Intelligent bolt-based composite connection structure pre-tightening state identification method

Abstract

The application provides a kind of composite material connecting structure pre-tightening state identification method based on intelligent bolt, comprising: 1. preparing intelligent bolt and carrying out composite material glue bolt structure vibration experiment, the bolt response signal of composite material glue bolt structure in working environment is collected by the micro sensing unit integrated in the bolt;2. short-time Fourier transform is used to process and analyze the intelligent bolt response signal, construct two-dimensional time-frequency spectrum as the input of residual network, extract the response signal characteristics to preliminarily identify the stability and nonlinear behavior evolution law of bolt system;3. based on residual network, the high-dimensional features of input time-frequency spectrum are extracted, a pre-tightening state identification model is established, and the bolt pre-tightening state identification result is output.The intelligent bolt monitoring system proposed in the application has the advantages of light weight, high precision and high robustness, and can provide effective technical support for online health monitoring and integrity evaluation of composite material connecting structure.

Description

Technical Field

[0001] This invention relates to the field of composite material health monitoring technology, and more specifically to a method for identifying the preload state of composite material connection structures based on smart bolts. Background Technology

[0002] With the increasing demands for lightweight structures in the aerospace, rail transportation, and automotive industries, carbon fiber reinforced resin matrix composites have gradually become the preferred material for manufacturing critical load-bearing structures due to their high specific strength, high specific modulus, and excellent corrosion resistance. However, structural connection problems are unavoidable during the assembly of composite components. Compared to traditional metal structures, composite material connection areas are more prone to becoming weak points, making connection technology a key technical issue in composite material engineering applications. Bolted connections, due to their high load-bearing capacity, strong disassembly capability, and mature assembly process, are widely used in the assembly of large composite material structures. Bolt preload, as a core control parameter in the connection structure, directly affects the interface contact state, load transfer path, and stress distribution characteristics. Insufficient preload may lead to contact surface slippage and fatigue failure, while excessive preload may cause interlaminar delamination or extrusion failure of the composite material. Therefore, achieving precise control and real-time monitoring of bolt preload is of great significance for ensuring the service safety and reliability of composite material connection structures.

[0003] The torque method is currently the mainstream method for controlling bolt preload. This method indirectly controls the preload by applying a specific torque value to the bolt. However, in actual assembly, due to the uncertainty of the friction coefficient and the influence of elastic interaction, even applying the same torque can lead to a preload deviation of ±30%. Bolted connections mainly face two typical failure modes during service: bolt loosening and hole edge damage. Currently commonly used methods for monitoring bolt tightness include fiber optic gratings, strain gauges, and ultrasonic testing. The fiber optic grating method mainly measures bolt preload based on the relative displacement of the Bragg length. Researchers have studied embedding fiber optic grating sensors into the bolt surface or body to create "smart bolts," or combining fiber optic grating sensors with washers to create "smart washers," etc., to determine the slow change in bolt preload by the change in wavelength threshold, suitable for monitoring the decrease in preload under long-term stable conditions. The ultrasonic method directly attaches an ultrasonic probe to the bolt head surface, utilizing the linear relationship between the ultrasonic wave time-of-flight difference and the preload to accurately measure the preload magnitude. Compared with the torque method, ultrasonic testing has significant advantages such as non-destructive testing, high integration, and high accuracy. Existing monitoring technologies have made significant progress in bolt preload and loosening detection, greatly improving the intelligence level of fasteners. However, the aforementioned methods still face many challenges in practical engineering applications, specifically: (1) Sensors developed for bolts often face bottlenecks such as difficulty in adapting to small-sized connection structures, easy interference with assembly accuracy, and insufficient reliability in harsh environments. Their sensor systems are generally bulky and have complex wiring, which not only significantly increases the added weight of the structure, but also has obvious limitations in terms of system lightweighting and adaptability to complex environments.

[0004] (2) Traditional non-destructive testing methods mostly rely on periodic offline shutdown testing, which is a "post-event" or "retrospective" inspection. It is impossible to conduct real-time and dynamic in-situ monitoring of bolt stress changes, crack initiation and corrosion development during the service of the aircraft and the alternating load on the skin.

[0005] Based on the aforementioned research gaps, it is necessary to propose an intelligent bolt assembly scheme for composite material connection structures to solve the lightweighting problem and realize the identification of the pre-tightening state of the connection structure. Summary of the Invention

[0006] To address the challenge of monitoring bolt preload in composite material connection structures, this invention proposes a preload identification method based on smart bolts. By integrating a micro-sensor unit within a titanium alloy bolt, an integrated design of sensor and fastener is achieved, acquiring dynamic response signals during the connection process while meeting lightweight structural requirements. This smart bolt can capture changes in friction and micro-slippage between the hole and the fastener under different preload states and different fit types (clearance / transition). Furthermore, short-time Fourier transform is used to perform time-frequency analysis on the original waveform, constructing a two-dimensional time-frequency spectrum as input to a neural network. Finally, a preload identification model is established based on a ResNet residual network. This method can provide crucial data support for the integrity assessment of composite material structures while meeting lightweight requirements, and is of great significance for promoting intelligent monitoring of aerospace structures.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A method for identifying the preload state of a composite material connection structure based on smart bolts includes the following steps: Step 1: Fabrication of smart bolts and vibration experiments on composite material screw structures A smart bolt was fabricated by integrating a micro-sensing unit inside the bolt to achieve structural health monitoring. The smart bolt was then assembled with composite materials of different mating forms to obtain a composite rubber-screw structure. Vibration experiments were conducted on the composite rubber-screw structure, and the micro-sensing unit collected the bolt response signal of the composite rubber-screw structure in the working environment. Step 2: Smart Bolt Response Data Processing and Analysis The response signal is processed using short-time Fourier transform (SFT) to obtain the joint distribution characteristics of the response signal in the time and frequency domains, and a time-frequency spectrum is generated. The time-frequency spectrum is then subjected to logarithmic decibel conversion and dynamic range limiting to enhance its contrast. The processed time-frequency spectrum is then uniformly scaled to a preset size and used as input features for a deep learning model. Four time-domain features—amplitude, energy, skewness, and kurtosis—are extracted from the response signal to measure the variation of the bolt signal in the transition fit-through hole specimen. The correlation between these time-domain features and the bolt preload state is established. By analyzing the trend and dispersion of the time-domain feature values, the stability and nonlinear behavior evolution of the composite material screw structure are preliminarily identified. Step 3: Deep Learning-Based Pre-Tightening State Recognition Model A pre-tightening state recognition model is established using a ResNet residual network. The ResNet consists of an initial convolutional layer, a pooling layer, multiple stacked residual modules, and a fully connected layer responsible for classification. The initial convolutional and pooling layers are used to perform preliminary feature extraction and spatial resolution reduction on the input temporal spectrogram, resulting in a feature map of the input temporal spectrogram. Multiple residual modules are stacked, and convolutional operations are used to progressively downsample the feature map size and increase the number of channels, thereby enhancing the expressive power of the bolt pre-tightening state features and extracting high-dimensional features from the input temporal spectrogram. The extracted high-dimensional features are then input into the fully connected layer to establish a nonlinear mapping relationship between the high-dimensional features and the bolt pre-tightening state, thus constructing a deep learning-based pre-tightening state recognition model. The bolt pre-tightening state identification result is obtained based on the output of the recognition model.

[0008] In step one, the smart bolt is assembled with composite materials of different mating forms to obtain a composite material screw structure. Different mating forms (gap / transition) are set to change the boundary conditions between the hole and the fastener and the damage mode around the hole. A vibration experiment based on signal control is carried out on the composite material screw structure, and the signals collected by the micro-sensing unit are recorded.

[0009] In step two, amplitude is the maximum value of the signal amplitude, reflecting the magnitude of the signal strength. As the preload increases, the bolt contact force increases, the signal strength increases, and the amplitude also increases accordingly. Energy represents the total energy contained in the signal, which is usually related to amplitude and duration. When the preload increases, the energy increases, indicating that the working state of the bolt system becomes more intense or complex. Skewness measures the asymmetry of the signal distribution. A decrease in skewness usually reflects that the signal distribution tends to be symmetrical, indicating the stability of the system state. Kurtosis measures the sharpness of the signal, that is, the peak height in the signal distribution. A decrease in kurtosis indicates that the suddenness of the signal is weakened, reflecting that the system gradually tends to stabilize from nonlinear behavior.

[0010] In step three, the residual network successfully overcomes the gradient decay problem of traditional deep neural networks when increasing depth by utilizing the shortcut connection method in the residual module, making it easier to train. The shortcut connection method adds the input and output of the residual module, thereby better capturing the details and contextual information of the features; the residual module is represented as shown in equation (1). (1) In the formula: It is the first The input of each residual module, It is the first The input of each residual module, It is the first The weight parameters of each residual module, It is the residual function, representing the feature information learned by the residual module. It is the ReLU activation function; The deep residual module features are In the formula: It is the first The input of each residual module, It is the first The output of each residual module.

[0011] The composite material is a carbon fiber reinforced resin matrix composite material.

[0012] Compared with the prior art, the present invention has the following advantages: 1. Strong feature representation capability: For nonlinear vibration signals caused by the anisotropy of composite materials, short-time Fourier transform is used to analyze the time-frequency characteristics of the signal and the original data is transformed into a two-dimensional time-frequency graph, which effectively preserves the key features of the signal in the time and frequency dimensions, significantly enhances the identification of acoustic response differences under different preload gradients, and provides high-quality input features for subsequent identification. 2. High recognition accuracy: Based on the deep learning model of residual network (ResNet), the model deeply mines the evolution law of preload hidden in the time spectrum, overcomes the limitations of traditional manual feature extraction in representing complex signals, and the deep learning model shows excellent preload level classification performance, realizing the recognition of preload state.

[0013] 3. Strong robustness: The method of this invention exhibits extremely high recognition accuracy under various typical assembly conditions such as clearance fit and transition fit, proving that the method of this invention has excellent environmental adaptability and engineering application reliability in complex assembly environments, and provides an important reference for online monitoring and quality assessment of composite material structures.

[0014] 4. Lightweight: While taking into account the above performance advantages, by embedding the miniature sensing unit inside the bolt, the additional weight and complex wiring problems brought about by traditional external sensors are avoided, which meets the lightweight design principle of composite material structure and provides a new approach for lightweight health monitoring. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the assembly of the smart bolt and the composite laminate. Figure 3 Arrangement of experimental equipment; Figure 4 (a) and (b) are the original signals and time-frequency diagrams of the transition fit-through hole under different pre-tightening states, respectively; Figure 5 In the middle (a) and (b), the original signal and time-frequency diagram of different fit forms under the same preload (0.4 N·m) are respectively. Figure 6 Box plot of traditional characteristic distribution of bolt signals in transition fit through hole specimens under different preload conditions; Figure 7 Here is a diagram of the ResNet18 network structure; Figure 8 In the diagram, (a), (b), and (c) are the confusion matrices of the preload identification results under the clearance fit, transition fit-through hole, and transition fit-thread conditions, respectively. Detailed Implementation

[0016] To better understand the technical solution of the present invention, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The same reference numerals in the drawings indicate elements with the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0017] The entire testing system includes a power amplifier, a signal generator, a signal acquisition system, a signal display, and a composite material screw structure. Two composite material plates are assembled using smart bolts. During mechanical loading experiments, different levels of preload are applied to the screw structure using a torque wrench. The positive and negative wires at both ends of the smart bolt strain electrodes are connected to the signal acquisition system to collect the response signals of the smart bolt under different preload states and different fit configurations.

[0018] like Figure 1 As shown, a method for identifying the preload state of a composite material connection structure based on smart bolts includes the following steps: Step 1: Fabrication of smart bolts and vibration experiments on composite material screw structures A smart bolt was fabricated by integrating a micro-sensing unit inside the bolt. During fabrication, ethyl α-cyanoacrylate was manually applied to connect the micro-sensing unit to the bolt substrate, ensuring good acoustic coupling performance while avoiding adverse effects on the overall structural strength of the bolt. A titanium bolt was selected, with a 3.1mm diameter hole in the center. A piezoelectric strain electrode, model PZT-53(5H), serving as the micro-sensing unit, was embedded inside the bolt, with its polarization direction along the cylindrical axis. The main material properties of this electrode are as follows: longitudinal electromechanical coupling coefficient... The relative permittivity in the polarization direction is 3400, and the dielectric loss is 2.3%; the piezoelectric strain constant is... , Two T300 prepreg woven composite panels were used for assembly. The laminate holes were arranged on one side, with the center of each hole 15 mm from the long side and 12.5 mm from the short side. The transition fit design included two forms: through holes (φ6.0 mm) and threaded holes (φ6.0 mm bottom hole); the clearance fit used a through hole design (φ6.2 mm). Two samples were prepared for each design, for a total of six samples. Figure 2 shows the composite laminate, the actual parts, and the assembly process.

[0019] In the experiment, a sinusoidal signal was output by the signal generator, with a vibration frequency set to 5Hz and an amplitude of 1V. This signal was then amplified and applied to the composite material screw structure. Different levels of preload were applied to the screw structure using a torque wrench, set to 0.4, 0.5, 0.6, and 0.7 N·m, representing four gradient conditions, to study the effect of preload variations on the bolt signal characteristics. The positive and negative wires of the piezoelectric strain gauge were connected to the signal acquisition system. This system employed an 8-channel acquisition platform based on the EXPRESS-8 acquisition card, manufactured by Physical Acoustics Corporation (PAC), and equipped with adjustable 2 / 4 / 6 type preamplifiers. The smart bolt data sampling rate was set to 2MHz, the acquisition time to 30s, the preamplifier gain to 40dB, and the threshold level to 42dB to minimize environmental noise interference while ensuring effective signal acquisition. The experimental setup is shown in Figure 3.

[0020] Step 2: Smart Bolt Response Data Processing and Analysis Short-time Fourier transform (SFT) was performed on the response signal of the smart bolt. The `signal.stft` function from the SciPy library was used to process the signal, with a sampling frequency of `fs` and a Hanning window selected to reduce spectral leakage. The length of each signal segment was set to 1024 samples (`nperseg=1024`), the overlap between segments was 512 samples (`noverlap=512`), and the number of FFT points was set to 2048 (`nfft=2048`) to improve frequency resolution and ensure good time-frequency balance. The calculated time-frequency spectrum Zxx was logarithmically transformed to obtain decibels. The dynamic range was limited to... The time-spectrum was scaled between 80dB and 0dB to enhance contrast and facilitate subsequent image processing. The final time-spectrum was uniformly scaled to 256×256 for visualization and as input to a deep learning network model. The standardized time-spectrum of the composite screw structure under different preload states and fit types was used as the dataset for the deep learning network model. The original signals and time-spectrum of the transition fit-through hole under different preload states are shown in Figures 4(a) and (b), demonstrating the nonlinear effect of the preload state on the response signal. Under the same preload state (0.4 N·m), the original signals and time-spectrum of different fit types are shown in Figures 5(a) and (b), showing that the signal impact peaks in the response signals of the smart bolt differ under different fit types. The signal duration collected on the transition fit and threaded specimen is longer, indicating a more complex contact friction process. The variation of bolt signals in transition fit-through hole specimens is measured using four time-domain characteristics: amplitude, energy, skewness, and kurtosis. Figure 6 The box plot is shown. Amplitude is the maximum value of the signal amplitude, reflecting the magnitude of the signal strength. As the preload increases, the bolt contact force increases, the signal strength increases, and the amplitude increases accordingly. Energy represents the total energy contained in the signal, usually related to amplitude and duration. Increased preload leads to increased energy, indicating that the working state of the bolt system becomes more intense or complex. Skewness measures the asymmetry of the signal distribution. A decrease in skewness usually reflects a tendency towards symmetry in the signal distribution, indicating the stability of the system state. Kurtosis measures the sharpness of the signal, i.e., the peak height in the signal distribution. A decrease in kurtosis indicates a reduction in the burstiness of the signal, reflecting a gradual shift from nonlinear behavior to stability in the system. From Figure 6As can be seen, the energy exhibits a significant upward trend as the preload increases from 0.4 N·m to 0.7 N·m, with a clear inflection point at 0.6 N·m. Simultaneously, the dispersion of energy within a group gradually increases with the increase of preload. The trends of amplitude, skewness, and kurtosis are basically consistent, all reaching their maximum mean at 0.5 N·m, while the variance of these three characteristics reaches its maximum at 0.7 N·m. Therefore, although traditional time-domain features can reflect the evolution of signals with preload in different structures to some extent, there is a strong nonlinear relationship between their characteristics and the preload state, and traditional time-domain features cannot accurately capture the statistical laws of microstructure evolution.

[0021] Step 3: Deep Learning-Based Pre-Tightening State Recognition Model This study utilizes a ResNet residual network to identify the pre-tightening state of bolts. The model chosen for this case is ResNet-18, a classic lightweight architecture in the ResNet family. Structurally, ResNet-18 consists of one initial convolutional layer, one max-pooling layer, and eight BasicBlock residual modules stacked together, which can be divided into five stages: Stage 1: Composed of a 7×7 convolutional layer (stride 2, output channels 64) and a 3×3 max-pooling layer (stride 2), used for preliminary feature extraction and spatial resolution reduction of the input image. Stages 2 to 5: Composed of stacked BasicBlock residual modules, each stage contains two BasicBlocks. Each stage uses a convolution operation with a stride of 2 to progressively downsample the feature map size, while the number of channels gradually increases from 64 to 128, 256, and 512 to enhance feature representation. The network structure of ResNet-18 is shown in Figure 7.

[0022] To validate the performance of the pre-tightened state recognition model, the dataset was divided into training and test sets in a 4:1 ratio. Considering the needs of sample size and model robustness evaluation, this case study introduced a five-fold cross-validation method. Specifically, the training set was divided into five subsets, with four subsets used for training each time and the remaining subset used for validation, repeated for five rounds, with each round having 50 training cycles. In evaluating the performance of the pre-tightened state recognition model, precision, recall, F1-score, and area under the curve (AUC) were selected as core metrics. Precision, recall, and F1-score were calculated using weighted averages to reduce the impact of class imbalance on the results; AUC was expanded to reflect the model's overall discriminative ability across classes using a macro-average approach. To improve the robustness and reliability of the statistical results, this case study used the Bootstrap resampling method to estimate the 95% confidence intervals for each evaluation metric. The specific method involves 1000 random samplings with replacement on the test set, generating a subsample of the same size as the original test set each time, and recalculating all indicators. Finally, based on the 2.5 and 97.5 percentiles of the resampled indicator distribution, a 95% confidence interval is constructed. The recognition results of the deep learning model under different fit types are shown in Table 1. The deep learning model exhibits high recognition accuracy under various fit types, especially in the "transition fit-through hole" and "transition fit-thread" conditions, where the accuracy rate reaches over 98%.

[0023] Table 1. Preload state identification results for different mating methods index Clearance fit Transition fit - through hole Transition fit - thread accuracy 0.9343 ± 0.0401 0.98.54 ± 0.0182 0.98.54 ± 0.0182 Accuracy 0.9395 ± 0.0355 0.9854 ± 0.0181 0.9862 ± 0.0160 Recall rate 0.9343 ± 0.0401 0.9854 ± 0.0182 0.9854 ± 0.0182 F1 score 0.9341 ± 0.0406 0.9854 ± 0.0183 0.9854 ± 0.0183 Macro AUC 0.9870 ± 0.0124 0.9997 ± 0.0007 1.0000 ± 0.0000 To evaluate the performance of the deep learning model in the task of classifying the preload level of bolt structures, this case study calculated the confusion matrix of the test set for three working conditions, as follows: Figure 8 As shown. Figure 8 Figure (a) shows that under clearance fit conditions, the model achieves 100% accuracy in identifying the 0.7 N·m preload state, and 97.06%, 97.06%, and 80.00% accuracy in identifying the 0.4 N·m, 0.5 N·m, and 0.6 N·m preload states, respectively. Among these, 0.4 N·m and 0.5 N·m are easily misidentified as 0.6 N·m, and 0.6 N·m is easily misidentified as 0.5 N·m. Figure 8 Figure (b) shows that in the case of transition fit-through hole, the deep learning model has a 100% accuracy rate in recognizing the pre-tightening state at the levels of 0.4 N·m and 0.5 N·m, and an accuracy rate of about 97% for both 0.6 N·m and 0.7 N·m. Confusion mainly occurs between these two levels. Figure 8Figure (c) shows that the model performs best in the transition fit-threaded case, achieving 100% accuracy in identifying preload levels of 0.4 N·m, 0.5 N·m, and 0.7 N·m, with only a 5.71% misclassification (misclassified as 0.5 N·m) at the 0.6 N·m level. Overall, this deep learning model demonstrates excellent preload level classification performance under all three fit conditions, particularly excelling in the transition fit condition, proving the invention's strong robustness and ability to maintain high accuracy under various typical assembly conditions. The main misclassifications of the model are concentrated between adjacent preload levels, possibly due to the similarity in acoustic or mechanical response characteristics of bolt structures at adjacent levels.

[0024] This invention innovatively proposes a method for identifying the preload state of composite material connection structures based on smart bolts. It fully utilizes the inherent property of bolts as a medium for inter-substructure transmission, employing time-frequency characterization and deep network learning to construct a nonlinear mapping relationship between high-dimensional features and bolt preload state levels, thus achieving preload state identification. This intelligent monitoring system, while meeting lightweight requirements, can provide crucial data support for the integrity assessment of composite material structures, and is of great significance for promoting intelligent monitoring of aerospace structures.

[0025] Final Note The embodiments described above are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying the preload state of a composite material connection structure based on smart bolts, characterized in that: Includes the following steps: Step 1: Fabrication of smart bolts and vibration experiments on composite material screw structures A smart bolt was fabricated by integrating a micro-sensing unit inside the bolt to achieve structural health monitoring. The smart bolt was then assembled with composite materials of different mating forms to obtain a composite rubber-screw structure. Vibration experiments were conducted on the composite rubber-screw structure, and the micro-sensing unit collected the bolt response signal of the composite rubber-screw structure in the working environment. Step 2: Smart Bolt Response Data Processing and Analysis The response signal is processed using short-time Fourier transform (SFT) to obtain the joint distribution characteristics of the response signal in the time and frequency domains, and a time-frequency spectrum is generated. The time-frequency spectrum is then subjected to logarithmic decibel conversion and dynamic range limiting to enhance its contrast. Finally, the processed time-frequency spectrum is uniformly scaled to a preset size and used as the input feature of the deep learning model. Four time-domain features—amplitude, energy, skewness, and kurtosis—of the response signal are extracted to measure the variation law of the bolt signal in the transition fit-through hole specimen. The correlation between the time-domain features and the bolt preload state is established. By analyzing the trend and dispersion of the time-domain feature values, the stability and nonlinear behavior evolution law of the composite material screw structure are preliminarily identified. Step 3: Deep Learning-Based Pre-Tightening State Recognition Model A pre-tightening state recognition model is established using a residual network. The residual network consists of an initial convolutional layer, a pooling layer, multiple stacked residual modules, and a fully connected layer responsible for classification. The initial convolutional and pooling layers are used to perform preliminary feature extraction and spatial resolution reduction on the input temporal spectrogram, resulting in a feature map of the input temporal spectrogram. Multiple residual modules are stacked, and convolution operations are used to progressively downsample the feature map size and progressively increase the number of channels, thereby enhancing the expressive power of the bolt pre-tightening state features and extracting high-dimensional features from the input temporal spectrogram. The extracted high-dimensional features are then input into the fully connected layer to establish a nonlinear mapping relationship between the high-dimensional features and the bolt pre-tightening state, thus realizing the construction of a deep learning-based pre-tightening state recognition model. The identification result of the bolt pre-tightening state is obtained based on the output of the recognition model.

2. The method for identifying the preload state of composite material connection structures based on smart bolts according to claim 1, characterized in that, In step one, the smart bolt is assembled with composite materials of different mating forms to obtain a composite material screw structure. Different mating forms are set to change the boundary conditions between the hole and the fastener and the damage mode around the hole. A vibration experiment based on signal control is carried out on the composite material screw structure, and the signals collected by the micro-sensing unit are recorded.

3. The method for identifying the preload state of composite material connection structures based on smart bolts according to claim 1, characterized in that: In step two, the amplitude is the maximum value of the signal amplitude, reflecting the magnitude of the signal strength. As the preload increases, the bolt contact force increases, the signal strength increases, and the amplitude also increases accordingly. Energy represents the total energy contained in the signal, which is usually related to the amplitude and duration. When the preload increases, the energy increases, indicating that the working state of the composite material screw structure becomes more intense or complex. Skewness measures the asymmetry of the signal distribution. A decrease in skewness usually reflects that the signal distribution tends to be symmetrical, indicating the stability of the composite material screw structure. Kurtosis measures the sharpness of the signal, that is, the peak height in the signal distribution. A decrease in kurtosis indicates that the suddenness of the signal is weakened, reflecting that the composite material screw structure gradually tends to stabilize from nonlinear behavior.

4. The method for identifying the preload state of composite material connection structures based on smart bolts according to claim 1, characterized in that: In step three, the residual network adds the input and output of the residual module using the shortcut connection method in the residual module, thereby better capturing the details and contextual information of the features; the residual module is represented as shown in equation (1). (1) In the formula: It is the first The input of each residual module, It is the first The input of each residual module, It is the first The weight parameters of each residual module, It is the residual function, representing the feature information learned by the residual module. It is the ReLU activation function; The deep residual module features are In the formula: It is the first The input of each residual module, It is the first The output of each residual module.

5. The method for identifying the preload state of composite material connection structures based on smart bolts according to claim 1, characterized in that: The composite material is a carbon fiber reinforced resin matrix composite material.

Images