Method for testing mechanical properties of carbon fiber reinforced composite material based on image recognition

Abstract

The invention discloses a method for testing mechanical properties of a carbon fiber reinforced composite material based on image recognition, and relates to the technical field of material tests.The method comprises the steps that a carbon fiber sample of a fiber laying layer is prepared, the width tolerance is controlled through laser cutting, and speckle patterns are sprayed on the surface; training a U-Net + + neural network based on the collected macro / micro image, and analyzing the dynamic change of the fiber angle in real time; fusing the acquired macro / micro image with the dynamic change of the fiber angle, and constructing a constitutive equation of dynamic fiber orientation distribution; and importing the constitutive equation into a finite element model, iteratively optimizing the fiber-matrix interface strength parameters through inversion calculation, performing blind test verification on the sample by using the optimized finite element model, and outputting a cross-scale correlation test report. According to the method, an interface strength parameter is subjected to inversion iteration optimization through a finite element model, a macro-micro strain field, a fiber slip angle and an interface damage parameter are integrated through a cross-scale correlation test report, and full-chain data support is provided for a composite material structure.

Description

Technical Field

[0001] The present invention relates to the technical field of material testing, in particular to a method for testing the mechanical properties of carbon fiber reinforced composite materials based on image recognition. Background Art

[0002] In recent years, carbon fiber reinforced composite materials and mechanical property testing have gradually tended to develop in the direction of multi-scale and high-precision. Macroscopic DIC, microscopic optical components and SEM observations have been heavily invested in strain field analysis and damage characterization. The finite element model combined with the inversion calculation algorithm has further promoted the calibration of interface strength parameters. Among the existing methods, most rely on single-scale data or offline fusion methods, which makes it difficult to achieve the spatiotemporal synchronization of macro / micro deformations, and seriously lacks the ability to capture the dynamic slip behavior of fibers in real time. The existing constitutive equations are usually based on the static homogenization assumption, but do not introduce the dynamic coupling mechanism of fiber orientation evolution, so that there are errors in the inversion of interface strength parameters.

[0003] The limitations of the finite element inversion calculation method, based on static constitutive equations, are constrained by theoretical assumptions, making it difficult to dynamically modify finite element model parameters using experimental data. When calibrating the fiber-matrix interface strength, it is easy to fall into a local optimal solution, which reduces the reliability of the prediction of the mechanical behavior of the composite material. Multi-scale image acquisition protocols lack sub-pixel spatiotemporal alignment capabilities, resulting in inaccurate cross-scale data matching, error accumulation, and an inability to accurately analyze the correlation between fiber slip angle and interface damage. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for testing the mechanical properties of carbon fiber reinforced composite materials based on image recognition to solve the problems of real-time dynamic monitoring of multi-scale damage evolution and cross-scale precise calibration of fiber-matrix interface strength parameters.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for testing the mechanical properties of carbon fiber reinforced composite materials based on image recognition, which comprises preparing a carbon fiber specimen of a fiber layup, controlling the width tolerance by laser cutting, and spraying a speckle pattern on the surface;

[0008] Perform sub-pixel coordinate mapping on the macro DIC camera, micro-optical components, and in-situ scanning electron microscope, establish a multi-scale image synchronous acquisition protocol, and output the fiber slip angle;

[0009] The sample is loaded in stages, a threshold is set, and the fiber slip angle change rate is calculated in real time. When the fiber slip angle change rate exceeds the threshold, the macro / micro image synchronous acquisition is triggered;

[0010] Based on the collected macro / micro images, the U-Net++ neural network is trained to analyze the dynamic changes of fiber angles in real time;

[0011] The collected macro / micro images are integrated with the dynamic changes of fiber angles to construct the constitutive equation of dynamic fiber orientation distribution;

[0012] The constitutive equation is imported into the finite element model, and the fiber-matrix interface strength parameters are iteratively optimized through inverse calculation. The optimized finite element model is used to perform blind test verification on the specimen, and a cross-scale correlation test report is output.

[0013] As a preferred embodiment of the method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to the present invention, the carbon fiber specimens for preparing the fiber layup are cut by laser to control the width tolerance and the speckle pattern is sprayed on the surface. The specific steps are as follows:

[0014] According to the fiber layup angle and thickness parameters, the prepreg layup process is used to prepare the carbon fiber reinforced resin matrix composite material plate;

[0015] Laser cutting equipment is used to cut composite material plates and control the width tolerance of the specimens;

[0016] A high-contrast speckle pattern is sprayed on the surface of the cut sample to pre-treat the surface flatness.

[0017] As a preferred solution of the method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to the present invention, the following specific steps are performed: sub-pixel coordinate system mapping is performed on the macroscopic DIC camera, microscopic optical components and in-situ scanning electron microscope, a multi-scale image synchronous acquisition protocol is established, and the fiber slip angle is output.

[0018] The macro DIC camera, microscopic optical components, and in situ SEM were independently calibrated;

[0019] Align the calibrated macro DIC camera and micro-optical assembly to the same area of the specimen, and establish a sub-pixel spatial coordinate system mapping relationship using a checkerboard calibration plate;

[0020] Based on the electron beam scanning signal of the in-situ scanning electron microscope and the exposure signal of the micro-optical component, a hardware-triggered synchronization mechanism is used to establish a time dimension alignment protocol;

[0021] According to the spatial coordinate system mapping relationship and time dimension alignment protocol, the macro / micro equipment is controlled to synchronously acquire images;

[0022] The synchronously collected macro / micro images are cross-scale registered to output fiber slip angle data.

[0023] As a preferred embodiment of the method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to the present invention, the sample is subjected to graded loading, a threshold is set, and the fiber slip angle change rate is calculated in real time. When the fiber slip angle change rate exceeds the threshold, the macro / micro image synchronous acquisition is triggered. The specific steps are as follows:

[0024] The specimen is divided into multiple loading stages according to the predicted mechanical property curve, and the target load value and the critical threshold value of the fiber slip angle change rate are set for each stage.

[0025] Apply axial load to the specimen step by step, and enter the load holding state after each loading level reaches the target load value;

[0026] Based on the fiber slip angle data, the current slip angle change rate is calculated in real time;

[0027] If the slip angle change rate exceeds the critical threshold of the fiber slip angle change rate, a hardware trigger signal is sent to the macro / micro image acquisition device; if it does not exceed the critical threshold of the fiber slip angle change rate, the current stage is continuously monitored;

[0028] Receive synchronous image data collected by macro / micro imaging equipment and store it in association with the current loading stage.

[0029] As a preferred solution of the method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to the present invention, wherein: the U-Net++ neural network is trained based on the collected macro / micro images to analyze the dynamic changes of fiber angles in real time. The specific steps are as follows:

[0030] Perform data preprocessing on stored macro / micro images;

[0031] The preprocessed macro / micro images are fed into the U-Net++ neural network, which is trained using dynamic weighted cross entropy loss with fiber slip angle data as the supervisory signal.

[0032] The trained U-Net++ neural network is used to analyze the dynamic changes of fiber angles in real time.

[0033] As a preferred solution of the method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to the present invention, the steps of fusing the collected macro / micro images with the dynamic changes of fiber angles to construct the constitutive equation of dynamic fiber orientation distribution are as follows:

[0034] Temporal alignment of macro / micro images to extract multi-scale fiber distribution characteristics under the same mechanical state;

[0035] Perform pixel-level spatial matching between the dynamic changes of fiber angles and the multi-scale fiber distribution characteristics of macro / micro images at the corresponding moment;

[0036] Based on the fiber angle dynamic change data and macro / micro image data after pixel-level spatial matching, the constitutive equation of dynamic fiber orientation distribution is constructed.

[0037] As a preferred solution of the method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to the present invention, the constitutive equation is introduced into the finite element model, and the fiber-matrix interface strength parameters are iteratively optimized by inverse calculation. The specific steps are as follows:

[0038] Define the constitutive equations as material properties and import them into the finite element model;

[0039] Run finite element simulations, extract simulation data and compare residuals with stored macro / micro image data;

[0040] When the residual exceeds the error range, the fiber-matrix interface strength parameters are adjusted using inverse calculation and the simulation is repeated;

[0041] When the residual does not exceed the error range, the optimized interface strength parameters are output and the finite element model is updated.

[0042] As a preferred solution of the method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to the present invention, wherein: the optimized finite element model is used to perform blind testing on the sample and output a cross-scale correlation test report. The specific steps are as follows:

[0043] Prepare new specimens with the same layup parameters as the original specimens as blind test specimens, use laser cutting to control the width tolerance, and spray speckle;

[0044] Input the new specimen into the optimized finite element model for blind test verification;

[0045] Through blind test verification data and measured data, cross-scale comparison is performed to generate a correlation test report.

[0046] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the method for testing the mechanical properties of carbon fiber reinforced composite materials based on image recognition as described in the first aspect of the present invention is implemented.

[0047] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the method for testing the mechanical properties of carbon fiber reinforced composite materials based on image recognition as described in the first aspect of the present invention is implemented.

[0048] The beneficial effects of the present invention are as follows: the present invention improves the accuracy and efficiency of mechanical property testing of composite materials through multi-scale collaborative architecture and dynamic data drive, realizes spatiotemporal synchronous acquisition of macroscopic DIC, microscopic optical components and SEM images based on hardware trigger protocol and sub-pixel coordinate system mapping, and combines the real-time analysis of fiber slip angle by U-Net++ neural network to solve the problem of accumulated cross-scale data registration errors in traditional methods, constructs dynamic constitutive equations and parameter inversion optimization, integrates real-time fiber slip angle data with macro-micro images, constructs constitutive equations containing fiber orientation evolution terms, and breaks through the limitations of static homogenization theory; it iteratively optimizes interface strength parameters through finite element model inversion, and integrates macro-micro strain fields, fiber slip angles and interface damage parameters in cross-scale correlation test reports to provide full-chain data support for composite material structures.

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0050] Figure 1 This is a flow chart of the method for testing the mechanical properties of carbon fiber reinforced composite materials based on image recognition in Example 1.

[0051] Figure 2 This is a flowchart for outputting a cross-scale association test report in Example 1. DETAILED DESCRIPTION

[0052] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0053] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0054] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0055] Example 1, with reference to Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a method for testing the mechanical properties of carbon fiber reinforced composite materials based on image recognition, comprising the following steps:

[0056] S1. Prepare carbon fiber specimens of fiber layup, control width tolerance by laser cutting, and spray speckle pattern on the surface.

[0057] Furthermore, according to the fiber layup angle and thickness parameters, a prepreg layup process is used to prepare a carbon fiber reinforced resin matrix composite material plate;

[0058] It should be noted that the fiber layup angle θ is defined as the arrangement angle of the prepreg unidirectional fiber layer relative to the specimen axis (e.g., 0°, ±45°, 90°). Its physical significance directly affects the anisotropic modulus (Ex / Ey) of the specimen. It is necessary to limit θ∈[0°,90°], and the angle difference between adjacent layers must be ≥30° to avoid interlaminar shear stress concentration (experimental data show that when the angle difference is <30°, the interlaminar shear strength decreases by 12%-18%).

[0059] The thickness parameter t refers to the nominal thickness of a single layer of prepreg (usually t = 0.125mm or 0.25mm). The thickness tolerance of a single layer is ≤ ±0.005mm, otherwise the final specimen thickness will fluctuate by more than ±2%.

[0060] Specifically, prepreg is used instead of wet lay-up, and its precise resin content is utilized to avoid porosity defects. The porosity of prepreg molding is ≤0.5%, while the porosity of wet lay-up is 1.2%-2.5%, resulting in an 18%-25% decrease in interlaminar shear strength. By reducing the porosity, the prepreg fiber alignment error is <0.5°, while the manual lay-up error is >2°, which increases the compressive strength of the 0° lay-up specimen from 1,520MPa to 1,680MPa, thereby improving the fiber orientation accuracy.

[0061] Laser cutting equipment is used to cut composite material plates and control the width tolerance of the specimens;

[0062] Specifically, the laser cutting parameters are controlled, the laser power P is set to 200W (wavelength 1064nm), the power density is >1.5×10 6The cutting speed v is set to v=5mm / s (for 2mm thick specimens), and the heat input Q is controlled to be ≤50J / mm² (when Q>80J / mm², the matrix resin thermally degrades and the interfacial shear strength decreases by 22%). The width tolerance Δw is limited to Δw=±0.05mm (compared to ±0.1mm for traditional mechanical cutting). This is achieved by a laser spot diameter of ≤0.03mm. For every 0.01mm reduction in width error, the strain measurement error in subsequent tensile testing can be reduced by 0.3%. Through the coordinated control of heat input Q and spot diameter, a "cold cutting" effect is achieved, suppressing the heat-affected zone (HAZ). Metallographic observation shows that the HAZ width at the laser-cut incision is ≤10μm (compared to 50-80μm for traditional waterjet cutting), and the interfacial shear strength retention rate is increased from 78% to 95%.

[0063] A high-contrast speckle pattern is sprayed on the surface of the cut sample to pre-treat the surface flatness.

[0064] It should be noted that high-contrast speckle spraying solves the problem of full-field strain measurement accuracy. Through the combined effect of two-component materials and surface pretreatment, the recognition stability of the speckle pattern is improved and the resolution capability of the DIC camera is enhanced. At the same pixel density (1200dpi), speckles with a contrast of >90% increase the displacement measurement resolution from 0.01 pixels to 0.005 pixels (the error is reduced by 50%), eliminate the interference of surface roughness, and after pretreatment, the strain calculation error of the surface with Ra≤0.8μm is reduced from 3.2% to 0.9%.

[0065] Specifically, the spraying material uses a double-component speckle of aluminum oxide and black graphite with a particle size distribution of 1-5μm (SEM measurement) and a contrast ratio of >90% (grayscale difference ≥180). The contrast ratio of traditional single-material speckle is only 60%-70%, resulting in a 1.5-fold increase in the DIC (digital image correlation) displacement analysis error. The surface flatness pretreatment uses mechanical polishing (sandpaper grit 800#) + acetone wiping to make the surface roughness Ra ≤ 0.8μm (untreated sample Ra = 1.5-2.2μm). When Ra > 1.2μm, the speckle pattern will produce local distortion, resulting in a full-field strain calculation error ≥ 3%.

[0066] S2. Perform sub-pixel coordinate mapping on the macro DIC camera, micro-optical components, and in-situ scanning electron microscope, establish a multi-scale image synchronous acquisition protocol, and output the fiber slip angle.

[0067] Furthermore, the macro DIC camera, microscopic optical components, and in situ SEM were independently calibrated;

[0068] It should be noted that the macro DIC camera is a strain measurement device based on optical imaging. It calculates the full-field strain distribution by tracking the displacement changes of the speckle pattern sprayed on the sample surface during the loading process. The microscopic optical component combines an optical microscope with digital imaging to observe the sample microstructure (such as fiber arrangement, crack propagation) and local strain distribution. The in-situ scanning electron microscope is an electron microscope with an integrated mechanical loading device. It can apply force / heat / electric loads to the sample in a vacuum or low vacuum environment and observe its microstructural evolution in real time.

[0069] Specifically, by independently calibrating the inherent distortion of the device and eliminating optical distortion, the image edge positioning error is reduced from ±3.2μm to ±0.5μm after microscope lens distortion correction.

[0070] Align the calibrated macro DIC camera and micro-optical assembly to the same area of the specimen, and establish a sub-pixel spatial coordinate system mapping relationship using a checkerboard calibration plate;

[0071] It should be noted that the core purpose of aligning the same area of the sample is to achieve spatial correlation of cross-scale data, solve the problem of multi-scale data faults caused by field of view deviation in traditional multi-scale observations, and use the checkerboard to establish a cross-scale coordinate system conversion matrix. Sub-pixel coordinate system mapping refers to the use of image interpolation algorithms to increase the physical space resolution to less than 1 / 10 of the device pixel size, ensuring that the positioning error is <0.05 pixels. The checkerboard calibration plate parameters refer to the use of chromium-deposited silicon substrates, grid size tolerance ±0.1μm, thermal expansion coefficient ≤0.5ppm / ℃ (20-50℃ temperature control environment), to avoid calibration errors >0.5μm caused by thermal drift.

[0072] Based on the electron beam scanning signal of the in-situ scanning electron microscope and the exposure signal of the micro-optical component, a hardware-triggered synchronization mechanism is used to establish a time dimension alignment protocol;

[0073] It should be noted that the electron beam scanning signal is a timing electrical signal that controls the SEM electron beam to scan the sample surface line by line, and determines the electron beam scanning path, dwell time and image acquisition rate. The exposure signal is an electrical pulse that controls the photosensitivity time of the microscope camera's photoelectric sensor (such as CMOS / CCD), and determines the start / end time of image acquisition and exposure time.

[0074] The hardware-triggered synchronization mechanism uses FPGA hardware to trigger the electron beam scanning signal and the exposure signal of the microscope optical component, achieving a trigger delay of ≤100ns between the devices and aligning the timing. The difference between the SEM electron beam scanning period (10ms) and the DIC camera exposure time (9.8ms) is less than 0.2ms, ensuring a time synchronization error of less than 2% cycle.

[0075] According to the spatial coordinate system mapping relationship and time dimension alignment protocol, the macro / micro equipment is controlled to synchronously acquire images;

[0076] It should be noted that spatial coordinate system mapping is to transform the imaging coordinates of the macro DIC camera, microscopic optical components, and in situ SEM into the same physical space through mathematical transformations (e.g., affine transformation, perspective transformation), ensuring that the pixel positions of the multi-scale images correspond one-to-one to the actual geometric positions of the specimen;

[0077] The time dimension alignment protocol ensures strict alignment (microsecond accuracy) of image acquisition moments of macro and micro devices through hardware triggering and clock synchronization, thus avoiding distortion of dynamic process observations due to time differences.

[0078] Specifically, the macro / micro devices are controlled to synchronously capture images. First, the spatial transformation matrix and time synchronization parameters are loaded, and the geometric overlap area between the SEM scanning path and the DIC field of view is pre-calculated. Second, after the SEM starts scanning, the FPGA sends a trigger pulse to the DIC camera and micro-optical component in real time, triggering the electron beam scanning signal and the exposure signal of the micro-optical component to be strictly aligned with the SEM electron beam scanning position. Then, the SEM collects nanoscale morphology images at 1kHz, the DIC camera synchronously records the full-field strain distribution at 1000fps, and the micro-optical component captures the mesoscale damage evolution at 500fps. Finally, the fiber slip coordinates of the SEM image are converted to the corresponding position of the DIC strain field according to the spatial mapping matrix, and the strain-slip timing relationship is established through timestamp alignment.

[0079] The synchronously collected macro / micro images are cross-scale registered to output fiber slip angle data.

[0080] It should be noted that cross-scale image registration solves the problem of macro-micro data fusion, and adopts SIFT feature point matching and affine transformation matrix joint optimization to achieve sub-pixel registration.

[0081] Specifically, the SIFT algorithm extracted >1000 feature points in the SEM image (5000×5000 pixels), with a matching error RMS <0.1 pixel (RMS >2 pixels when not aligned), resulting in a displacement mapping error of <0.05% between the macroscopic strain field and the microscopic fiber slip. The fiber slip angle was calculated using the aligned displacement gradient tensor, and its standard deviation was reduced from ±2.1° to ±0.3°.

[0082] S3. Perform graded loading on the specimen, set a threshold, and calculate the fiber slip angle change rate in real time. When the fiber slip angle change rate exceeds the threshold, trigger the synchronous acquisition of macro / micro images.

[0083] Furthermore, the specimen is divided into multiple loading stages according to the predicted mechanical property curve, and the target load value and the critical threshold of the fiber slip angle change rate are set for each stage;

[0084] It should be noted that the mechanical property prediction curve is defined as a load-displacement curve based on the constitutive equation of the composite material and the fitting of historical test data. It is used to predict the nonlinear response of the specimen at each loading stage. The hierarchical loading stage division solves the problems of loading data overload and omission of key events. The loading stage is adaptively divided based on the mechanical prediction curve, replacing the traditional constant amplitude loading, and achieving "dense sampling in high load resolution range (such as near the yield point) and sparse sampling in low risk range".

[0085] The target load values are divided based on the nonlinear characteristics of the material constitutive curve. According to the tensile test data, when the tangent modulus of the stress-strain curve drops to 90% of the initial elastic modulus, the first-level target load is defined to avoid premature entry into the plastic phase. The second-level target load is defined in the strain hardening phase, and the third-level target load is defined in the softening phase to improve the data resolution of the damage initiation interval.

[0086] The critical threshold of the fiber slip angle change rate is the critical rate criterion for interface shear failure, and its calculation is based on the material interface properties.

[0087] Apply axial load to the specimen step by step, and enter the load holding state after each loading level reaches the target load value;

[0088] It should be noted that the axial load parameters for step-by-step loading are dynamically adjusted based on the material yield strength. The upper and lower limits of the load increment must be controlled. Exceeding the upper limit will cause the interface shear stress to surge beyond the critical value (e.g., carbon fiber / epoxy resin system) during single-stage loading, leading to irreversible damage. Exceeding the lower limit will result in excessive levels and data redundancy.

[0089] The load-holding state solves the interference of transient noise and stress relaxation, and uses the holding stage to suppress the signal drift caused by the viscoelastic effect of the material, while providing a stable mechanical environment for cross-scale image acquisition.

[0090] Based on the fiber slip angle data, the current slip angle change rate is calculated in real time;

[0091] Specifically, the sliding window difference method is used to calculate the instantaneous rate of change, replacing the traditional offline post-processing.

[0092] If the slip angle change rate exceeds the critical threshold of the fiber slip angle change rate, a hardware trigger signal is sent to the macro / micro image acquisition device; if it does not exceed the critical threshold of the fiber slip angle change rate, the current stage is continuously monitored;

[0093] It should be noted that the current slip angle change rate is calculated in real time and compared with the set critical threshold of the fiber slip angle change rate. The slip angle change rate is used as the trigger indicator of damage evolution, making the fiber slip angle change rate data acquisition more intelligent and realizing adaptive test control based on damage characteristics. If the slip angle change rate exceeds the critical threshold of the fiber slip angle change rate, the fiber slip angle change rate is used as a trigger signal to synchronize the macro / micro image acquisition with the damage evolution process, thereby establishing a multi-scale damage correlation relationship. If the fiber slip angle change rate does not exceed the critical threshold, the current stage is continuously monitored, and the sampling frequency is adaptive. In the low load stage, low-frequency sampling (10Hz) is used to reduce the amount of data. In the high load stage, the sampling is increased to 1kHz to capture potential damage events (such as microcrack initiation).

[0094] Receive synchronous image data collected by macro / micro imaging equipment and store it in association with the current loading stage.

[0095] It should be noted that by associating and storing image data with the loading stage, it helps to analyze the feature evolution of different damage stages in the subsequent process and improve the accuracy of damage identification and modeling.

[0096] Failure mode recognition and early warning based on DIC+AE multimodal signal fusion.

[0097] Furthermore, acoustic emission (AE) signals were collected synchronously during the specimen loading process to construct a joint monitoring of DIC strain field and AE time-frequency characteristics;

[0098] It should be noted that the acoustic emission sensor (bandwidth 20kHz-1MHz, sensitivity ≥80dB) is placed at the clamping end of the sample, and the elastic wave signal generated by the interface damage is captured by the piezoelectric ceramic probe. The characteristic parameters of the AE signal (including energy, count, amplitude, and rise time) form a multimodal correlation with the DIC strain field data.

[0099] Specifically, wavelet transform was used to extract the time-frequency characteristics of the AE signal. The original AE signal was subjected to a five-layer wavelet packet decomposition (mother wavelet db5), and the sub-band energy ratio of the third layer (125-250kHz) and the fourth layer (62.5-125kHz) was extracted as a damage sensitivity indicator. When the sub-band energy ratio threshold exceeded 1.5 times the baseline value (undamaged state), it was determined to be the onset of interface debonding. Combined with the area with a local strain gradient ≥0.15 / mm² in the DIC full-field strain data, the damage initiation position was located (spatial positioning accuracy ±0.5mm).

[0100] The AE feature parameters and DIC strain field data are input into the improved 3D ResNet-50 network for multimodal fusion;

[0101] Specifically, the network input channels were designed, with channel 1 being the DIC full-field strain tensor (512×512×3), and channel 2 being the AE time-frequency map (converted to a 256×256 grayscale image via STFT). An attention mechanism was used to dynamically weight multimodal features, and an SE module was introduced into the bottleneck layer to increase the AE feature weight to 0.78 (initial weight 0.5) during the damage initiation stage. The output layer adopted a dual-branch structure: branch 1 predicted the failure mode category (fiber fracture / interface debonding / matrix cracking), and branch 2 outputted the damage warning level (levels 1-5, corresponding to a residual strength attenuation rate of 5%-30%).

[0102] S4. Based on the collected macro / micro images, the U-Net++ neural network is trained to analyze the dynamic changes of fiber angles in real time.

[0103] Furthermore, data preprocessing is performed on the stored macro / micro images;

[0104] It should be noted that macro / micro images refer to image data at two different scales. Macro images are used to observe overall structural changes, such as the overall slip of fiber bundles and interfacial cracking, while micro images provide local details, such as interfacial failure between fibers and the matrix and the initiation and propagation of microcracks. Since the damage mechanism of fiber-reinforced materials usually develops at multiple scales, the introduction of the concept of macro / micro images is reasonable and necessary.

[0105] Since the original collected images may have problems such as noise, uneven lighting, and low contrast, data preprocessing can improve image quality and make subsequent network training more robust.

[0106] The preprocessed macro / micro images are fed into the U-Net++ neural network, which is trained using dynamic weighted cross entropy loss with fiber slip angle data as the supervisory signal.

[0107] It should be noted that U-Net++ is an improved U-Net network with added dense skip connections, including better information flow transmission capabilities and fine-grained feature fusion capabilities. It is particularly suitable for segmentation and feature extraction tasks. Because the change of fiber slip angle involves complex texture features, U-Net++ is more suitable for processing this task than ordinary CNN networks. Therefore, it is reasonable to use U-Net++ for image parsing.

[0108] Dynamic fiber angle changes refer to the angle changes that occur in fibers during loading due to stress. Since the deformation of fiber-reinforced materials is affected by factors such as the load path and the mechanical properties of the matrix-fiber interface, angle changes are important parameters characterizing damage evolution. Therefore, real-time analysis of these dynamic changes can provide key data support for material failure prediction.

[0109] The dynamic weighted cross-entropy loss function is a variant of the cross-entropy loss. By dynamically adjusting the category weights, the network can pay more attention to a small number of categories (such as damaged areas) and improve segmentation accuracy. In the fiber slip angle analysis task, the number of pixels in the damaged area is relatively small.

[0110] Specifically, the dynamic weight cross entropy loss function and weight adjustment formula are:

[0111]

[0112] ;

[0113] in, Expressed as the loss function value, Expressed as the total number of categories, Represents the current category index, It represents the starting index of the summation, starting from the first category (c=1), It is expressed as traversing and summing all categories. Expressed as dynamic class weights, Represented as the one-hot encoding of the true label, Indicates that the model predicts that the sample belongs to the category The probability of Expressed as the effect of controlling for category frequency, Represents the influence of controlling the training process, It is represented as the total number of samples in the training set, Represented as category The number of samples, Indicates the current training round, starting from 1 and increasing.

[0114] It should be noted that the dynamic category weights are obtained through the organic combination of category frequency statistics and training process attenuation terms, which achieves precise balance of category distribution, adaptive training control and strong parameter interpretability.

[0115] The trained U-Net++ neural network is used to analyze the dynamic changes of fiber angles in real time.

[0116] It should be noted that after the training is completed, U-Net++ is used to process the new macro / micro image data in real time to obtain the dynamic changes of the fiber slip angle.

[0117] S5. The collected macro / micro images are integrated with the dynamic changes of fiber angles to construct the constitutive equation of dynamic fiber orientation distribution.

[0118] Furthermore, the macro / micro images are temporally aligned to extract multi-scale fiber distribution features under the same mechanical state;

[0119] It should be noted that by temporally aligning macro / micro images, we ensure consistency in the temporal dimension of image data at different scales, enabling subsequent analysis to be based on the same loading state. Since macro / micro images may drift in time due to sensor sampling frequency and lighting conditions, temporal alignment is key to ensuring data matching.

[0120] The key to extracting multi-scale fiber distribution characteristics under the same mechanical state is to ensure the temporal and spatial consistency of image data at different scales (macro and micro) and to accurately reflect the stress state of the material at that moment.

[0121] Perform pixel-level spatial matching between the dynamic changes of fiber angles and the multi-scale fiber distribution characteristics of macro / micro images at the corresponding moment;

[0122] It should be noted that the pixel-level matching method is used to accurately match the time-synchronized image data with the angle change data in space, thereby ensuring the consistency of the image data and the angle change data during the construction of the constitutive equation. After adopting the pixel-level matching method, the fiber angle change at each pixel point can be accurately associated with its corresponding microstructure, thereby improving the accuracy of the data and making the constitutive equation more consistent with the actual material.

[0123] The dynamic change of fiber angle refers to the instantaneous arrangement of fibers under stress. This distribution determines the local stress transfer path and overall mechanical properties of the material. Compared with the static orientation distribution, the dynamic distribution can more accurately describe the deformation mechanism during loading and can be used to predict the evolution trend of the material.

[0124] Based on the fiber angle dynamic change data and macro / micro image data after pixel-level spatial matching, the constitutive equation of dynamic fiber orientation distribution is constructed.

[0125] It should be noted that in material mechanics, the constitutive equation describes the stress-strain relationship of a material under external forces. By fusing image data and angle change information, a dynamic constitutive equation that changes with time and load is constructed to improve its prediction accuracy.

[0126] Through pixel-level data fusion, the constitutive equation can be dynamically adjusted over time, space, and loading conditions to better conform to the material's true mechanical behavior. High-precision image data is used for real-time analysis, allowing the constitutive equation to be dynamically corrected based on real observational data. Dynamic fiber orientation distribution enables finite element analysis (FEA) or other numerical simulation methods to more accurately simulate the material's evolution process, improving the agreement between the calculated results and experimental data.

[0127] Specifically, the constitutive equation is in the form of:

[0128] ;

[0129] in, Represented as a composite material in Cartesian coordinates , The stress component in the direction, dimension / unit is Pascal (Pa), Expressed as the component directions of the stress tensor, Expressed as the component directions of the stress tensor, Expressed as the component directions of the strain tensor, Expressed as the component directions of the strain tensor, It is expressed as a fourth-order tensor, which characterizes the material stiffness characteristics, and its value varies with the fiber slip angle Dynamic changes, dimension / unit is Pascal (Pa), Represented as a composite material in Cartesian coordinates The strain component in the direction, dimension / unit is dimensionless (or %), It is expressed as the real-time slip angle of the fiber relative to the initial layup direction, with the dimension / unit being degree (°). Expressed as the shear stress at the fiber-matrix interface versus the slip angle The function of the change, the dimension / unit is Pascal (Pa), It is expressed as the rate of change of the fiber slip angle with time, and its dimension / unit is radian / second (rad / s). Expressed as a dynamic stiffness tensor, It is expressed as interface damage dynamics.

[0130] The AE cumulative energy E_AE is introduced into the dynamic constitutive equation as a damage state variable, and the expression is:

[0131] ;

[0132] in, Expressed as the acoustic emission-mechanical coupling coefficient, Expressed as the acoustic emission energy release rate, Expressed as the damage energy quantified by acoustic emission technology, It is the core parameter that describes the time accuracy of the dynamic process.

[0133] S6. Import the constitutive equation into the finite element model, iteratively optimize the fiber-matrix interface strength parameters through inverse calculation, use the optimized finite element model to perform blind test verification on the specimen, and output a cross-scale correlation test report.

[0134] Furthermore, the constitutive equations are defined as material properties and imported into the finite element model;

[0135] It should be noted that the constitutive equations of dynamic fiber orientation distribution are directly imported into the finite element model as material properties, which enables the finite element calculation to reflect the evolution of the real microstructure;

[0136] The finite element model refers to dividing a continuum into a finite number of small units and establishing discretized mathematical equations on the divided units to approximate the entire physical behavior. It is necessary to establish a composite material containing fibers and a matrix, define its material properties and structural shape, and import it into the finite element model.

[0137] Run finite element simulations, extract simulation data and compare residuals with stored macro / micro image data;

[0138] It should be noted that the residual analysis is carried out by combining the finite element calculation results with the experimental macro / micro image data, rather than relying solely on numerical calculations. By establishing the geometric structure of the fiber-matrix composite material, dividing the finite element mesh, selecting the appropriate unit type (such as tetrahedral unit, hexahedral unit), using a refined mesh in the interface area to improve the calculation accuracy, and using an adaptive meshing method to dynamically adjust the mesh density during the calculation process, applying boundary conditions (fixing part of the material to prevent rigid body displacement), and applying loads, applying axial tension, shear or bending loads, which are consistent with the actual experimental loading conditions. Due to the nonlinear behavior of the interface of the composite material, a nonlinear finite element solver is used. In each calculation step, the fiber orientation distribution and the interface slip mechanical response are calculated, and the data are recorded. The simulation is run to obtain mechanical information such as stress, strain, displacement, and fiber orientation.

[0139] After the simulation calculation is completed, the stress field and strain field information of different loading stages are extracted, the slip displacement of the interface elements is extracted, and the trend of the interface shear stress change is calculated. The fiber direction data in the simulation is processed using an image recognition algorithm (such as U-Net++) to obtain the fiber orientation distribution. Since the time steps of the experimental data and the simulation data may be different, the two need to be aligned in time.

[0140] When the residual exceeds the error range, the fiber-matrix interface strength parameters are adjusted using inverse calculation and the simulation is repeated;

[0141] It should be noted that in composite materials, the fiber-matrix interface strength parameter is an important mechanical parameter to measure the interaction between the fiber and the matrix. It determines the overall mechanical properties of the composite material, especially its performance under loads such as tension, shear, and compression. It includes interface shear strength, interface bond strength, and interface fracture toughness. Interface shear strength indicates the shear resistance between the fiber and the matrix, that is, whether the fiber will slip or debond from the matrix when the interface is subjected to shear load. Interface bond strength describes the adhesion between the fiber and the matrix, which determines the bonding strength between the two. Interface fracture toughness describes the energy required for interface destruction and is a comprehensive reflection of interface strength.

[0142] Residual refers to the error between the simulation results and the experimental data. Interpolation methods (such as linear interpolation and spline interpolation) are used to resample the experimental data to match the simulation time step, and residual calculations are performed on the mechanical parameters (stress, strain, slip, etc.). For the fiber orientation distribution, pixel-level error calculation is used. When the residual exceeds the error range, inversion calculation is used to adjust the fiber-matrix interface strength parameters and re-simulate.

[0143] When the residual does not exceed the error range, the optimized interface strength parameters are output and the finite element model is updated;

[0144] It should be noted that when the residual does not exceed the error range, the simulation results are considered to be consistent with the measured data, the final optimized fiber-matrix interface strength parameters are output, and the finite element model is updated.

[0145] Prepare new specimens with the same layup parameters as the original specimens as blind test specimens, use laser cutting to control the width tolerance, and spray speckle;

[0146] It should be noted that by constraining the preparation of new specimens with the same layup parameters, the influence of material batch differences on the verification results is eliminated, the width tolerance is controlled to the sub-micron level by laser cutting, and combined with speckle spraying, the accuracy of capturing local deformation by the digital image correlation (DIC) algorithm is improved.

[0147] Simultaneously analyze the differences in AE parameters between simulation and experiment when comparing residuals;

[0148] Specifically, the difference analysis of simulation and experimental AE parameters is realized through spatiotemporal synchronous mapping and multimodal data fusion. First, based on the hardware-triggered timestamp alignment and spatial coordinate transformation matrix, the interface shear stress field output by the finite element simulation is matched with the experimental AE energy mutation event at the same loading stage and spatial position; then, the time series correlation curve between the simulated interface stress peak and the experimental AE energy release rate is extracted. When the simulated stress value exceeds ±7% of the corresponding AE characteristic point in the experiment, the inversion calculation is triggered, and the genetic algorithm is used to dynamically adjust the interface strength parameters to synchronously optimize the AE event rate correlation and stress-strain residual of the simulation and experiment. Finally, the cross-scale consistency verification of multi-physics field data (mechanics + acoustic emission) is achieved through iterative approximation.

[0149] Input the new specimen into the optimized finite element model for blind test verification;

[0150] It should be noted that blind test verification refers to relying on the optimized finite element model to make predictions when the specific performance of the sample is unknown, and comparing the predicted results with the actual experimental measurement results to evaluate the accuracy of the finite element model. Blind testing is different from conventional verification tests. It examines the predictive ability of the finite element model for new samples without directly using the original test data as input. It uses the optimized finite element model for blind testing, introduces multi-scale damage criteria (such as the micro-interface debonding criterion), and uses parameter inversion to ensure that the model maintains high accuracy in untrained data (blind test samples).

[0151] Through blind test verification data and measured data, cross-scale comparison is performed to generate a correlation test report.

[0152] It should be noted that cross-scale refers to the comparison of the mechanical properties of specimens at different scales (macro / micro). The performance of composite materials is often affected by factors such as microscopic fiber orientation and interfacial bonding. Therefore, it is necessary to test and verify at both the microscopic (fiber orientation, interfacial slip) and macroscopic (overall mechanical response) levels. By comparing cross-scale data, the relationship between macroscopic properties (such as strength and modulus) and microstructure (such as interfacial damage and fiber slip) can be revealed, thereby improving the finite element model.

[0153] The core of the correlation test report lies in cross-scale data fusion and confidence quantification. First, the macroscopic strain field distribution of the blind test sample is synchronously collected through DIC, and the microscopic damage characteristics are obtained by combining in-situ observation with a scanning electron microscope (SEM). Then, a multi-source data fusion algorithm is used to map and align the macroscopic strain field and microscopic damage parameters according to spatial coordinates. By defining a cross-scale damage contribution index, the nonlinear effect of microscopic defects on the macroscopic mechanical response is quantified. Finally, Monte Carlo simulation is used to calculate the confidence interval and generate a structured report containing a "strain-damage" correlation cloud map, a confidence heat map, and the weights of key failure modes.

[0154] This embodiment also provides a computer device, which is suitable for the case of a carbon fiber reinforced composite material mechanical property testing method based on image recognition, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the carbon fiber reinforced composite material mechanical property testing method based on image recognition proposed in the above embodiment.

[0155] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0156] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for testing the mechanical properties of carbon fiber reinforced composite materials based on image recognition as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0157] In summary, the present invention improves the accuracy and efficiency of mechanical property testing of composite materials through: multi-scale collaborative architecture and dynamic data drive, realizes spatiotemporal synchronous acquisition of macroscopic DIC, microscopic optical components and SEM images based on hardware trigger protocol and sub-pixel coordinate system mapping, combines the real-time analysis of fiber slip angle by U-Net++ neural network, solves the problem of cross-scale data registration error accumulation in traditional methods, constructs dynamic constitutive equations and parameter inversion optimization, integrates real-time fiber slip angle data with macro and micro images, constructs constitutive equations containing fiber orientation evolution terms, breaks through the limitations of static homogenization theory; it iteratively optimizes interface strength parameters through finite element model inversion, and integrates macro and micro strain fields, fiber slip angles and interface damage parameters in cross-scale correlation test reports to provide full-chain data support for composite material structures.

[0158] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for testing the mechanical properties of carbon fiber reinforced composite materials based on image recognition, characterized by: include, Prepare carbon fiber specimens of fiber layup, use laser cutting to control width tolerance, and spray speckle pattern on the surface; Perform sub-pixel coordinate mapping on the macro DIC camera, micro-optical components, and in-situ scanning electron microscope, establish a multi-scale image synchronous acquisition protocol, and output the fiber slip angle; The sample is loaded in stages, a threshold is set, and the fiber slip angle change rate is calculated in real time. When the fiber slip angle change rate exceeds the threshold, the macro / micro image synchronous acquisition is triggered; Based on the collected macro / micro images, the U-Net++ neural network is trained to analyze the dynamic changes of fiber angles in real time; The collected macro / micro images are integrated with the dynamic changes of fiber angles to construct the constitutive equation of dynamic fiber orientation distribution; The constitutive equation is imported into the finite element model, and the fiber-matrix interface strength parameters are iteratively optimized through inverse calculation. The optimized finite element model is used to perform blind test verification on the specimen, and a cross-scale correlation test report is output.

2. The method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to claim 1, wherein: The carbon fiber sample of the fiber layer is prepared by laser cutting to control the width tolerance and spraying the speckle pattern on the surface. The specific steps are as follows: According to the fiber layup angle and thickness parameters, the prepreg layup process is used to prepare the carbon fiber reinforced resin matrix composite material plate; Laser cutting equipment is used to cut composite material plates and control the width tolerance of the specimens; A high-contrast speckle pattern is sprayed on the surface of the cut sample to pre-treat the surface flatness.

3. The method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to claim 2, wherein: The sub-pixel coordinate system mapping of the macro DIC camera, micro-optical components and in-situ scanning electron microscope is performed, a multi-scale image synchronous acquisition protocol is established, and the fiber slip angle is output. The specific steps are as follows: The macro DIC camera, microscopic optical components, and in situ SEM were independently calibrated; Align the calibrated macro DIC camera and micro-optical assembly to the same area of the specimen, and establish a sub-pixel spatial coordinate system mapping relationship using a checkerboard calibration plate; Based on the electron beam scanning signal of the in-situ scanning electron microscope and the exposure signal of the micro-optical component, a hardware-triggered synchronization mechanism is used to establish a time dimension alignment protocol; According to the spatial coordinate system mapping relationship and time dimension alignment protocol, the macro / micro equipment is controlled to synchronously acquire images; The synchronously collected macro / micro images are cross-scale registered to output fiber slip angle data.

4. The method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to claim 3, wherein: The sample is loaded in stages, a threshold is set, and the fiber slip angle change rate is calculated in real time. When the fiber slip angle change rate exceeds the threshold, the macro / micro image synchronous acquisition is triggered. The specific steps are as follows: The specimen is divided into multiple loading stages according to the predicted mechanical property curve, and the target load value and the critical threshold value of the fiber slip angle change rate are set for each stage. Apply axial load to the specimen step by step, and enter the load holding state after each loading level reaches the target load value; Based on the fiber slip angle data, the current slip angle change rate is calculated in real time; If the slip angle change rate exceeds the critical threshold of the fiber slip angle change rate, a hardware trigger signal is sent to the macro / micro image acquisition device; if it does not exceed the critical threshold of the fiber slip angle change rate, the current stage is continuously monitored; Receive synchronous image data collected by macro / micro imaging equipment and store it in association with the current loading stage.

5. The method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to claim 4, characterized in that: The U-Net++ neural network is trained based on the collected macro / micro images to analyze the dynamic changes of fiber angles in real time. The specific steps are as follows: Perform data preprocessing on stored macro / micro images; The preprocessed macro / micro images are fed into the U-Net++ neural network, which is trained using dynamic weighted cross entropy loss with fiber slip angle data as the supervisory signal. The trained U-Net++ neural network is used to analyze the dynamic changes of fiber angles in real time.

6. The method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to claim 5, characterized in that: The collected macro / micro images are integrated with the dynamic changes of fiber angles to construct the constitutive equation of dynamic fiber orientation distribution. The specific steps are as follows: Temporal alignment of macro / micro images to extract multi-scale fiber distribution characteristics under the same mechanical state; Perform pixel-level spatial matching between the dynamic changes of fiber angles and the multi-scale fiber distribution characteristics of macro / micro images at the corresponding moment; Based on the fiber angle dynamic change data and macro / micro image data after pixel-level spatial matching, the constitutive equation of dynamic fiber orientation distribution is constructed.

7. The method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to claim 6, characterized in that: The constitutive equation is introduced into the finite element model, and the fiber-matrix interface strength parameters are iteratively optimized through inverse calculation. The specific steps are as follows: Define the constitutive equations as material properties and import them into the finite element model; Run finite element simulations, extract simulation data and compare residuals with stored macro / micro image data; When the residual exceeds the error range, the fiber-matrix interface strength parameters are adjusted using inverse calculation and the simulation is repeated; When the residual does not exceed the error range, the optimized interface strength parameters are output and the finite element model is updated.

8. The method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition according to claim 7, characterized in that: The optimized finite element model is used to perform blind testing on the sample and output a cross-scale correlation test report. The specific steps are as follows: Prepare new specimens with the same layup parameters as the original specimens as blind test specimens, use laser cutting to control the width tolerance, and spray speckle; Input the new specimen into the optimized finite element model for blind test verification; Through blind test verification data and measured data, cross-scale comparison is performed to generate a correlation test report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for testing mechanical properties of carbon fiber reinforced composite materials based on image recognition are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for testing the mechanical properties of carbon fiber reinforced composite materials based on image recognition according to any one of claims 1 to 8 are implemented.

Images