sequences, the system struggles to fully exploit the spatiotemporal continuity and intrinsic correlation of the material response. Consequently, it becomes difficult to accurately assess the electron-hole separation efficiency and adjust the mechanical excitation parameters accordingly in a closed-loop manner. Ultimately, this results in significant deviations in the assessment of electron-hole separation efficiency of piezoelectric composite systems under complex actual working conditions, the inability to adaptively optimize excitation parameters, unstable catalytic performance, wasted mechanical energy, and shortened long-term service life of materials. Summary of the Invention

[0004] In view of the above-mentioned prior art, this application is hereby proposed. Embodiments of this application provide a system for vibration response image analysis and dynamic processing of piezoelectric composite systems, including:

[0005] Data acquisition module: used to acquire the vibration response image sequence of the piezoelectric composite material under mechanical vibration excitation, as well as the full-field displacement data acquired synchronously with the vibration response image sequence. The frequency, waveform and amplitude of the mechanical vibration excitation are all controllable.

[0006] Neural network processing module: used to input the vibration response image sequence and the full-field displacement data into a preset single encoder-single decoder parallel multi-task head structure, and jointly output the enhanced vibration feature map and the sub-pixel level displacement field in the form of a single channel of equivalent displacement modulus;

[0007] The fusion strain module is used to stitch the enhanced vibration feature map, the sub-pixel level displacement field in the form of the equivalent displacement modulus single channel, and the instantaneous displacement velocity field obtained by the difference of the sub-pixel level displacement fields of adjacent frames on the same pixel grid in the channel dimension, and then fuse them through at least one layer of spatiotemporal convolution to obtain the fused feature tensor.

[0008] The calculation module is used to perform channel dimension analysis on the fused feature tensor, extract displacement-related features, and reconstruct it into a dual-channel displacement component containing horizontal displacement u and vertical displacement s. Based on the dual-channel displacement component, the real-time local strain field of the piezoelectric composite material is calculated. According to the real-time local strain field and the piezoelectric voltage coefficient of the piezoelectric composite material, the electron-hole separation efficiency of the piezoelectric composite material is calculated.

[0009] Excitation adjustment module: used to dynamically adjust the frequency, waveform or amplitude of the mechanical vibration excitation based on the obtained electron-hole separation efficiency.

[0010] According to another aspect of this application, an electronic device is provided, including a memory and a processor, the memory being used to store computer-executable instructions, and the processor being used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method described above.

[0011] According to another aspect of this application, a computer storage medium is provided that stores computer-executable instructions thereon, which, when executed by a processor, implement the steps of the method described above.

[0012] Compared with the prior art, the vibration response image analysis and dynamic processing system of the piezoelectric composite system according to the embodiments of this application adopts a single encoder-single decoder structure and jointly outputs the enhanced vibration feature map and the sub-pixel level displacement field in the form of a single channel of equivalent displacement modulus. This avoids the defects of traditional serial processing that ignores image texture and dynamic information, and can capture the spatiotemporal continuity characteristics of piezoelectric composite materials in real time, reduce evaluation bias, and support high-precision real-time local strain field calculation.

[0013] By stitching together the channel dimensions and fusing spatiotemporal convolutions, a fused feature tensor is formed. The dual-channel displacement components are analyzed and reconstructed, and the real-time local strain field is calculated. This allows for better exploration of intrinsic correlations and significantly improves the spatiotemporal resolution and accuracy of the real-time local strain field. Based on the real-time local strain field and the piezoelectric voltage coefficient, the separation efficiency is calculated. The excitation adjustment module dynamically adjusts the excitation parameters accordingly to achieve adaptive optimization. This avoids performance instability caused by evaluation distortion, resulting in improved electron-hole separation efficiency, more stable catalytic performance, reduced energy consumption, and extended lifespan of the piezoelectric composite material. Attached Figure Description

[0014] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0015] Figure 1 This is a schematic diagram of the overall framework of the piezoelectric composite system vibration response image analysis and dynamic processing system of the present invention.

[0016] Figure 2 This is a schematic diagram of the dual-channel displacement component reconstruction process of the piezoelectric composite system vibration response image analysis and dynamic processing system of the present invention.

[0017] Figure 3 This is a schematic diagram of the dynamic adjustment of mechanical vibration parameters in the vibration response image analysis and dynamic processing system of the piezoelectric composite system of the present invention. Detailed Implementation

[0018] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0019] Example 1:

[0020] Reference Figures 1-3 As an embodiment of the present invention, a system for image analysis and dynamic processing of vibration response of piezoelectric composite systems is provided:

[0021] Existing systems, when processing piezoelectric composite systems, typically obtain subpixel-level displacement fields by directly regressing image sequences using the DIC algorithm, then perform simple spatial differentiation to calculate local strain, and finally obtain the average or local polarization electric field through the piezoelectric constitutive equation. This simplistic sequential processing ignores the rich texture, intensity gradient, and dynamic texture evolution information inherent in the vibration images themselves. This leads to a significant decrease in the accuracy of the extracted displacement and strain fields due to external interference, resulting in a distorted assessment of electron-hole separation efficiency. Ultimately, this leads to unstable catalytic performance, energy waste, and premature material failure in practical applications of piezoelectric composite systems.

[0022] Figure 1 The figure illustrates a vibration response image analysis and dynamic processing system for a piezoelectric composite system according to an embodiment of this application, including: a data acquisition module, a neural network processing module, a fusion strain module, a calculation module, and an excitation adjustment module.

[0023] Data acquisition module: used to acquire the vibration response image sequence of piezoelectric composite material under mechanical vibration excitation, as well as the full-field displacement data acquired synchronously with the vibration response image sequence. The frequency, waveform and amplitude of the mechanical vibration excitation are all controllable.

[0024] Furthermore, in actual operation, the data acquisition module applies mechanical vibration excitation to the piezoelectric composite material through a vibration generator. The frequency, waveform, and amplitude of this vibration excitation can be adjusted in real time through a preset control algorithm to simulate environmental vibrations in actual application scenarios. Specifically, the vibration generator can be a piezoelectric ceramic actuator or an electromagnetic vibration table. For example, when using a piezoelectric ceramic actuator, initial excitation parameters are first set (e.g., frequency of 10Hz, sinusoidal waveform, amplitude of 0.5mm). Then, the response of the piezoelectric composite material is monitored through a feedback loop, and the parameters are fine-tuned to ensure the accuracy of the excitation. The data acquisition module also uses a high-speed camera (e.g., an industrial camera with a frame rate of 1000fps) to acquire vibration response image sequences. These image sequences capture the dynamic deformation of the material surface. For example, in a typical experiment, the image sequence can be represented as a series of grayscale image matrices, with each frame measuring 512×512 pixels and a sequence length of 100 frames, recording the continuous process of the material from rest to vibration. In addition, the full-field displacement data acquired synchronously with the image sequence is obtained through a laser Doppler oscillometer (LDV) or a digital image correlation (DIC) sensor. This data is represented in the form of a two-dimensional displacement field, ensuring that the timestamp of the full-field displacement data is strictly synchronized with the frame rate of the image sequence.

[0025] In this invention, the data acquisition module is designed to meet the requirements for controllability of excitation parameters and data synchronization in existing piezoelectric composite material vibration response analysis. This approach improves the robustness and accuracy of data acquisition. For example, in noisy environments, the signal can be amplified by increasing the amplitude, improving the signal-to-noise ratio for subsequent feature extraction. In contrast, alternative techniques include using mechanical percussion devices or acoustic exciters with fixed frequencies, but these methods lack parameter flexibility and cannot adapt to variations in the material's own frequency, leading to increased data deviation. This invention chooses this approach because it seamlessly integrates with the subsequent neural network processing module, ensuring the spatiotemporal consistency of the input data.

[0026] The neural network processing module is used to input vibration response image sequences and full-field displacement data into a preset single encoder-single decoder parallel multi-task head structure, and jointly output an enhanced vibration feature map and a sub-pixel level displacement field in the form of a single channel of equivalent displacement modulus.

[0027] Figure 2 This is a schematic diagram of the dual-channel displacement component reconstruction process of the piezoelectric composite system vibration response image analysis and dynamic processing system of the present invention;

[0028] This invention designs a single encoder-single decoder parallel multi-task head optimization network structure in the neural network processing module, using vibration response image sequences as the main input and full-field displacement data as auxiliary supervision or direct splicing input, to achieve joint learning of two tasks: vibration feature enhancement and sub-pixel level displacement regression.

[0029] Furthermore, the first decoder head used to output the enhanced vibration feature map and the second decoder head used to output the sub-pixel displacement field share the multi-scale feature output of a single encoder. This operation is designed based on the need for multi-task learning efficiency and feature sharing in existing piezoelectric composite vibration response analysis. The advantage of this operation is that it reduces parameter redundancy and improves the generalization ability of the model. In contrast, alternative techniques include using independent encoders to provide features for each decoder head, but this method has a large number of parameters, unstable training, and cannot fully utilize the shared low-level features. This invention chooses this operation because it can be seamlessly integrated with the subsequent fusion strain module to ensure that the enhanced vibration feature map and the sub-pixel displacement field maintain consistency in the spatiotemporal dimensions.

[0030] Furthermore, the loss function of the second decoder head includes a displacement modulus loss term and a spatial smoothing constraint term. This operation is designed based on the requirements for modulus accuracy and spatial continuity in existing sub-pixel level displacement field regression. The advantage of this operation is that it enhances the physical consistency of the displacement field and avoids modulus deviation and non-smooth regions caused by noise. Alternative techniques include using only mean square error loss, but this method ignores the modulus constraint, resulting in discontinuities in the displacement field in low-texture regions. This invention chooses this operation because it can directly constrain the output in the form of an equivalent displacement modulus single channel.

[0031] Furthermore, the loss function of the first decoder head includes a feature reconstruction loss term and a strain gradient regularization loss term. The strain gradient regularization loss term is calculated based on the enhanced vibration feature map through an intermediate strain field derived from a preset auxiliary calculation layer. The strain gradient regularization loss term is used to constrain the spatial distribution characteristics of the intermediate strain field to maintain spatial consistency with the strain field of the preset true value.

[0032] This invention introduces a feature reconstruction loss term and a strain gradient regularization loss term in the first decoder head because of the need for reconstruction accuracy and physical constraints in existing vibration feature enhancement. The advantage of this operation is that it improves the semantic richness and strain correlation of the vibration feature map. Alternative techniques include using only adversarial loss, but this approach is unstable during training and cannot directly incorporate strain physics knowledge. This invention chooses this operation because it can bridge the vibration feature map with the intermediate strain field through an auxiliary computation layer, ensuring the effectiveness of the enhanced vibration feature map in subsequent fusion.

[0033] It should be noted that the strain gradient regularization loss term is calculated based on the intermediate strain field derived from the enhanced vibration feature map through a preset auxiliary calculation layer, including:

[0034] Based on the enhanced vibration feature map, an intermediate strain field corresponding to the real-time local strain field is derived through a preset auxiliary calculation layer. This operation is designed based on the need for implicit physical constraints in existing feature enhancement. The advantage of this operation is that it bridges low-level features and high-order strain information, avoiding overfitting from direct regression. In contrast, alternative techniques include using fully connected layers to derive the strain field, but this method has a large number of parameters and cannot preserve spatial locality. This invention chooses this operation because it can decompose the calculation step by step to ensure the physical consistency of the intermediate strain field.

[0035] The preset auxiliary calculation layer uses convolution operations, and the specific formula is as follows:

[0036] ;

[0037] in, For the intermediate strain field channel k at position The strain value at that point represents the intermediate strain intensity derived from the vibration characteristic diagram. The weight matrix for the auxiliary calculation layer represents the weights of channel k with respect to input channel c, indicating the linear mapping coefficients from features to strain. For the enhanced vibration feature map, channel c is located at... The eigenvalues ​​at a given location represent the semantic intensity of the vibration image. The value of the bias term channel k for the auxiliary calculation layer represents the reference offset of the intermediate strain field, and C is the number of channels of the key feature enhancement tensor, representing the total number of feature dimensions after fusion.

[0038] Spatial differentiation is performed on the intermediate strain field to obtain the intermediate strain gradient field. This operation is designed based on the sensitivity requirement for gradient information in existing strain field analysis. The advantage of this operation is that it captures the local rate of change of the intermediate strain field and improves the precision of regularization constraints. In contrast, alternative techniques include using Fourier transform to obtain the frequency domain gradient, but this method is computationally complex and cannot maintain pixel-level accuracy directly in the spatial domain. This invention chooses this operation because it can be seamlessly integrated with pixel-by-pixel error calculation, ensuring the physical continuity of the gradient field.

[0039] Spatial differentiation operations employ the central difference method, with the specific formula as follows:

[0040] ;

[0041] For the x-direction, or similarly for the y-direction, where, For the intermediate strain gradient field channel k at position The directional gradient component at a given location represents the local rate of change of the intermediate strain gradient field. is the pixel spacing, representing the distance between pixels in the image grid.

[0042] The pixel-wise error between the intermediate strain gradient field and the gradient field of the preset true strain field is calculated. This operation is designed based on the requirement for gradient consistency in existing loss functions. The advantage of this operation is that it strengthens the spatial distribution constraint of the intermediate strain field and avoids the smooth transition of simple mean square error. In contrast, alternative techniques include using L1 norm error, but this method is sensitive to outliers and cannot highlight the physical importance of gradient peaks. This invention chooses this operation because it can directly quantify gradient bias and ensure the physical interpretability of regularized loss.

[0043] The pixel-by-pixel error is expressed as the difference of squares, and the specific formula is as follows:

[0044] ;

[0045] in, The strain gradient regularization loss term represents the global quantization of the gradient field bias based on pixel-wise error. K is the number of strain field channels, and H and W are the height and width (in pixels) representing the spatial dimensions of the strain field. For the intermediate strain field channel k at position The gradient vector at point represents the calculated gradient distribution. The gradient field channel k of the preset true strain field is located at... The gradient vector at point represents the reference gradient distribution.

[0046] The fusion strain module is used to stitch the enhanced vibration feature map, the sub-pixel displacement field in the form of a single channel of the equivalent displacement modulus, and the instantaneous displacement velocity field obtained by the difference of the sub-pixel displacement fields of adjacent frames on the same pixel grid in the channel dimension, and then fuse them through at least one layer of spatiotemporal convolution to obtain the fused feature tensor.

[0047] It should be noted that the process of obtaining the instantaneous displacement velocity field includes the following steps:

[0048] The subpixel-level displacement field of the current frame is subtracted from the subpixel-level displacement field of the previous frame or the previous N frames at the same pixel position to obtain the displacement difference field.

[0049] Dividing the displacement difference field by the corresponding inter-frame time interval Δt yields the instantaneous displacement velocity field, which is strictly aligned with the enhanced vibration feature map and the equivalent displacement modulus single channel on the pixel grid.

[0050] The instantaneous displacement-velocity field also includes an equivalent velocity modulus single channel, and the instantaneous displacement-velocity field is used as an additional temporal feature input when concatenating the channel dimensions.

[0051] In this embodiment, the idea of ​​introducing an instantaneous displacement velocity field stems from the fact that during the vibration of piezoelectric composite materials, not only does the magnitude of the displacement (modulus) contain deformation information, but its rate of change over time (i.e., velocity) also directly reflects the instantaneous acceleration and local stress state of the material. Especially under high-frequency or non-harmonic vibration conditions, the displacement modulus alone will lose important dynamic phase information, leading to lag or distortion in subsequent strain calculations near the vibration peak. By introducing an instantaneous displacement velocity field through channel-dimensional splicing, not only is explicit temporal derivative information added to the fused feature tensor, making spatiotemporal convolution more likely to capture the characteristics of the vibration acceleration direction, but it can also provide a significant response in the transition region where the displacement modulus is small (displacement is close to zero but velocity is large), avoiding the signal "dead zone" that occurs when only the modulus is used.

[0052] At the same time, it maintains strict alignment with the enhanced vibration feature map and displacement modulus on the same pixel grid and timestamp, without the need for additional interpolation or resampling.

[0053] It should be noted that the signal "dead zone" refers to the moment when the displacement inevitably crosses zero (i.e., the material returns to its equilibrium position) within one cycle of vibration. At this time, the modulus is approximately 0 (or very small), and the displacement channel of this frame is almost entirely close to 0, providing almost no useful information. However, in reality, the piezoelectric composite material is often at its maximum velocity (and acceleration) at this time, with the highest local stress / strain change rate, making it one of the moments with the strongest piezoelectric response. If only the modulus is considered, this crucial dynamic information will be "flattened" to near zero, and the network will hardly see the signal, which is equivalent to a signal "dead zone". However, by introducing an instantaneous velocity field (or velocity modulus), even if the displacement modulus is close to zero, the velocity channel can still provide a large value, thereby recovering the dynamic characteristics of this critical moment and avoiding information loss.

[0054] Specifically, the enhanced vibration feature map, the subpixel-level displacement field in the form of a single channel of the equivalent displacement modulus, and the instantaneous displacement velocity field obtained by the difference between the subpixel-level displacement fields of adjacent frames are concatenated in the same pixel grid along the channel dimension. In other words, in this case, the instantaneous displacement velocity field V (also a single channel) is obtained by subtracting the subpixel-level displacement fields of adjacent frames and dividing by the corresponding inter-frame time interval Δt. Then, the enhanced vibration feature map E, the equivalent displacement modulus, and the instantaneous displacement velocity field are concatenated in the channel dimension to form a C+2 channel tensor, and then the fused feature tensor is obtained by at least one layer of spatiotemporal convolution.

[0055] The only existing alternative is to directly estimate the velocity field from the original image sequence using optical flow. However, this method is extremely sensitive to image noise, and the calculated instantaneous displacement velocity field cannot be guaranteed to be consistent with the sub-pixel displacement field regressed by the neural network at the pixel level, resulting in amplified errors after fusion. This invention adopts the method of "differentiating the displacement field frame by frame + dividing by Δt", which is the simplest and most direct method with the least amount of computation, and can naturally guarantee complete spatiotemporal consistency with the sub-pixel displacement field output by the neural network. Therefore, it is the optimal choice.

[0056] After at least one layer of spatiotemporal convolution fusion, a fused feature tensor is obtained, including:

[0057] Based on the concatenated tensor, the dynamic weighted coefficient vector of each channel is calculated through global average pooling and a fully connected layer. This operation stems from the need for dynamic evaluation of feature importance in existing channel attention mechanisms. Its advantage lies in adaptively highlighting key channel information and improving fusion efficiency. In contrast, alternative techniques include static weight allocation, but this method cannot adapt to the dynamic changes in the vibration sequence, leading to uneven feature utilization. This invention chooses this operation because it can seamlessly connect with subsequent channel-by-channel multiplication, ensuring the contextual relevance of the weights.

[0058] Specifically, global average pooling averages the concatenated tensors in the spatial dimension to obtain the channel description vector, which is then mapped to a dynamic weighted coefficient vector through a fully connected layer. The channel description vector represents the spatial average intensity of each channel, and the fully connected layer introduces non-linear activation to capture the inter-channel dependencies.

[0059] The dynamic weighted coefficient vector is multiplied channel-by-channel with the concatenated tensor to dynamically allocate and enhance the weights of the equivalent displacement modulus channel, instantaneous displacement velocity channel, and vibration semantic feature channel in the tensor, resulting in a key feature enhancement tensor. This operation is designed based on the need for channel recalibration in existing attention mechanisms. The advantage of this operation is that it suppresses noisy channels and amplifies relevant features, improving the input quality of subsequent convolutions. In contrast, alternative techniques include element-level additive fusion, but this method ignores the differences between channels, leading to feature conflicts. This invention chooses this operation because it can dynamically adjust the vibration characteristics of piezoelectric composite materials and enhance spatiotemporal correlation.

[0060] Specifically, channel-wise multiplication achieves weighted scaling of features by multiplying the dynamic weighting coefficient of each channel with the corresponding channel element, highlighting the equivalent displacement modulus channel (capturing displacement amplitude), the instantaneous displacement velocity channel (reflecting temporal dynamics), and the vibration semantic feature channel (containing texture information).

[0061] A 3D convolutional kernel is used to process the key feature enhancement tensor to obtain a fused feature tensor, where the size of the convolutional kernel in the time dimension t is a preset number of bits T. This operation is designed based on the need to capture spatiotemporal continuity in existing video processing. The advantage of this operation is that it integrates temporal and spatial information to form a more robust feature representation. In contrast, alternative techniques include 2D convolutional sequences, but this method ignores the direct fusion of the time dimension, resulting in the loss of dynamic evolution. This invention chooses this operation because it can handle continuous frames of vibration sequences, ensuring the spatiotemporal consistency of the fused feature tensor.

[0062] The specific formula for 3D convolution is:

[0063] ;

[0064] in, To fuse the feature tensor Channel in time ,Location The value represents the feature intensity after spatiotemporal fusion. For the 3D convolution kernel Output channel to input channel c, time offset k, spatial offset The weight, Enhance the c-th channel of the tensor for key features in time ,Location The value of represents the input feature, T is the kernel size in the time dimension, represents the temporal range captured, and P and Q are the kernel sizes in the spatial dimension.

[0065] For example, assume the size of the concatenated tensor X is [5, 16, 512, 512] (5 time frames, 16 channels, 512×512 space), where channels 1-4 are equivalent displacement moduli, 5-8 are instantaneous displacement velocities, and 9-16 are vibration semantic features; global average pooling is used to calculate the channel description vectors, such as the average value of channel 1 being 0.05 (representing the average intensity of the displacement moduli); a fully connected layer is used to obtain a dynamic weighted coefficient vector, such as the coefficient of channel 1 being 0.8 (high weight, because the displacement moduli are crucial); channel-by-channel multiplication is used to obtain Y, such as the intensity of channel 1 of Y being 0.8 times that of the original (enhancing the displacement channels); 3D convolution (T=3, P=Q=3) is used to process Y to obtain the fused feature tensor Z, where a certain output channel of the fused feature tensor Z integrates the local 9-pixel neighborhood features of the first 3 frames, and finally the fused feature tensor Z captures the spatiotemporal evolution of displacement and velocity in the sequence.

[0066] The calculation module is used to perform channel dimension analysis on the fused feature tensor, extract displacement-related features, and reconstruct them into dual-channel displacement components containing horizontal displacement u and vertical displacement s. Based on the dual-channel displacement components, the real-time local strain field of the piezoelectric composite material is calculated. According to the real-time local strain field and the piezoelectric voltage coefficient of the piezoelectric composite material, the electron-hole separation efficiency of the piezoelectric composite material is calculated.

[0067] This invention introduces a two-stage design of channel dimension parsing and lightweight convolutional reconstruction in the computing module. It uses the fused feature tensor as the main input and the equivalent displacement modulus and instantaneous displacement velocity as auxiliary information to achieve the extraction of displacement-related features and the joint reconstruction of dual-channel displacement components.

[0068] The process of extracting displacement-related features and reconstructing them into dual-channel displacement components containing horizontal displacement u and vertical displacement s includes the following steps:

[0069] Channel dimension analysis is performed on the fused feature tensor to extract displacement-related features containing equivalent displacement modulus information, instantaneous displacement velocity information, and displacement direction decoupling features. This operation is designed based on the need for decoupling multi-channel information of the fused feature tensor in existing piezoelectric composite vibration response analysis. The advantage of this operation is that it can selectively separate the core features directly related to displacement and avoid interference from irrelevant noise channels. In contrast, alternative techniques include using a global attention mechanism to weight all channels, but this method has high computational cost, is prone to overfitting, and cannot efficiently separate specific displacement-related information. This invention chooses this operation because it can be seamlessly connected with subsequent lightweight convolution to ensure that the extracted features maintain consistency in the spatiotemporal dimensions.

[0070] Based on displacement-related features, these features are mapped and reconstructed into dual-channel displacement components through single-layer or multi-layer lightweight convolution operations. This operation is designed to meet the need for efficient mapping in existing displacement reconstruction. Its advantage lies in achieving parameter-efficient feature mapping through lightweight design (such as depthwise separable convolution), avoiding parameter explosion caused by traditional fully connected layers, while preserving the spatial locality of displacement components. In contrast, alternative techniques include using multilayer perceptrons (MLPs) for global mapping, but this method ignores spatial local relationships, resulting in discontinuities in the reconstructed displacement components at material edges. This invention chooses this operation because it can directly utilize the locality advantage of convolution, significantly improving the spatiotemporal resolution and accuracy of reconstruction.

[0071] For example, assuming the fused feature tensor is a four-dimensional tensor with shape [10, 256, 128, 128] (where 10 is the number of time series frames, 256 is the number of channels, and 128×128 is the spatial resolution), after channel dimension parsing, the first 64 channels can be extracted as equivalent displacement modulus information (corresponding to modulus values ​​ranging from 0 to 5 pixels), the middle 64 channels as instantaneous displacement velocity information (corresponding to velocity values ​​ranging from -10 to 10 pixels / frame), and the last 128 channels as displacement direction decoupling features (corresponding to direction angle encoding from -π to π), thus obtaining a refined displacement-related feature sub-tensor [10, 256, 128,

[128] significantly reduces computational complexity; then, based on this displacement-related feature subtensor, two layers of lightweight convolution (the first layer uses a 3×3 depth convolution kernel, and the second layer uses a 1×1 point convolution kernel) are used to map the equivalent displacement modulus information to an amplitude basis, the instantaneous displacement velocity information to a dynamic adjustment factor, and the displacement direction decoupling feature to a direction vector, and finally reconstruct a dual-channel displacement component [10, 2, 128, 128], where the first channel is the horizontal displacement u (value range -5 to 5 pixels) and the second channel is the vertical displacement s (value range -5 to 5 pixels), ensuring that the reconstruction result is strictly aligned with the original fused feature tensor in terms of spatiotemporal resolution.

[0072] Based on dual-channel displacement components, the real-time local strain field of piezoelectric composite materials is calculated, including:

[0073] Based on the local grayscale or brightness variation characteristics of the vibration response image sequence, a real-time noise level analysis is performed on the image sequence to obtain the estimated signal-to-noise ratio of the vibration response image sequence; this operation quantifies the noise level by calculating the standard deviation of the grayscale difference between adjacent pixels in the vibration response image sequence.

[0074] The dual-channel displacement components are differentiated using the finite difference method, which employs a preset template. This operation calculates spatial partial derivatives by applying convolution templates to the horizontal displacement u and vertical displacement s channels.

[0075] The estimated signal-to-noise ratio is compared with a preset signal-to-noise ratio threshold, and the size of the preset template is dynamically adjusted based on the comparison result; this operation determines the switching of the template size by judging the threshold.

[0076] When the estimated signal-to-noise ratio is less than the preset signal-to-noise ratio threshold, a preset template of the first template size is selected. This operation selects a larger template (such as 5×5) for low estimated signal-to-noise ratio scenarios to suppress noise amplification.

[0077] When the estimated signal-to-noise ratio is greater than or equal to the preset signal-to-noise ratio threshold, a preset template of the second template size is selected, wherein the first template size is larger than the second template size; this operation selects a smaller template (such as 3×3) to preserve details for scenarios with high estimated signal-to-noise ratio.

[0078] This invention introduces a noise-adaptive finite difference template dynamic adjustment scheme when calculating real-time local strain fields. Using dual-channel displacement components as the main input and the noise level of the vibration response image sequence as auxiliary information, it optimizes the accuracy of real-time local strain field calculations. This scheme is based on observations of noise sensitivity in existing piezoelectric composite vibration response analysis: in high-noise environments, directly using a fixed small template in finite difference analysis easily leads to spurious peaks in the strain field, while a large template, although smoothing noise, blurs details. Therefore, dynamically adjusting the template size by estimating the signal-to-noise ratio of the vibration response image sequence in real time can balance noise suppression and detail preservation. This significantly improves the spatiotemporal robustness of the real-time local strain field, avoids deviations in electron-hole separation efficiency assessment caused by noise amplification, and is computationally efficient, requiring only a few additional noise analysis steps. In contrast, alternative techniques include using a fixed template in finite difference analysis, but the fixed template cannot adapt to varying noise, resulting in larger errors at low signal-to-noise ratios.

[0079] Based on the real-time local strain field and the piezoelectric voltage coefficient of the piezoelectric composite material, the electron-hole separation efficiency of the piezoelectric composite material is calculated. This calculation is based on the constitutive relationship of the piezoelectric composite material under plane stress, directly mapping the strain field to the electron-hole separation efficiency. The specific formula is as follows:

[0080] ;

[0081] in, The electron-hole separation efficiency per pixel at the current moment (value range 0–1). and piezoelectric voltage coefficient of piezoelectric composite material , where is the material's intrinsic constant, E is Young's modulus (N / m²), and v is Poisson's ratio. and These are the dimensionless normal strain components (in the x-direction and s-direction) of the actual local strain field. These strain components are derived from the displacement gradient tensor obtained by the aforementioned calculation module through finite difference calculation of the dual-channel displacement components u and s based on small deformation theory, and then symmetrically calculated. Specifically:

[0082] , , ;

[0083] The characteristic composite electric field strength of the piezoelectric composite material (unit: V / m) is a material intrinsic parameter, which is used as a known constant input after being calibrated through a static bending experiment.

[0084] For example, assume the real-time local strain field is a four-dimensional tensor of [10, 3, 128, 128] (10 represents the time frame, and the 3 channels are respectively...). , and (Typical strain range: -0.01 to 0.01) Input material parameters as follows:

[0085] = ;

[0086] ;

[0087] V=0.33;

[0088] ;

[0089] Substituting directly into the formula, we can obtain the electron-hole separation efficiency tensor [10, 1, 128, 128] (value range 0-0.95), which is close to the electron-hole separation efficiency in the high strain region (such as the position of the vibration wave peak). Approximately 0.95 (highly efficient separation), electron-hole separation efficiency in the low strain region It is close to 0.1.

[0090] Excitation adjustment module: used to dynamically adjust the frequency, waveform or amplitude of mechanical vibration excitation based on the obtained electron-hole separation efficiency.

[0091] Figure 3 This is a schematic diagram illustrating the dynamic adjustment of mechanical vibration parameters in the piezoelectric composite system vibration response image analysis and dynamic processing system of the present invention. Based on the obtained electron-hole separation efficiency, the frequency, waveform, or amplitude of the mechanical vibration excitation is dynamically adjusted, including at least one of the following scenarios:

[0092] When the electron-hole separation efficiency is less than or equal to the first preset threshold, increase the amplitude or frequency of the mechanical vibration excitation, or switch the waveform of the mechanical vibration excitation to a sine wave containing the intrinsic frequency components of the piezoelectric composite material.

[0093] When the electron-hole separation efficiency is greater than or equal to the second preset threshold and is characterized as entering the saturation region, the amplitude or frequency of the mechanical vibration excitation is reduced, or the waveform of the mechanical vibration excitation is switched to a low-power waveform; wherein, the second preset threshold is greater than the first preset threshold.

[0094] When the electron-hole separation efficiency is greater than the first preset threshold and less than the second preset threshold, the amplitude, frequency and waveform of the current mechanical vibration excitation are maintained.

[0095] For example, suppose in a green building project, after a piezoelectric composite exterior wall panel operates under continuous vibration conditions for a week, the system detects a slow decline in electron-hole separation efficiency. Although the efficiency value calculated in a single instance does not fall below the first preset threshold, the system detects through continuous monitoring that this decline is gradually intensifying. When the efficiency remains below the normal range for 2 hours and the rate of decline begins to accelerate, the system automatically adjusts the frequency of the mechanical vibration excitation from the initial 10Hz to the material's intrinsic frequency of 15Hz and switches to a sine wave containing resonant components. After receiving the adjustment notification, the project maintenance personnel promptly verified the material response, confirmed the recovery of catalytic performance, avoided further panel failure leading to interruption of the self-cleaning function, prevented potential pollutant accumulation and damage to the building's aesthetics, and ensured the long-term stability and environmental degradation efficiency of the exterior wall system.

[0096] In this embodiment, regarding the various thresholds in excitation adjustment, the first preset threshold is determined by statistically analyzing a large amount of electron-hole separation efficiency data collected from piezoelectric composite materials under standard vibration conditions, calculating the average value and confidence interval, and comprehensively evaluating them in conjunction with the technical specifications and efficiency tolerance range provided by the material manufacturer. The second preset threshold is derived based on long-term observation of the material's saturation region behavior, combined with stability analysis of the piezoelectric polarization field distribution, mainly considering the material's performance under different excitation intensities. The saturation region characterization is determined by analyzing the characteristic differences of the efficiency-electric field curves, combined with material fatigue experience and response mode analysis, aiming to effectively distinguish the transition between the linear region and the saturation region. In addition, regarding the above-mentioned first preset threshold, second preset threshold, and saturation region characterization, in practical applications, the initial thresholds can also be determined based on small-scale experimental data, and then continuously optimized through continuous monitoring and feedback, or machine learning methods can be introduced to obtain better threshold combinations through training on historical response data. This embodiment does not specifically limit these methods.

[0097] It should be noted that the adjustment mechanism of traditional piezoelectric composite system vibration response image analysis and dynamic processing systems usually adopts a fixed parameter control method, which is difficult to adapt to complex and ever-changing vibration response states, leading to untimely or over-adjustment. This embodiment designs a multi-level adjustment mechanism based on electron-hole separation efficiency, establishing a refined adjustment system that includes increasing mechanical vibration excitation, decreasing mechanical vibration excitation, and maintaining the current mechanical vibration excitation. By comprehensively considering the current efficiency value and saturation region characteristics, a three-dimensional judgment of the material state is achieved. This multi-level dynamic adjustment mechanism overcomes the single mode of traditional adjustment systems, realizing gradual and process-oriented optimization of vibration excitation. It can take corresponding level adjustment measures according to the nature, degree, and development trend of efficiency. It significantly improves the precision and intelligence of the piezoelectric composite system response, reduces material failure caused by untimely adjustment, and avoids unnecessary energy waste caused by over-adjustment, achieving optimal allocation of piezoelectric composite material resources.

[0098] Furthermore, the electron-hole separation efficiency calculated by this invention can directly reflect the photoelectrocatalytic performance of the piezoelectric composite material. For example, in pollutant degradation applications, higher separation efficiency can significantly reduce the electron-hole recombination rate, thereby improving the catalytic degradation rate. As a future direction, an external current source can be integrated into the system to further drive the directional separation of electrons and holes by applying a suitable current, thereby improving the overall photoelectrocatalytic efficiency under conditions of insufficient vibration excitation.

[0099] Example 2:

[0100] In one embodiment of the present invention, which differs from the previous embodiment, the electronic device includes one or more processors and a memory.

[0101] A processor can be a central processing unit (CPU) or other form of processing unit with data processing and / or instruction execution capabilities, and can control other components in an electronic device to perform desired functions.

[0102] The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0103] In one example, the electronic device may also include input devices and output devices, which are interconnected via a bus system and / or other forms of connection mechanisms (not shown). In addition, depending on the specific application, the electronic device may include any other suitable components.

[0104] Example 3:

[0105] Embodiments of this application may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps described in the "Exemplary Methods" section above according to the various embodiments of this application.

[0106] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0107] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not restrict the application from being implemented using the specific details described above.

[0108] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0109] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0110] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0111] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A piezoelectric composite system vibration response image analysis and dynamic processing system, characterized in that, include: Data acquisition module: used to acquire the vibration response image sequence of the piezoelectric composite material under mechanical vibration excitation, as well as the full-field displacement data acquired synchronously with the vibration response image sequence. The frequency, waveform and amplitude of the mechanical vibration excitation are all controllable. Neural network processing module: used to input the vibration response image sequence and the full-field displacement data into a preset single encoder-single decoder parallel multi-task head structure, and jointly output the enhanced vibration feature map and the sub-pixel level displacement field in the form of a single channel of equivalent displacement modulus; The fusion strain module is used to stitch the enhanced vibration feature map, the sub-pixel level displacement field in the form of the equivalent displacement modulus single channel, and the instantaneous displacement velocity field obtained by the difference of the sub-pixel level displacement fields of adjacent frames on the same pixel grid in the channel dimension, and then fuse them through at least one layer of spatiotemporal convolution to obtain the fused feature tensor. The calculation module is used to perform channel dimension analysis on the fused feature tensor, extract displacement-related features, and reconstruct it into a dual-channel displacement component containing horizontal displacement u and vertical displacement s. Based on the dual-channel displacement component, the real-time local strain field of the piezoelectric composite material is calculated. According to the real-time local strain field and the piezoelectric voltage coefficient of the piezoelectric composite material, the electron-hole separation efficiency of the piezoelectric composite material is calculated. Excitation adjustment module: used to dynamically adjust the frequency, waveform or amplitude of the mechanical vibration excitation based on the obtained electron-hole separation efficiency.

2. The piezoelectric composite system vibration response image analysis and dynamic processing system of claim 1, wherein: The process of obtaining the instantaneous displacement velocity field includes the following steps: The subpixel-level displacement field of the current frame is subtracted from the subpixel-level displacement field of the previous frame or the previous N frames at the same pixel position to obtain the displacement difference field. Divide the displacement difference field by the corresponding inter-frame time interval Δt to obtain the instantaneous displacement velocity field, wherein the instantaneous displacement velocity field is strictly aligned with the enhanced vibration feature map and the equivalent displacement modulus single channel on the pixel grid. The instantaneous displacement velocity field also includes an equivalent velocity modulus single channel, and the instantaneous displacement velocity field is used as an additional temporal feature input when the channel dimensions are stitched together.

3. The piezoelectric composite system vibration response image analysis and dynamic processing system of claim 1, wherein: In the neural network processing module, the first decoder head for outputting the enhanced vibration feature map and the second decoder head for outputting the sub-pixel displacement field share the multi-scale feature output of the single encoder; and the loss function of the second decoder head includes a displacement modulus loss term and a spatial smoothing constraint term.

4. The vibration response image analysis and dynamic processing system for piezoelectric composite systems according to claim 3, characterized in that: The loss function of the first decoder head includes a feature reconstruction loss term and a strain gradient regularization loss term. The strain gradient regularization loss term is calculated based on the intermediate strain field derived from the enhanced vibration feature map through a preset auxiliary calculation layer. The strain gradient regularization loss term is used to constrain the spatial distribution characteristics of the intermediate strain field to maintain spatial consistency with the strain field of the preset true value.

5. The vibration response image analysis and dynamic processing system for piezoelectric composite systems according to claim 4, characterized in that: The strain gradient regularization loss term is calculated based on the intermediate strain field derived from the enhanced vibration feature map through a preset auxiliary calculation layer, including: Based on the enhanced vibration feature map, an intermediate strain field corresponding to the real-time local strain field is derived through the preset auxiliary calculation layer. The intermediate strain gradient field is obtained by performing spatial differentiation on the intermediate strain field. Calculate the pixel-by-pixel error between the intermediate strain gradient field and the gradient field of the preset true strain field.

6. The vibration response image analysis and dynamic processing system for piezoelectric composite systems according to claim 1, characterized in that: The extraction of displacement-related features and their reconstruction into dual-channel displacement components including horizontal displacement u and vertical displacement s includes the following steps: Channel dimension parsing is performed on the fused feature tensor to extract displacement-related features containing equivalent displacement modulus information, instantaneous displacement velocity information, and displacement direction decoupling features; Based on the displacement-related features, they are mapped and reconstructed into the dual-channel displacement components through single-layer or multi-layer lightweight convolution operations.

7. The vibration response image analysis and dynamic processing system for piezoelectric composite systems according to claim 6, characterized in that: The calculation of the real-time local strain field of the piezoelectric composite material based on the dual-channel displacement components includes: Based on the local grayscale or brightness variation characteristics of the vibration response image sequence, a real-time noise level analysis is performed on the image sequence to obtain the estimated signal-to-noise ratio of the vibration response image sequence. The dual-channel displacement components are differentiated using a finite difference method, which uses a preset template. The estimated signal-to-noise ratio is compared with a preset signal-to-noise ratio threshold, and the size of the preset template is dynamically adjusted based on the comparison result; When the estimated signal-to-noise ratio is less than the preset signal-to-noise ratio threshold, a preset template of the first template size is selected. When the estimated signal-to-noise ratio is greater than or equal to the preset signal-to-noise ratio threshold, a preset template of the second template size is selected, wherein the first template size is greater than the second template size.

8. The vibration response image analysis and dynamic processing system for piezoelectric composite systems according to claim 1, characterized in that: The spatiotemporal convolution fusion process, which involves at least one layer, yields a fused feature tensor, including: Based on the spliced ​​tensor, the dynamic weighted coefficient vector of each channel is calculated through global average pooling and fully connected layers; The dynamic weighted coefficient vector is multiplied channel by channel with the concatenated tensor to dynamically allocate and enhance the weights of the equivalent displacement modulus channel, instantaneous displacement velocity channel and vibration semantic feature channel in the tensor, thereby obtaining the key feature enhancement tensor. The key feature enhancement tensor is processed using a three-dimensional convolution kernel to obtain the fused feature tensor, wherein the size of the convolution kernel in the time dimension t is a preset number of bits T.

9. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method as described in any one of claims 1 to 8.

10. A computer storage medium storing computer-executable instructions thereon, characterized in that: When the computer-executable instructions are executed by a processor, they implement the steps of the method as described in any one of claims 1 to 8.

Images