A method and system for monitoring defects in composite materials based on fiber grating sensors

Abstract

The application discloses a composite material defect monitoring method and system based on a fiber grating sensor. The method comprises the following steps: implanting the fiber grating sensor in the composite material and forming an array, then acquiring the center wavelength offset measured by each fiber grating sensor in real time by using a signal demodulation device, calculating the corresponding stress gradient, temperature change rate or vibration frequency offset according to the center wavelength offset, constructing a defect feature map, finally performing defect detection by using a defect detection module, and performing multi-level early warning according to the result of the defect detection. The application solves the technical problems that the implantation process of the fiber grating sensor in the composite material is complex, the interface bonding strength between the sensor and the base material is difficult to guarantee, and multi-parameter and multi-scale defect monitoring is difficult to realize in the prior art, and more accurate identification and positioning of defects in the composite material structure are realized.

Description

Technical Field

[0001] This invention belongs to the field of composite material structural health monitoring technology, specifically relating to a composite material defect monitoring method and system based on fiber optic grating sensors. Background Technology

[0002] Composite materials are widely used in aerospace, civil engineering, and shipbuilding due to their excellent mechanical properties, designability, and corrosion resistance. However, composite materials are prone to defects such as delamination, cracking, and debonding during manufacturing and use, which can seriously affect the load-bearing capacity and safety of structures. Therefore, real-time, online monitoring of composite material structures to promptly detect and locate defects is of great significance for ensuring the safe operation of structures.

[0003] Traditional non-destructive testing techniques, such as ultrasonic testing and X-ray testing, suffer from drawbacks such as low efficiency, high cost, and difficulty in achieving online monitoring. Fiber Bragg grating (FBG) sensors, with their advantages of small size, light weight, resistance to electromagnetic interference, corrosion resistance, and distributed measurement capabilities, have been widely used in the field of composite material structural health monitoring in recent years. However, in existing technologies, the implantation process of FBG sensors in composite materials is complex, the interfacial bonding strength between the sensor and the matrix material is difficult to guarantee, and it is challenging to achieve multi-parameter, multi-scale defect monitoring. Summary of the Invention

[0004] To address the problems existing in the background technology, the present invention provides a method and system for detecting defects in composite materials based on fiber Bragg grating sensors. This solves the technical problems in the prior art, such as the complex implantation process of fiber Bragg grating sensors in composite materials, the difficulty in ensuring the interfacial bonding strength between the sensor and the matrix material, and the difficulty in achieving multi-parameter and multi-scale defect monitoring.

[0005] The technical solution adopted in this invention is:

[0006] I. A method for detecting defects in composite materials based on fiber Bragg grating sensors:

[0007] S1. Select several fiber Bragg grating sensors and process each fiber Bragg grating sensor with a coating material to obtain the fiber Bragg grating sensor after the corresponding process.

[0008] S2. Divide the monitoring area according to the structure of the composite material, implant the processed fiber optic grating sensor into the divided monitoring area and record the implantation position of each fiber optic grating sensor. Connect all the implanted fiber optic grating sensors into an array to obtain a fiber optic grating sensor array.

[0009] S3. The center wavelength offset measured by each fiber optic grating sensor in the fiber optic grating sensor array is obtained in real time through the signal demodulation device.

[0010] S4. Obtain the corresponding stress gradient, temperature change rate, or vibration frequency offset based on the center wavelength offset measured in real time by each fiber Bragg grating sensor.

[0011] S5. Construct a defect feature map based on the stress gradient, temperature change rate, and vibration frequency offset at the same time.

[0012] S6. Collect continuous defect features from the current time and the preset time period prior to it. Figure 1 The input is fed into a pre-trained defect recognition model for defect detection, and the real-time defect detection results of the composite material are obtained. Multi-level early warnings are then issued based on the defect detection results.

[0013] In step S1, the fiber Bragg grating sensor includes a fiber Bragg grating stress sensor, a fiber Bragg grating temperature sensor, and a fiber Bragg grating vibration sensor; the coating material includes epoxy resin and polyimide resin; the process adopts a precision coating process, which includes electrostatic spraying and dip coating.

[0014] Step S2 specifically involves:

[0015] S21. Based on the structure of the composite material, determine the stress concentration area and potential defect distribution area of ​​the composite material, and designate the stress concentration area and potential defect distribution area as the key monitoring area, and the rest as the global monitoring area.

[0016] S22. Several fiber Bragg grating stress sensors are implanted in the stress concentration areas of the key monitoring areas using a high-density method; several fiber Bragg grating stress sensors, fiber Bragg grating temperature sensors, and fiber Bragg grating vibration sensors are implanted in the potential defect distribution areas of the key monitoring areas using a high-density method; several fiber Bragg grating stress sensors, fiber Bragg grating temperature sensors, and fiber Bragg grating vibration sensors are implanted in the global monitoring area using a low-density method.

[0017] S23. Record the position of all implanted fiber Bragg grating sensors and connect all implanted fiber Bragg grating sensors into an array to obtain a fiber Bragg grating sensor array.

[0018] Step S21 further includes: dividing the stress concentration area into a unidirectional stress area and a multidirectional stress area; for the unidirectional stress area, multiple fiber optic stress sensors are arranged along the stress direction; for the multidirectional stress area, a fiber optic stress sensor is arranged along each stress direction.

[0019] In step S22, the high density refers to arranging 10-15 fiber Bragg grating sensors per square meter, and the low density refers to arranging 3-5 fiber Bragg grating sensors per square meter.

[0020] The implantation methods in step S22 include interlayer implantation, intra-seam implantation, and three-fiber braided implantation; the implantation process includes prepreg laying, resin transfer molding, and fiber winding.

[0021] The fiber optic grating sensor array in step S23 can be in the form of a linear array, a planar array, or a three-dimensional array.

[0022] The stress gradient and temperature change rate are set according to the following formulas:

[0023] dε / dx=1 / (λ B (1-P e ))·d(Δλ B ) / dx

[0024] dT / dt=1 / K T ·d(Δλ A ) / dt

[0025] Where dε / dx is the stress gradient; dT / dt is the rate of temperature change; P e λ is the effective elastic-optical coefficient of the optical fiber; B d(Δλ) is the center wavelength of the fiber grating. B ) / dx is the rate of change of the center wavelength offset along the x-direction, as measured by the fiber optic stress sensor; K T d(Δλ) is the temperature sensitivity coefficient. A ) / dt is the rate of change of the center wavelength offset over time as measured by the fiber Bragg grating temperature sensor.

[0026] The specific steps for obtaining the vibration frequency offset are as follows: the center wavelength offset measured in real time by the fiber optic grating vibration sensor is processed to convert the center wavelength offset into a frequency spectrum. One or more main vibration frequencies of the composite material structure are identified from the frequency spectrum. The average value of the main vibration frequencies is subtracted from the reference vibration frequency measured in the defect-free state of the composite material structure to obtain the vibration frequency offset.

[0027] Step S5 specifically involves:

[0028] The composite material is first divided into three-dimensional meshes, which are then divided into N×N meshes.

[0029] S51. At the same time, normalize all the stress gradients, temperature change rates and vibration frequency offsets obtained, and obtain each normalized stress gradient, temperature change rate and vibration frequency offset.

[0030] S52. The values ​​of each grid are filled in sequentially according to the normalized values ​​of stress gradient, temperature change rate and vibration frequency offset at the location. For locations where no value exists, the value is filled with 0. All the filled values ​​are arranged in a two-dimensional N×N feature map to obtain an N×N stress gradient feature map, temperature change rate feature map and vibration frequency offset feature map respectively. The three feature maps are spliced ​​together to obtain the defect feature map at a single time.

[0031] Follow these steps to build a defect dataset for pre-training the defect identification model:

[0032] The composite material is first divided into three-dimensional meshes, consisting of N×N grids.

[0033] F1. At the same time, after normalizing all the stress gradients, temperature change rates and vibration frequency offsets obtained, we can obtain each normalized stress gradient, temperature change rate and vibration frequency offset.

[0034] F2. The values ​​of each grid are filled in sequentially according to the normalized values ​​of stress gradient, temperature change rate and vibration frequency offset at the location. For locations where no value exists, the value is filled with 0. All the filled values ​​are arranged in a two-dimensional N×N feature map to obtain an N×N stress gradient feature map, temperature change rate feature map and vibration frequency offset feature map respectively. The three feature maps are spliced ​​together to obtain the defect feature map at a single time.

[0035] F3. At the same time, label each grid in N×N according to the condition of the composite material to obtain a label map.

[0036] F4. At the same time, the defect feature map is used as input and the label map is used as output to obtain the sample data at a single time. The sample data at several consecutive time points are used as the defect data.

[0037] F5. Several defect data points are aggregated to obtain a defect dataset.

[0038] The defect identification model includes a first spatial feature extraction model, a second spatial feature extraction model, a temporal feature extraction model, a multi-source spatiotemporal fusion model, and a classification model. The defect feature map at the current moment is input into the spatial feature extraction model for processing to obtain a spatial feature map. Continuous defect feature maps from the current moment and previous preset time periods are input into the temporal feature extraction model for processing to predict a defect feature map at a future moment. The predicted defect feature map is then input into the second spatial feature extraction model for processing to obtain a spatiotemporal feature map, a spatial feature map, and a spatiotemporal feature map. Figure 1The input is fed into a multi-source spatiotemporal fusion model for processing to obtain multi-source spatiotemporal fusion features. The multi-source spatiotemporal fusion features are then fed into a classification model for processing to obtain defect detection results.

[0039] The first and second spatial feature extraction models both use ResNet networks; the temporal feature extraction model uses LSTM networks; and the classification model uses multi-layer fully connected networks.

[0040] The multi-source spatiotemporal fusion model is set according to the following formula:

[0041] Output=A⊙F fused

[0042] A = G⊙Softmax(S); F fused =[F s ,F t ]

[0043] G=σ(FC1(W g Std(F s ))+FC2(U g ·Avg(BN(F t )))+b g S = tanh(W) a Flatten(M))+b a

[0044] M=φ(∑ i=1 N ∑ j=1 P (W ij (F s,i ★F t,j )))

[0045] F s,i ★F t,j =F s,i ·F t,j / ((||F s,i ||2·||F t,j ||2+ε) α )

[0046] Where Output represents the multi-source spatiotemporal fusion feature output by the multi-source spatiotemporal fusion model; A represents the final gated feature; ⊙ represents element-wise multiplication; F fused G represents the fusion feature; G represents the statistical gating feature; Softmax() represents the Softmax activation function; S represents the collaborative gating feature; F s Represents spatial feature map; F t Represents a spatiotemporal feature map; [F s ,F t ] represents the spatial feature map Fs Spatiotemporal feature map F t Perform splicing; σ() represents the Sigmoid activation function; FC1 and FC2 both represent fully connected layers; W g U g and W a All represent weights; Std() represents standard deviation; Avg() represents mean; BN() represents batch normalization; tanh() represents activation function; Flatten() represents flattened layer; M represents co-current tensor; b g and b a All represent bias; φ() represents the nonlinear transformation function; i and j are both indices; N represents the spatial feature map F. s The total number of channels; P represents the spatiotemporal feature map F. t The total number of channels; F s,i This indicates that by analyzing the spatial feature map F s The feature vector obtained by applying a fully connected layer to the i-th channel; F t,j This indicates that by analyzing the spatiotemporal feature map F t The feature vector obtained by applying a fully connected layer to the j-th channel; W ij Represents the weight; ||F s,i ||2 represents F s,i Take the L2 norm; ||F t,j ||2 pairs of F t,j Take the L2 norm; ε is a constant to prevent the denominator from being zero; α represents the learnable parameter; F s,i ·F t,j This represents the dot product of two eigenvectors.

[0047] II. A composite material defect monitoring system based on fiber Bragg grating sensors:

[0048] A fiber Bragg grating sensor array is composed of several fiber Bragg grating sensors embedded in a composite material.

[0049] The signal demodulation device is used to acquire the center wavelength offset measured by each fiber optic grating sensor in the fiber optic grating sensor array in real time.

[0050] The data processing module calculates the corresponding stress gradient, temperature change rate, or vibration frequency shift based on the center wavelength offset measured in real time by each fiber Bragg grating sensor, and constructs a defect feature map.

[0051] The defect detection module stores pre-trained defect recognition models and performs defect detection.

[0052] The early warning module provides multi-level warnings based on defect detection results.

[0053] The beneficial effects of this invention are:

[0054] 1. Advanced features: This invention uses special coating materials and processes to treat fiber optic grating sensors, which improves the interfacial bonding strength and durability between the sensor and the composite matrix, and solves the problems of easy sensor detachment and unstable signal in the prior art.

[0055] 2. Operability: The operation process of this invention, from sensor placement and signal demodulation to data processing and early warning, is clear and easy to implement.

[0056] 3. Originality and innovation: The composite material defect identification model based on multi-parameter and multi-scale monitoring proposed in this invention can more accurately identify and locate defects in composite material structures.

[0057] 4. The defect feature map constructed by this invention and the defect recognition model proposed have higher defect detection accuracy, stronger multi-source spatiotemporal information fusion capability and better generalization performance, which can effectively improve the recognition rate of small defects in complex scenarios. Attached Figure Description

[0058] Figure 1 This is a flowchart of the method of the present invention.

[0059] Figure 2 This is a block diagram of the system of the present invention.

[0060] Figure 3 This is an external model diagram of the composite material wing in the embodiment.

[0061] Figure 4 This is a diagram of the internal structure of the composite material wing in the embodiment. Detailed Implementation

[0062] The present invention will now be described in more detail with reference to the accompanying drawings and embodiments. However, the present invention is not limited thereto. For those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention. Contents not described in detail in this specification are prior art known to those skilled in the art.

[0063] Example 1:

[0064] like Figure 1 As shown, the composite material defect monitoring method of this embodiment includes the following steps:

[0065] S1. Select several fiber Bragg grating sensors and process each fiber Bragg grating sensor with a coating material to obtain the processed fiber Bragg grating sensors.

[0066] In step S1, the fiber Bragg grating sensor includes a fiber Bragg grating stress sensor, a fiber Bragg grating temperature sensor, and a fiber Bragg grating vibration sensor; the coating material includes epoxy resin and polyimide resin; the process adopts a precision coating process, which includes electrostatic spraying and dip coating.

[0067] In practical applications, these fiber Bragg grating sensors possess advantages such as high precision and resistance to electromagnetic interference, making them suitable for monitoring composite materials. Epoxy resin exhibits good adhesion and chemical corrosion resistance, while polyimide resin possesses properties such as high temperature resistance and high strength.

[0068] In practice, the selected fiber Bragg grating sensor has a center wavelength in the range of 1520nm to 1570nm to ensure compatibility. The coating material undergoes the following processing:

[0069] Under normal operating conditions, epoxy resin coating is used: dip coating is adopted, the sensor is immersed in low viscosity environmental resin, pulled up at a uniform speed, and the coating thickness is controlled at 10-20μm, and then cured at 80℃ for 2 hours.

[0070] Under high-temperature conditions, polyimide resin coating is used: electrostatic spraying is employed with a spraying pressure of 0.3-0.5 MPa, and the coating thickness is controlled at 15-25 μm. Curing is then completed under stepped heating conditions (150℃ / 1h + 200℃ / 1h). The treated sensor requires interfacial bonding strength testing to ensure that its shear strength with the composite matrix is ​​not less than 20 MPa.

[0071] S2. Divide the monitoring area according to the structure of the composite material, implant the processed fiber optic grating sensor into the divided monitoring area and record the implantation position of each fiber optic grating sensor. Connect all the implanted fiber optic grating sensors into an array to obtain a fiber optic grating sensor array.

[0072] Composite materials include those used in the aerospace field.

[0073] S21. Based on the structure of the composite material, determine the stress concentration area and potential defect distribution area of ​​the composite material, and designate the stress concentration area and potential defect distribution area as the key monitoring area, and the rest as the global monitoring area.

[0074] Furthermore, stress concentration areas (such as holes, corners, joints, etc.) and potential defect distribution areas (such as ply overlaps, abrupt changes in material properties, etc.) can be determined through finite element analysis or based on historical fault data.

[0075] For example, stress concentration areas are often found at the corners and around holes in composite materials, while potential defect distribution areas may be found at the joints of materials or in areas where impurities exist inside.

[0076] S22. Several fiber Bragg grating stress sensors are implanted in a high-density manner in the stress concentration areas of the key monitoring areas; several fiber Bragg grating stress sensors, fiber Bragg grating temperature sensors, and fiber Bragg grating vibration sensors are implanted in a high-density manner in the potential defect distribution areas of the key monitoring areas. This allows for more accurate monitoring of stress, temperature, and vibration changes in the key monitoring areas; several fiber Bragg grating stress sensors, fiber Bragg grating temperature sensors, and fiber Bragg grating vibration sensors are implanted in a low-density manner in the global monitoring area, realizing macroscopic monitoring of the overall state of the composite material.

[0077] In practice, if temperature is not an important monitoring factor, fiber optic temperature sensors may not be implanted in areas with potential defects in key monitoring areas.

[0078] High density involves 10-15 fiber Bragg grating sensors per square meter, while low density involves 3-5 fiber Bragg grating sensors per square meter. Implantation methods include interlayer implantation, intra-gap implantation, and three-fiber braided implantation. Implantation processes include prepreg laying, resin transfer molding, and fiber winding.

[0079] The stress concentration area is divided into uniaxial stress area and multiaxial stress area. For the uniaxial stress area, multiple fiber optic stress sensors are arranged along the stress direction. For the multiaxial stress area, each fiber optic stress sensor is arranged along the corresponding stress direction, so that one fiber optic stress sensor is arranged along each stress direction, thus forming a cross-monitoring network.

[0080] For example, in a composite material plate subjected to uniaxial tension, multiple fiber Bragg grating stress sensors are arranged sequentially along the tensile direction; for example, in the joint area, fiber Bragg grating stress sensors are arranged in multiple directions to form a cross-monitoring network.

[0081] In practice, the choice of implantation method depends on the composite material manufacturing process:

[0082] Interlayer implantation: Applicable to laminated composite materials. During the prepreg laying process, fiber Bragg grating sensors are placed between designated layers, ensuring that the fiber Bragg grating sensor axis is aligned with the fiber direction or arranged according to monitoring requirements, followed by curing.

[0083] Intrasuture implantation: Suitable for composite material structures requiring suture reinforcement. A fiber optic grating sensor is guided into the suture path using a guide needle and implanted along with the suture.

[0084] Three-fiber braiding implantation: Applicable to the braiding of composite materials. In the three-fiber braiding process, a fiber Bragg grating sensor is woven as an independent "yarn" into a designated location in the preform.

[0085] After implantation, the fiber optic grating sensor can be confirmed to be undamaged, accurately positioned, and well integrated with the fiber by microscopic or X-ray inspection.

[0086] S23. Record the positions of all implanted fiber Bragg grating sensors and connect all implanted fiber Bragg grating sensors into an array to obtain a fiber Bragg grating sensor array. Fiber Bragg grating sensor arrays can take the form of linear arrays, planar arrays, and three-dimensional arrays.

[0087] Furthermore, after obtaining the fiber Bragg grating sensor array, it is then packaged and protected as necessary to ensure its normal operation; and performance testing and verification are carried out on the implanted fiber Bragg grating sensor array to ensure that it meets the monitoring requirements.

[0088] Fiber Bragg grating sensors are connected in series or parallel using fiber optic fusion splicing to form an array. After the entire array is laid out, a protective coating is applied to the outside of the array, and it is then encapsulated to protect it from external factors in the operating environment. After encapsulation, the entire sensor array undergoes initial performance testing, including insertion loss testing (which should be below 3dB) and reflectance spectroscopy checks, to ensure that the signal from each fiber Bragg grating sensor is clear and usable.

[0089] S3. The center wavelength offset measured by each fiber optic grating sensor in the fiber optic grating sensor array is obtained in real time through the signal demodulation device.

[0090] In practice, the signal demodulation device uses wavelength demodulation technology, such as tunable laser method, CCD spectrometer method, etc.

[0091] In practice, the signal demodulation device uses a Micron Optics SM125 or a commercially available demodulator with equivalent performance. The demodulator scans the laser wavelength, detects the reflection spectrum of each fiber Bragg grating sensor, and uses a peak detection algorithm to calculate the center wavelength offset Δλ in real time. B The sampling frequency should be no less than 100Hz to meet the requirements of dynamic monitoring. The demodulated data is transmitted to the data processing module in real time via Ethernet or USB interface.

[0092] S4. Obtain the corresponding stress gradient, temperature change rate, or vibration frequency offset based on the center wavelength offset measured in real time by each fiber Bragg grating sensor.

[0093] Step S4 specifically involves: obtaining the corresponding stress gradient based on the center wavelength offset measured in real time by each fiber Bragg grating stress sensor; obtaining the corresponding temperature change rate based on the center wavelength offset measured in real time by each fiber Bragg grating temperature sensor; and obtaining the corresponding vibration frequency offset based on the center wavelength offset measured in real time by each fiber Bragg grating vibration sensor.

[0094] The stress gradient and temperature change rate are set according to the following formulas:

[0095] dε / dx=1 / (λ B (1-P e ))·d(Δλ B ) / dx

[0096] dT / dt=1 / K T ·d(Δλ A ) / dt

[0097] Where dε / dx is the stress gradient; dT / dt is the rate of temperature change; P e λ is the effective elastic-optical coefficient of the optical fiber; B d(Δλ) is the center wavelength of the fiber grating. B ) / dx is the rate of change of the center wavelength offset along direction x (fiber axial direction) as measured by the fiber optic stress sensor; K T d(Δλ) is the temperature sensitivity coefficient. A ) / dt is the rate of change of the center wavelength offset over time as measured by the fiber Bragg grating temperature sensor; Δλ B The center wavelength offset measured by the fiber optic grating stress sensor; Δλ A This refers to the center wavelength offset measured by the fiber Bragg grating temperature sensor.

[0098] In the specific calculation, the data processing module receives the real-time ΔλB data stream and automatically calls the corresponding formula to perform the calculation according to the sensor type (stress, temperature). The derivative is solved approximately by the first-order finite difference method.

[0099] The specific steps for obtaining the vibration frequency offset are as follows: the center wavelength offset measured in real time by the fiber optic grating vibration sensor is processed to convert the center wavelength offset into a frequency spectrum. One or more main vibration frequencies of the composite material structure are identified from the frequency spectrum. The average value of the main vibration frequencies is subtracted from the reference vibration frequency measured in the defect-free state of the composite material structure to obtain the vibration frequency offset.

[0100] In practice, once a primary vibration frequency is identified, it can be directly subtracted from the reference vibration frequency. If multiple primary vibration frequencies are identified, the average frequency is taken and then subtracted from the reference vibration frequency.

[0101] S5. Construct a defect feature map based on the stress gradient, temperature change rate, and vibration frequency offset at the same time.

[0102] Step S5 specifically involves: dividing the composite material of the three-dimensional structure into a three-dimensional network beforehand, dividing it into N×N three-dimensional cube meshes.

[0103] S51. At the same time, normalize all the stress gradients, temperature change rates and vibration frequency offsets obtained, and obtain each normalized stress gradient, temperature change rate and vibration frequency offset.

[0104] Step S51 specifically involves: normalizing all obtained stress gradients to obtain each normalized stress gradient; normalizing all obtained temperature change rates to obtain each normalized temperature change rate; and normalizing all obtained vibration frequency offsets to obtain each normalized vibration frequency offset. Normalization is performed to a value between 0 and 1.

[0105] S52. The values ​​of each grid are filled in sequentially according to the normalized values ​​of stress gradient, temperature change rate and vibration frequency offset at the location. For locations where no value exists, the value is filled with 0. All the filled values ​​are arranged in a two-dimensional N×N feature map to obtain an N×N stress gradient feature map, temperature change rate feature map and vibration frequency offset feature map respectively. The three feature maps are spliced ​​in the channel dimension to obtain an N×N×3 defect feature map at a single moment.

[0106] Step S52 is as follows: The values ​​of each grid are filled according to the stress gradient normalized to their location, with zeros filled for locations where no values ​​exist. Then, all values ​​are arranged in the form of a two-dimensional N×N feature map to obtain an N×N stress gradient feature map; The values ​​of each grid are filled according to the temperature change rate normalized to their location, with zeros filled for locations where no values ​​exist. Then, all values ​​are arranged in the form of a two-dimensional N×N feature map to obtain an N×N temperature change rate feature map; The values ​​of each grid are filled according to the vibration frequency offset normalized to their location, with zeros filled for locations where no values ​​exist. Then, all values ​​are arranged in the form of a two-dimensional N×N feature map to obtain an N×N vibration frequency offset feature map; Then, the three feature maps are stitched together in the channel dimension to obtain an N×N×3 defect feature map at a single moment.

[0107] In practice, if a location with a value covers multiple grids, then all of these grids will be filled with that value.

[0108] The defect identification model includes a first spatial feature extraction model, a second spatial feature extraction model, a temporal feature extraction model, a multi-source spatiotemporal fusion model, and a classification model. The defect feature map at the current moment is input into the spatial feature extraction model for processing to obtain a spatial feature map. Continuous defect feature maps from the current moment and previous preset time periods are input into the temporal feature extraction model for processing to predict the defect feature map at a future moment. The predicted defect feature map is then input into the second spatial feature extraction model for processing to obtain a spatiotemporal feature map, a spatial feature map, and spatiotemporal features. Figure 1 The input is fed into a multi-source spatiotemporal fusion model for processing to obtain multi-source spatiotemporal fusion features. These features are then fed into a classification model for processing to obtain defect detection results. In specific implementation, the defect detection results are N×N 0 and 1 grid result images.

[0109] In practice, the time feature extraction model is first trained using the continuous defect feature maps of a certain time T1 and the preset time period before each defect data as input, and the defect feature map of time T2 after the preset time interval of time T1 as output. After training, the model is frozen, and then the defect recognition model is trained.

[0110] In this embodiment, the temporal feature extraction model (LSTM) is first trained based on the defect feature map of the previous 10 seconds of each defect data moment to predict the defect feature map of the next moment. After training, the model is frozen. Then, the defect recognition model is trained as a whole to obtain the trained defect recognition model.

[0111] The first and second spatial feature extraction models both use a ResNet50 network (with the last fully connected layer removed); the temporal feature extraction model uses an LSTM network; and the classification model uses a multi-layer fully connected network.

[0112] The multi-source spatiotemporal fusion model is set according to the following formula:

[0113] Output=A⊙F fused

[0114] A = G⊙Softmax(S); F fused =[F s ,F t ]

[0115] G=σ(FC1(W g Std(F s ))+FC2(U g ·Avg(BN(F t )))+b g S = tanh(W) a Flatten(M))+ba

[0116] M=φ(∑ i=1 N ∑ j=1 P (W ij (F s,i ★F t,j )))

[0117] F s,i ★F t,j =F s,i ·F t,j / ((||F s,i ||2·||F t,j ||2+ε) α )

[0118] Where Output represents the multi-source spatiotemporal fusion feature output by the multi-source spatiotemporal fusion model; A represents the final gated feature; ⊙ represents element-wise multiplication; F fused G represents the fusion feature; G represents the statistical gating feature; Softmax() represents the Softmax activation function; S represents the collaborative gating feature; F s Represents spatial feature map; F t Represents a spatiotemporal feature map; [F s ,F t ] represents the spatial feature map F s Spatiotemporal feature map F t Concatenation is performed along the channel dimension; σ() represents the Sigmoid activation function; FC1 and FC2 both represent fully connected layers; W g U g and W a All represent weights; Std() represents standard deviation; Avg() represents mean; BN() represents batch normalization; tanh() represents activation function; Flatten() represents flattened layer; M represents co-current tensor; b g and b a All represent bias; φ() represents the nonlinear transformation function; i and j are both indices; N represents the spatial feature map F. s The total number of channels; P represents the spatiotemporal feature map F. t The total number of channels; F s,i This indicates that by analyzing the spatial feature map F s The feature vector obtained by applying a fully connected layer projection transformation to the i-th channel; F t,j This indicates that by analyzing the spatiotemporal feature map F t The feature vector obtained by applying a fully connected layer projection transformation to the j-th channel; W ij Represents the weight; ||F s,i ||2 represents Fs,i Take the L2 norm; ||F t,j ||2 pairs of F t,j Take the L2 norm; ε is a constant to prevent the denominator from being zero; α represents the learnable parameter; F s,i ·F t,j This represents the dot product of two eigenvectors.

[0119] In specific implementation, the spatial feature map F s The same fully connected layer is applied to each channel in the spatiotemporal feature map F, and the parameters of the same fully connected layer are shared; t Each channel also applies the same new fully connected layer and shares the parameters of the same new fully connected layer.

[0120] When constructing the defect dataset, defective (with known defect locations) and defect-free composite materials are used. Fiber Bragg grating sensors are implanted into both defective and defect-free composite materials, and then the defect dataset is constructed based on the values ​​obtained from the fiber Bragg grating sensors.

[0121] Follow these steps to build a defect dataset for pre-training the defect identification model:

[0122] The composite material with a three-dimensional structure is first divided into a three-dimensional network, which is divided into N×N cubic meshes.

[0123] F1. At the same time, normalize all the stress gradients, temperature change rates and vibration frequency offsets obtained, and obtain each normalized stress gradient, temperature change rate and vibration frequency offset.

[0124] F2. The values ​​of each grid are filled in sequentially according to the normalized values ​​of stress gradient, temperature change rate and vibration frequency offset at the location. For locations where no value exists, zero is filled in, thus obtaining an N×N stress gradient feature map, temperature change rate feature map and vibration frequency offset feature map respectively. The three feature maps are spliced ​​in the channel dimension to obtain an N×N×3 defect feature map at a single moment.

[0125] F3. At the same time, label each grid in N×N according to the condition of the composite material to obtain a label map.

[0126] Step F3 specifically is as follows: According to whether there are problems with the composite materials at the actual positions where each fiber Bragg grating sensor is located, label the positions where the fiber Bragg grating sensors are located. Assign label 1 to the fiber Bragg grating sensors at the positions where there are problems with the composite materials, and assign label 0 to the fiber Bragg grating sensors at the positions where there are no problems with the composite materials; in the N×N grid, the value of each grid is filled according to the label value marked at the position, and for the positions without labels, fill them with 0, thus obtaining an N×N label map.

[0127] Furthermore, the three-dimensional composite material with a three-dimensional structure can be re-divided into a three-dimensional network in advance, divided into M×M cube grids, where M < N. Then, according to whether there are problems with the composite materials at the actual positions where each fiber Bragg grating sensor is located, label the positions where the fiber Bragg grating sensors are located. Assign label 1 to the fiber Bragg grating sensors at the positions where there are problems with the composite materials, and assign label 0 to the fiber Bragg grating sensors at the positions where there are no problems with the composite materials; in the M×M grid, the value of each grid is filled according to the label value marked at the position, and for the positions without labels, fill them with 0, thus obtaining an M×M label map. Using the M×M label map can further achieve the purpose of reducing the model calculation amount during model training or inference.

[0128] F4. At the same moment, take the defect feature map as the input and the label map as the output, thus obtaining sample data at a single time, and the sample data at continuous times is used as defect data.

[0129] F5. Aggregate several defect data to obtain a defect data set.

[0130] The core of the multi-source spatio-temporal fusion model is a dual-path gated attention mechanism, aiming to achieve adaptive and refined fusion of spatial and temporal features.

[0131] The first step: Feature co-perception. The model first calculates the collaborative tensor M through M = φ(∑ i=1 N ∑ j=1 P (W ij (F s,i ★F t,j ))). Here, F s,i ★F t,j calculates the correlation between spatial channel i and spatio-temporal channel j, revealing which spatial patterns are associated with the dynamic evolution of "when" (for example, stress concentration at a specific position is synchronized with vibration mutation at a specific frequency band). The learnable weight W ij further evaluates the importance of different spatio-temporal feature combinations. This step essentially enables the model to automatically discover deep cross-modal coupling relationships.

[0132] Step 2: Generate dynamic attention gating. The model generates two attention weights in parallel:

[0133] 1. Statistical gating feature G: Standard deviation Std(F) based on spatial features s The mean of the spatiotemporal features Avg(BN(F)) t )) Calculation. Focus on regions with drastic spatial changes and overall spatiotemporal anomalies with coarse-grained attention, providing a stable attention baseline.

[0134] 2. Cooperative gating feature S: Obtained by nonlinear transformation of the cooperative tensor M. It encodes complex patterns of cross-modal interactions in a fine-grained manner and can identify weak correlation signals that cannot be detected by single-modal statistics alone.

[0135] The final gating feature A is the element-wise product of G and Softmax(S) (A=G⊙Softmax(S)). This means that the model's ultimate focus is determined by the dynamic modulation of the underlying statistical anomaly signal and the higher-level cross-modal cooperative signal.

[0136] Step 3: Gated fusion. The original stitching features F... fused =[F s ,F t Multiply element-wise with the final gated feature A (Output = A ⊙ F) fused This enables the enhancement of key spatiotemporal features and the suppression of non-key information, outputting refined multi-source spatiotemporal fusion features for use by the classifier.

[0137] Compared to traditional fusion or simple attention mechanisms, multi-source spatiotemporal fusion models have significant advantages:

[0138] 1. Deep Fusion and Refined Attention: Existing technologies often simply splice or weighted average multimodal features, failing to fully explore their internal correlations. The multi-source spatiotemporal fusion model explicitly models the paired interactions between spatial and spatiotemporal feature channels through a collaborative tensor M, and utilizes a dual-gating mechanism (G and S) to achieve multi-granular, adaptive attention from "statistical anomalies" to "correlation patterns," enabling more accurate location and identification of defects characterized by complex spatiotemporal coupling.

[0139] 2. High sensitivity to weak defect signals: The signals of early defects in composite materials (such as microcracks) are often weak and masked by noise. The collaborative computation path (S) of the multi-source spatiotemporal fusion model can amplify signals that change synchronously and correlatedly in terms of spatial and spatiotemporal characteristics. Even if the absolute change of a single mode is not large, it can be effectively captured, thereby significantly improving the detection rate of early defects.

[0140] 3. Stronger generalization and interpretability: through learnable weights W ijThe model can adapt to the association patterns of different structures and damage modes, and has stronger generalization ability. At the same time, analyzing the distribution of G and S can provide some interpretability for diagnosis (for example, it can intuitively see which region's spatial mutation and which spatiotemporal frequency band's anomaly the model focuses on).

[0141] In this embodiment, the loss function of the defect identification model is cross-entropy loss, the optimizer is Adam, and the training is conducted for 100 rounds.

[0142] The defect identification model proposed in this invention, based on multi-parameter and multi-scale monitoring, has higher identification and localization accuracy compared to existing defect identification models.

[0143] Furthermore, the defect identification model can also use existing defect identification models to perform defect identification.

[0144] S6. Collect continuous defect features from the current time and the preset time period prior to it. Figure 1 The input is fed into a pre-trained defect recognition model for defect detection, and the real-time defect detection results of the composite material are obtained. Multi-level early warnings are then issued based on the defect detection results.

[0145] Step S6 specifically involves: collecting continuous defect features from the current time and the preset time period prior to it. Figure 1 The input is fed into a pre-trained defect recognition model for defect detection, predicting a real-time N×N label map of the composite material. The proportion of the number of grids with a value of defect 1 in the N×N label map is calculated out of the total number of grids labeled 0 and 1 at the location of the fiber grating sensor (excluding grids without labels at that location). The defect category is obtained based on the preset proportion.

[0146] In this embodiment, the defect category is determined according to the preset division ratio as follows: if the ratio is between 0 and 0.05, it is classified as no defect; if the ratio is between 0.05 and 0.2, it is classified as a minor defect; if the ratio is between 0.2 and 0.5, it is classified as a moderate defect; and if the ratio is between 0.5 and 1, it is classified as a severe defect.

[0147] Furthermore, the real-time N×N label map of the composite material is predicted, and the location of the mesh for each defect 1 is given.

[0148] In this embodiment, the data sampling time interval is 0.5 seconds, and a defect feature map is collected and acquired every 0.5 seconds.

[0149] The multi-level early warning system includes three levels: Level 1 for minor defects, Level 2 for moderate defects, and Level 3 for severe defects. Minor defects include fiber breakage and microcracks in the matrix. Moderate defects include localized debonding and small-scale cracks. Severe defects include large-scale delamination and crack propagation.

[0150] The specific implementation method of multi-level early warning is as follows:

[0151] Level 1 Warning (Minor Defect): In the monitoring system's visualization interface, the corresponding fiber optic grating sensor location is marked in yellow (based on the grid value of defect 1), and the text warning "Minor damage detected, please pay attention" is displayed. The system log records the event, but no audible alarm is triggered. Recommended action: Increase the monitoring frequency in this area; it can be inspected during the next scheduled maintenance.

[0152] Level 2 Warning (Moderate Defect): In the visualization interface, the corresponding fiber Bragg grating sensor location marker turns orange, and the text warning "Moderate damage detected" is displayed. An intermittent buzzer alarm is emitted. Recommended Action: Non-destructive testing should be arranged within 1-2 weeks to confirm the defect and a repair plan should be developed.

[0153] Level 3 Warning (Severe Defect): In the visual interface, the corresponding fiber optic grating sensor location marker turns red, and the text warning "Severe damage detected, structural integrity threatened" is displayed. A continuous buzzer alarm is emitted, and the pre-assigned maintenance supervisor is automatically notified via SMS / email. Recommended Action: Immediately stop using the structure and initiate emergency repair procedures.

[0154] This invention enables the monitoring of defects in composite materials at the macroscopic, mesoscopic, and microscopic scales, respectively.

[0155] Macroscale: Monitor the overall stress distribution, temperature field and vibration modes of composite material structures, and identify large-scale defects (such as large-scale delamination, crack propagation, etc.), that is, identify severe defects.

[0156] Mesoscale: Monitor the stress, temperature and vibration characteristics of local areas to identify medium-sized defects (such as local debonding, small-scale cracks, etc.), i.e., identify medium-sized defects.

[0157] Microscale: By using high-precision sensors and signal processing technology, we monitor changes in micro-stress and temperature to identify minute defects (such as fiber breakage, matrix microcracks, etc.), that is, to identify minor defects.

[0158] like Figure 2 As shown, this embodiment also provides a monitoring system for implementing the above-described composite material defect monitoring method, comprising:

[0159] A fiber Bragg grating sensor array is composed of several fiber Bragg grating sensors embedded in a composite material.

[0160] The signal demodulation device is used to acquire the center wavelength offset measured by each fiber optic grating sensor in the fiber optic grating sensor array in real time.

[0161] The data processing module calculates the corresponding stress gradient, temperature change rate, or vibration frequency shift based on the center wavelength shift measured in real time by each fiber Bragg grating sensor, and constructs a defect feature map.

[0162] The defect detection module stores pre-trained defect recognition models and performs defect detection.

[0163] The early warning module provides multi-level warnings based on defect detection results.

[0164] Furthermore, this system is integrated into an industrial computer or embedded platform, has Ethernet or 4G / 5G communication capabilities, and supports remote monitoring and data uploading to the cloud platform.

[0165] Example 2:

[0166] like Figure 3 and Figure 4 As shown, this embodiment uses composite material wing monitoring in the aerospace field as an example for its setup:

[0167] 1) The arrangement scheme of the designed fiber Bragg grating sensor array is as follows:

[0168] Joint area: Fiber grating stress sensors are arranged in a high-density manner along the stress direction to form a cross monitoring network.

[0169] Wing root and wingtip: Fiber grating stress and vibration sensors are arranged in a high-density manner to monitor fatigue damage and vibration characteristics.

[0170] Straight sections of the wing: Fiber grating stress, temperature, and vibration sensors are uniformly arranged in a low-density manner for global monitoring.

[0171] 2) Sensor type:

[0172] Fiber Bragg grating stress sensor: Monitors stress distribution on the wing to identify crack propagation. Fiber Bragg grating temperature sensor: Monitors temperature distribution on the wing to identify debonding zones. Fiber Bragg grating vibration sensor: Monitors vibration characteristics of the wing to identify stiffness reduction.

[0173] 4) Data monitoring, identification, and early warning

[0174] Multi-parameter monitoring: Real-time monitoring of the stress distribution, temperature field, and vibration characteristics of the wing, identifying crack propagation and debonding areas caused by fatigue loads.

[0175] Multi-scale monitoring: From the macroscopic stress distribution of the entire wing to the microscopic stress changes of local joints, comprehensively assess the health status of the wing.

[0176] Defect identification: Using a trained defect identification model, defects are detected and their severity is assessed.

[0177] Early warning system: Based on the defect detection results, it issues early warning signals in a timely manner to provide a basis for maintenance decisions.

[0178] This embodiment uses a composite material wing as the research object to construct a wing defect dataset. The defect localization accuracy and defect identification accuracy of the wing defect dataset in the defect identification model proposed in this invention reached 91.7% and 94.3%, respectively. This demonstrates that the defect identification model based on multi-parameter, multi-scale monitoring proposed in this invention has advanced performance and significant advantages in composite material defect monitoring. Through multi-parameter analysis, it comprehensively captures defect features, greatly improving identification accuracy and reducing false positives and false negatives. Multi-scale monitoring allows it to adapt to defects of different sizes, making it widely applicable. It can efficiently handle complex composite material structures, providing strong support for product quality control.

[0179] This invention employs special coating materials and processes to treat fiber Bragg grating sensors, improving their interfacial bonding strength and durability with the composite matrix. It solves the problems of easy sensor detachment and unstable signal in existing technologies. Furthermore, the proposed composite material defect identification model based on multi-parameter and multi-scale monitoring can more accurately identify and locate defects in composite material structures.

[0180] The above embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. A method for monitoring defects in composite materials based on fiber Bragg grating sensors, characterized in that, The steps include the following: S1. Select several fiber Bragg grating sensors, and process each fiber Bragg grating sensor with a coating material to obtain the fiber Bragg grating sensor after the corresponding process. S2. Divide the monitoring area according to the structure of the composite material, implant the processed fiber optic grating sensor into the divided monitoring area and record the implantation position of each fiber optic grating sensor. Connect all the implanted fiber optic grating sensors into an array to obtain a fiber optic grating sensor array. S3. The center wavelength offset measured by each fiber grating sensor in the fiber grating sensor array is obtained in real time through the signal demodulation device; S4. Obtain the corresponding stress gradient, temperature change rate or vibration frequency offset based on the center wavelength offset measured in real time by each fiber Bragg grating sensor. S5. Construct a defect feature map based on the stress gradient, temperature change rate, and vibration frequency offset at the same time. Specifically, step S5 involves: dividing the entire composite material into three-dimensional networks, which are then divided into N×N grids. S51. At the same time, after normalizing all the stress gradients, temperature change rates and vibration frequency offsets obtained, we can obtain each normalized stress gradient, temperature change rate and vibration frequency offset. S52. The values ​​of each grid are filled in sequentially according to the normalized values ​​of stress gradient, temperature change rate and vibration frequency offset at the location. For locations where no value exists, the value is filled with 0. All the filled values ​​are arranged in a two-dimensional N×N feature map to obtain an N×N stress gradient feature map, temperature change rate feature map and vibration frequency offset feature map respectively. The three feature maps are spliced ​​together to obtain the defect feature map at a single moment. S6. Input the continuous defect feature maps of the current time and the preset time period before it into the pre-trained defect recognition model for defect detection, obtain the real-time defect detection results of composite materials, and provide multi-level early warning based on the defect detection results. The defect identification model includes a first spatial feature extraction model, a second spatial feature extraction model, a temporal feature extraction model, a multi-source spatiotemporal fusion model, and a classification model. The defect feature map at the current moment is input into the spatial feature extraction model for processing to obtain a spatial feature map. The continuous defect feature maps within the current moment and the preset time period before it are input into the temporal feature extraction model for processing to predict the defect feature map at a future moment. The predicted defect feature map is then input into the second spatial feature extraction model for processing to obtain a spatiotemporal feature map. The spatial feature map and the spatiotemporal feature map are input together into the multi-source spatiotemporal fusion model for processing to obtain multi-source spatiotemporal fusion features. The multi-source spatiotemporal fusion features are then input into the classification model for processing to obtain the defect detection result. The first and second spatial feature extraction models both use ResNet networks; the temporal feature extraction model uses LSTM networks; and the classification model uses multi-layer fully connected networks. The multi-source spatiotemporal fusion model is set according to the following formula: Output=A⊙F fused A=G⊙Softmax(S);F fused =[F s ,F t ] G=σ(FC1(W g Std(F s ))+FC2(U g ·Avg(BN(F t )))+b g );S=tanh(W a Flatten(M))+b a M=φ(∑ i=1 N ∑ j=1 P (W ij (F s,i ★F t,j ))) F s,i ★F t,j =F s,i ·F t,j / ((||F s,i ||2·||F t,j ||2+ε) α ) Where Output represents the multi-source spatiotemporal fusion feature output by the multi-source spatiotemporal fusion model; A represents the final gated feature; ⊙ represents element-wise multiplication; F fused G represents the fusion feature; G represents the statistical gating feature; Softmax() represents the Softmax activation function; S represents the collaborative gating feature; F s Represents spatial feature map; F t Represents a spatiotemporal feature map; [F s ,F t ] represents the spatial feature map F s Spatiotemporal feature map F t Perform splicing; σ() represents the Sigmoid activation function; FC1 and FC2 both represent fully connected layers; W g U g and W a All represent weights; Std() represents standard deviation; Avg() represents mean; BN() represents batch normalization; tanh() represents activation function; Flatten() represents flattened layer; M represents co-current tensor; b g and b a All represent bias; φ() represents the nonlinear transformation function; i and j are both indices; N represents the spatial feature map F. s The total number of channels; P represents the spatiotemporal feature map F. t The total number of channels; F s,i This indicates that by analyzing the spatial feature map F s The feature vector obtained by applying a fully connected layer to the i-th channel; F t,j This indicates that by analyzing the spatiotemporal feature map F t The feature vector obtained by applying a fully connected layer to the j-th channel; W ij Represents the weight; ||F s,i ||2 represents F s,i Take the L2 norm; ||F t,j ||2 pairs of F t,j Take the L2 norm; ε is a constant to prevent the denominator from being zero; α represents the learnable parameter; F s,i ·F t,j This represents the dot product of two eigenvectors.

2. The method for detecting defects in composite materials based on fiber optic grating sensors according to claim 1, characterized in that: In step S1, the fiber Bragg grating sensor includes a fiber Bragg grating stress sensor, a fiber Bragg grating temperature sensor, and a fiber Bragg grating vibration sensor; the coating material includes epoxy resin and polyimide resin; the process adopts a precision coating process, which includes electrostatic spraying and dip coating.

3. The method for monitoring defects in composite materials based on a fiber optic grating sensor according to claim 2, characterized in that, Step S2 specifically involves: S21. Based on the structure of the composite material, determine the stress concentration area and potential defect distribution area of ​​the composite material, and take the stress concentration area and potential defect distribution area as the key monitoring area, and take the rest as the global monitoring area. S22. Several fiber optic stress sensors are implanted in the stress concentration areas of the key monitoring areas using a high-density method; several fiber optic stress sensors, fiber optic temperature sensors, and fiber optic vibration sensors are implanted in the potential defect distribution areas of the key monitoring areas using a high-density method; several fiber optic stress sensors, fiber optic temperature sensors, and fiber optic vibration sensors are implanted in the global monitoring area using a low-density method. S23. Record the position of all implanted fiber Bragg grating sensors and connect all implanted fiber Bragg grating sensors into an array to obtain a fiber Bragg grating sensor array.

4. The method for detecting defects in composite materials based on a fiber Bragg grating sensor according to claim 3, characterized in that: Step S21 further includes: dividing the stress concentration area into a unidirectional stress area and a multidirectional stress area; for the unidirectional stress area, multiple fiber optic stress sensors are arranged along the stress direction; for the multidirectional stress area, a fiber optic stress sensor is arranged along each stress direction. In step S22, the high density is 10-15 fiber Bragg grating sensors per square meter, and the low density is 3-5 fiber Bragg grating sensors per square meter. The implantation methods in step S22 include interlayer implantation, intra-seam implantation, and three-fiber braided implantation; the implantation process includes prepreg laying, resin transfer molding, and fiber winding. The fiber optic grating sensor array in step S23 can be in the form of a linear array, a planar array, or a three-dimensional array.

5. The method for detecting defects in composite materials based on a fiber Bragg grating sensor according to claim 3, characterized in that: The stress gradient and temperature change rate are set according to the following formulas: dε / dx=1 / (λ B (1-P e ))·d(Dl B ) / dx dT / dt=1 / K T ·d(Δλ A ) / dt Where dε / dx is the stress gradient; dT / dt is the rate of temperature change; P e λ is the effective elastic-optical coefficient of the optical fiber; B d(Δλ) is the center wavelength of the fiber grating. B ) / dx is the rate of change of the center wavelength offset along the x-direction, as measured by the fiber optic stress sensor; K T d(Δλ) is the temperature sensitivity coefficient. A ) / dt is the rate of change of the center wavelength offset over time as measured by the fiber Bragg grating temperature sensor; The specific steps for obtaining the vibration frequency offset are as follows: the center wavelength offset measured in real time by the fiber optic grating vibration sensor is processed to convert the center wavelength offset into a frequency spectrum. One or more main vibration frequencies of the composite material structure are identified from the frequency spectrum. The average value of the main vibration frequencies is subtracted from the reference vibration frequency measured in the defect-free state of the composite material structure to obtain the vibration frequency offset.

6. The method for monitoring defects in composite materials based on fiber optic grating sensors according to claim 1, characterized in that, Follow these steps to build a defect dataset for pre-training the defect identification model: The composite material is first divided into three-dimensional meshes, consisting of N×N grids. F1. At the same time, after normalizing all the stress gradients, temperature change rates and vibration frequency offsets obtained, we can obtain each normalized stress gradient, temperature change rate and vibration frequency offset. F2. The values ​​of each grid are filled in sequentially according to the normalized values ​​of stress gradient, temperature change rate and vibration frequency offset at the location. For locations where no value exists, the value is filled with 0. All the filled values ​​are arranged in a two-dimensional N×N feature map to obtain an N×N stress gradient feature map, temperature change rate feature map and vibration frequency offset feature map respectively. The three feature maps are spliced ​​together to obtain the defect feature map at a single time. F3. At the same time, label each grid in N×N according to the condition of the composite material to obtain the label map; F4. At the same time, the defect feature map is used as input and the label map is used as output to obtain the sample data at a single time. The sample data at several consecutive times are used as the defect data. F5. Several defect data points are aggregated to obtain a defect dataset.

7. A monitoring system for implementing the composite material defect monitoring method according to any one of claims 1-6, characterized in that, include: A fiber optic grating sensor array is composed of several fiber optic grating sensors embedded in a composite material and connected together. A signal demodulation device is used to acquire the center wavelength offset measured by each fiber grating sensor in the fiber grating sensor array in real time. The data processing module calculates the corresponding stress gradient, temperature change rate, or vibration frequency shift based on the center wavelength shift measured in real time by each fiber Bragg grating sensor, and constructs a defect feature map. The defect detection module stores pre-trained defect recognition models and performs defect detection. The early warning module provides multi-level warnings based on defect detection results.

Images