A method for monitoring seismic damage of ancient building mortise and tenon joint

Abstract

The application provides a kind of ancient building mortise and tenon joint seismic damage monitoring method, belong to ancient building wood structure health monitoring and seismic performance evaluation technical field, comprising: according to the preset deployment scheme in ancient building mortise and tenon joint deployment MEMS accelerometer and 60GHz millimeter wave radar;Edge computing unit realizes the data synchronous acquisition and preprocessing of MEMS accelerometer and 60GHz millimeter wave radar by self-defined synchronization protocol, obtains vibration-displacement time series synchronization dataset;Edge computing unit extracts 6 core feature parameters capable of representing mortise and tenon joint damage state from vibration-displacement time series synchronization dataset;Lightweight intelligent diagnosis model analyzes 6-dimensional feature vector, outputs the damage state and confidence of mortise and tenon joint;The position of each mortise and tenon joint, damage diagnosis result and temperature and humidity data and system running state are uploaded to management platform, and the management platform visually displays the received data and carries out hierarchical early warning according to damage diagnosis result.

Description

Technical Field

[0001] This invention relates to the field of health monitoring and seismic performance assessment of ancient wooden structures, and in particular to a method for monitoring seismic damage to mortise and tenon joints in ancient buildings. Background Technology

[0002] The main forms of damage to mortise and tenon joints in ancient buildings under earthquake loading include tenon pull-out, mortise cracking, joint loosening, and component misalignment. Traditional inspection methods rely on manual inspections or post-earthquake mapping, which suffer from problems such as strong lag, high subjectivity, and difficulty in quantification.

[0003] Although the current national standard "Technical Standard for Maintenance and Reinforcement of Ancient Wooden Structures" (GB / T 50165-2020) has made qualitative classifications of the degree of damage to nodes, it lacks quantitative criteria applicable to automated monitoring systems, which makes it impossible to achieve real-time early warning and digital management.

[0004] In recent years, some studies have attempted to introduce sensors for structural health monitoring, including: (1) Single accelerometer solution: The overall stiffness degradation is assessed by the change of modal parameters, but it cannot locate node-level damage and is susceptible to temperature and humidity drift interference; (2) Laser displacement sensor solution: It can measure the slippage of the mortise and tenon joints in a non-contact manner with an accuracy of ±0.05mm, but it requires strict alignment of the line of sight. In the complex roof frame of ancient buildings, it often fails due to component obstruction and insufficient light. (3) Acoustic emission (AE) monitoring scheme: Theoretically, it can capture the signal of microcracks in wood, but due to the porous anisotropy of wood, the signal attenuation is severe, the environmental noise (such as wind noise and tourist activities) is large, and there is a lack of reliable coupling methods. There are no successful long-term engineering application cases yet. (4) Visual recognition solutions (such as DIC): Although non-contact, they rely on stable lighting and clear vision, making them difficult to deploy indoors or in dense rafter areas.

[0005] Furthermore, existing systems largely rely on cloud-based analysis and lack local real-time diagnostic capabilities, failing to meet the protection requirements of "early detection and early intervention" for ancient buildings. The closest current solution involves attaching MEMS accelerometers to the surface of beams and columns, combined with threshold alarms to detect abnormal vibrations. However, this method only reflects macroscopic dynamic characteristics and cannot identify mesoscopic damage modes unique to mortise and tenon joints, such as slippage and pull-out. It also fails to address environmental interference and long-term power supply issues. Summary of the Invention

[0006] To address the problems of existing technologies, this invention provides an integrated technical solution that combines "macroscopic vibration + mesoscopic displacement" dual-scale sensor network deployment, edge data fusion, lightweight intelligent diagnosis, environmentally adaptive low-power control, and visual early warning. This solution overcomes the shortcomings of existing technologies and enables non-destructive, accurate, long-term, and intelligent monitoring of seismic damage to mortise and tenon joints in ancient buildings.

[0007] To achieve the above objectives, the present invention provides a method for monitoring seismic damage to mortise and tenon joints in ancient buildings, comprising: S1. Dual-scale lossless sensor network deployment: MEMS accelerometers and 60GHz millimeter-wave radar are deployed at the mortise and tenon joints of ancient buildings according to the preset deployment plan.

[0008] S2. Edge Data Synchronization Acquisition and Preprocessing: Deploy edge computing units and connect them to MEMS accelerometers and 60GHz millimeter-wave radar. The edge computing units use a custom synchronization protocol to achieve synchronous data acquisition and preprocessing between the MEMS accelerometers and the 60GHz millimeter-wave radar, resulting in a vibration-displacement time-series synchronized dataset.

[0009] S3. Edge Environment Adaptive Compensation and Core Feature Extraction: Temperature and humidity sensors are integrated into the edge computing unit to collect and monitor environmental temperature and humidity data in real time. The edge computing unit uses a pre-calibrated reference frequency... Six core feature parameters characterizing the damage state of mortise and tenon joints were extracted from the vibration-displacement time-series synchronization dataset, along with the temperature compensation coefficient k, to construct a 6-dimensional feature vector. ,in, To correct the frequency offset, For damping ratio, For energy spectrum entropy, For maximum slip, For residual displacement, Let be the area of ​​the hysteresis loop.

[0010] S4. Lightweight Intelligent Damage Diagnosis at the Edge: A lightweight intelligent diagnosis model is constructed and trained based on an LSTM-CNN hybrid network. The trained lightweight intelligent diagnosis model is then deployed to the edge computing unit. The lightweight intelligent diagnosis model receives a 6-dimensional feature vector. Output the damage status and confidence level of the mortise and tenon joint. The damage status includes intact (Level-0), slightly loose (Level-1), obviously pulled out (Level-2), and joint failure (Level-3).

[0011] S5. Visualized Early Warning: The location of each mortise and tenon joint, damage diagnosis results, temperature and humidity data, and system operating status are uploaded to the management platform. The management platform visualizes the received data and provides graded early warnings based on the damage diagnosis results.

[0012] Optionally, step S1 specifically includes: A reversible silicone base is used to attach the MEMS accelerometer to the top surface of the beam or the side surface of the column adjacent to the mortise and tenon joint. The X-axis is along the longitudinal direction of the component, the Y-axis is along the transverse direction, and the Z-axis is along the vertical direction, so as to realize the acquisition of three-dimensional acceleration signals.

[0013] The 60GHz millimeter-wave radar is mounted on the side of the beam or column where the mortise and tenon joint is located using a detachable aluminum alloy bracket. It is then fixed with expansion screws or reversible silicone to ensure that the sensor beam is perpendicularly pointed to the area where the mortise and tenon joint is located.

[0014] Optionally, the edge computing unit uses a Raspberry Pi CM4 paired with a Coral USB Accelerator AI accelerator stick.

[0015] Accordingly, in step S2, the edge computing unit achieves synchronized data acquisition and preprocessing between the MEMS accelerometer and the 60GHz millimeter-wave radar through a custom synchronization protocol, including: The edge computing unit simultaneously sends trigger signals to the MEMS accelerometer and the 60GHz millimeter-wave radar, triggering the MEMS accelerometer and the 60GHz millimeter-wave radar to start data acquisition.

[0016] The collected data is transmitted in real time to the local storage module of the edge computing unit via a USB interface.

[0017] High-frequency noise is removed from the MEMS acceleration signal by using a moving average filter.

[0018] Median filtering was used to remove outliers from the 60GHz millimeter-wave radar displacement signal.

[0019] Optionally, in step S3, according to a pre-calibrated reference frequency... and temperature compensation coefficient k Six core feature parameters characterizing the damage state of mortise and tenon joints were extracted from the vibration-displacement time-series synchronization dataset, including: (1) Calculate the corrected frequency offset rate .

[0020] Calculate the dominant frequency of the power spectral density of MEMS acceleration signals .

[0021] For the main frequency To correct and eliminate temperature drift, the formula is as follows: .

[0022] in, The frequency after temperature compensation. T For real-time temperature,T 0 is the reference temperature.

[0023] Calculate the corrected frequency offset The formula is used to characterize the degree of stiffness degradation caused by damage, as follows: .

[0024] in, As the reference frequency, k This is the temperature compensation coefficient.

[0025] (2) Calculate the damping ratio .

[0026] A fast Fourier transform was performed on the MEMS acceleration signal to obtain the power spectral density curve.

[0027] Find the point on the power spectral density curve where the power value or amplitude is the largest, and denote the frequency value corresponding to this point on the frequency axis as the resonant frequency. f n .

[0028] Find the power value or amplitude at the resonance peak on the power spectral density curve. Two points, each a factor of 1, are denoted by their corresponding frequency values ​​on the frequency axis. and ,in, .

[0029] The damping ratio is calculated using the half-power bandwidth method, as shown in the following formula: .

[0030] (3) Calculate the energy spectral entropy .

[0031] The probability distribution is obtained by normalizing the spectrum of the MEMS acceleration signal after performing a fast Fourier transform. p i .

[0032] Calculate the energy spectral entropy using the Shannon entropy formula : .

[0033] Where N is the number of frequency sampling points.

[0034] (4) Calculate the maximum slip. .

[0035] The formula for calculating the difference of extreme values ​​of the relative displacement data sequence d(t) measured by a 60 GHz millimeter-wave radar is as follows: .

[0036] (5) Calculate the residual displacement .

[0037] The formula for calculating the magnitude of displacement failure after the monitoring period ends is as follows: .

[0038] in, For the initial displacement of the window, This represents the window displacement.

[0039] (6) Calculate the area of ​​the hysteresis loop. .

[0040] A virtual force F(t) = M is constructed by combining the vibration acceleration a(t) with the pre-calibrated equivalent mass M of the mortise and tenon joint. a(t).

[0041] Construct a displacement-force coordinate system, normalize d(t) and F(t) to the same time axis, and plot the hysteresis curve.

[0042] The area enclosed by the curve can be calculated using numerical integration, as shown in the following formula: .

[0043] Optionally, a lightweight intelligent diagnostic model is constructed and trained based on an LSTM-CNN hybrid network, including: A lightweight intelligent diagnostic model was constructed using an LSTM-CNN hybrid network. The model structure is as follows: the input layer has 6 nodes; the CNN feature extraction layer has 1 convolutional layer and 1 pooling layer, with a convolutional kernel size of 3×1, 16 kernels, a stride of 1, and a pooling kernel size of 2×1; the LSTM layer has 32 nodes; the fully connected layer has 16 nodes; and the output layer has 4 nodes.

[0044] Through laboratory low-cycle repeated loading tests on mortise and tenon joints and field measurements, dual-scale monitoring data were collected for mortise and tenon joints of different tree species and joint types under four damage states: intact (Level-0), slightly loose (Level-1), obviously pulled tenon (Level-2), and joint failure (Level-3). A total of 10,000 samples were constructed, with 2,500 samples for each damage state, and the samples were divided into training and testing sets in an 8:2 ratio.

[0045] The transfer learning method was used to optimize the model's adaptability: the model trained on samples of Phoebe zhennan wood was used as the pre-trained model, and fine-tuning was performed on sample data of other tree species and node forms.

[0046] Optionally, the method further includes: (1) Event triggering mechanism.

[0047] Under normal circumstances, the system is in sleep monitoring mode, maintaining only low-frequency monitoring by the MEMS accelerometer. When the detected vibration peak value is greater than the vibration threshold, the system starts the 60GHz millimeter-wave radar for high-frequency acquisition and executes the complete diagnostic process. When the vibration peak value is continuously less than the vibration threshold for a preset duration, the system automatically switches back to sleep monitoring mode.

[0048] (2) Power supply scheme design.

[0049] It uses a combination of a 10000mAh lithium battery and a 5W flexible solar charging panel.

[0050] (3) Data upload strategy.

[0051] It integrates a LoRa / NB-IoT dual-mode communication module, which automatically switches the communication mode according to the network coverage of the area where the ancient building is located: NB-IoT mode is used in areas with good network coverage, and LoRa mode is used in remote areas.

[0052] The system transmits data through scheduled uploads and abnormal trigger uploads: Under normal circumstances, damage diagnosis results and key monitoring data are uploaded once per hour; when the diagnosis result is one of the following: slight loosening Level-1, obvious tenon pull-out Level-2, and node failure Level-3, the upload process is immediately triggered, and the damage status, node location, and confidence level are uploaded to the management platform.

[0053] By adopting the above technical solution, the present invention has at least the following beneficial effects: 1. This invention adopts a completely non-destructive deployment method, which is suitable for the protection of ancient buildings: all sensors are fixed by reversible silicone adhesive or detachable brackets, with no drilling or adhesive residue, and will not damage the wooden components of ancient buildings after removal, strictly following the principle of "minimal intervention" in cultural relic protection.

[0054] 2. This invention offers flexible deployment and overcomes the limitations of complex environments: It employs 60GHz millimeter-wave radar to achieve non-line-of-sight displacement measurement, eliminating the need to align with mortise and tenon joints and allowing it to penetrate thin wood surfaces. This solves the deployment challenges in environments with complex roof trusses, component obstructions, and dim lighting inside ancient buildings, making it suitable for monitoring the needs of various types of wooden ancient buildings.

[0055] 3. This invention has strong anti-interference capabilities and high monitoring reliability: It introduces an environmental adaptive compensation mechanism, using a temperature compensation coefficient... k The calibration and real-time correction effectively eliminate structural frequency drift caused by temperature changes, avoiding false alarms; compared with single sensor monitoring, dual-scale data fusion provides more comprehensive information dimensions.

[0056] 4. This invention employs local intelligent diagnosis with strong real-time performance: It uses a lightweight intelligent diagnostic model deployed on an edge computing unit to achieve local real-time diagnosis of damage status without relying on remote data transmission and manual analysis.

[0057] 5. This invention features low power consumption, long-term operation, and controllable cost: It adopts an event-triggered acquisition mechanism and a lithium battery and solar charging power supply scheme, which can operate permanently under light conditions, realizing unattended long-term monitoring.

[0058] 6. The invention has strong targeting for damage identification: targeting typical damage patterns of mortise and tenon joints, and combining comprehensive analysis of 6-dimensional feature vectors (including correction frequency, residual displacement, hysteresis loop area, etc.), a four-level damage assessment system corresponding to the "Technical Standard for Maintenance and Reinforcement of Ancient Wooden Structures" (GB / T 50165-2020) is constructed, which significantly improves the accuracy of damage judgment. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is a schematic diagram of a method for monitoring seismic damage to mortise and tenon joints in ancient buildings, provided as an embodiment of the present invention.

[0061] Figure 2 This is a schematic diagram of an actual deployment scenario for a monitoring device for mortise and tenon joints in ancient buildings, provided as an embodiment of the present invention.

[0062] Figure descriptions: 1. MEMS accelerometer; 2. 60GHz millimeter-wave radar; 3. Edge computing unit; 4. Solar charging panel; 5. Wooden column; 6. Wooden beam; 7. Mortise and tenon joint; 8. Lithium battery; 9. Temperature and humidity sensor. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] In view of the shortcomings of existing monitoring schemes in the above-mentioned background art, the core technical problem to be solved by the present invention is as follows: 1. Laser displacement sensors rely on optical paths. In the complex roof truss environment inside ancient buildings, installation is limited due to problems such as component obstruction and dim lighting, making it impossible to achieve effective deployment throughout the entire area. Furthermore, they are susceptible to environmental interference and have poor stability.

[0065] 2. Acoustic emission technology suffers from rapid signal attenuation and low signal-to-noise ratio in wood, making it difficult to effectively extract damage characteristic signals and lacking reliability, thus failing to meet the needs of long-term engineering monitoring.

[0066] 3. Existing sensor monitoring solutions often lack local intelligent diagnostic capabilities, relying on remote data transmission and manual analysis, resulting in large data delays and low levels of intelligence, thus failing to achieve real-time damage assessment.

[0067] 4. Existing technologies do not have a dedicated identification mechanism for the typical damage mode of mortise and tenon joints, namely "slippage-pullout-loosening". The damage assessment is not targeted enough and is prone to misjudgment and omission.

[0068] The core reason for the above problems is that existing monitoring technologies are not fully adapted to the complex internal environment of ancient buildings and the unique damage evolution characteristics of mortise and tenon joints, and have failed to achieve a precise match between monitoring technologies and the needs of ancient building protection; at the same time, there are shortcomings in the integrated application of intelligent diagnosis and low-power long-term operation technologies.

[0069] After research, the inventors solved the above-mentioned technical problems using the following technical solution: 1. For the first time, 60GHz millimeter-wave radar was applied to non-line-of-sight relative displacement monitoring of mortise and tenon joints in ancient buildings, using its penetrability to overcome the installation limitations of traditional optical sensors.

[0070] 2. A dual-scale feature fusion framework including "vibration response + relative displacement" was constructed. In particular, a hysteresis loop area calculation method based on virtual force was proposed to replace unreliable micro-monitoring methods such as acoustic emission.

[0071] 3. A lightweight edge AI diagnostic model for typical damage modes of mortise and tenon joints was developed, and combined with an environmentally adaptive temperature compensation algorithm, it achieved accurate local real-time state classification.

[0072] 4. By adopting an event triggering mechanism based on vibration threshold and a solar power supply strategy, long-term low-power unmanned monitoring of ancient buildings was achieved.

[0073] 5. A mapping mechanism between the qualitative level of national standards and the quantitative indicators of sensors has been established, achieving seamless integration of monitoring results with cultural relic restoration standards.

[0074] like Figure 1 As shown, this embodiment of the invention provides a method for monitoring seismic damage to mortise and tenon joints in ancient buildings, including: S1. Deployment of dual-scale lossless sensor networks.

[0075] In this step, a MEMS accelerometer 1 and a 60GHz millimeter-wave radar 2 are deployed at the mortise and tenon joint 7 of the ancient building to construct a dual-scale non-destructive sensing network, thereby enabling the accurate acquisition of multi-dimensional monitoring data of the mortise and tenon joint 7. All devices are deployed in a non-destructive manner.

[0076] S1.1 Macroscopic Scale: MEMS accelerometer 1 deployment.

[0077] An industrial-grade MEMS accelerometer 1 (size 5×5×2mm, IP67 protection, range ±2g~±8g, sampling rate ≥1kHz) was selected.

[0078] like Figure 2 As shown, a reversible silicone base is used to attach the MEMS accelerometer 1 to the top surface of the wooden beam 6 adjacent to the mortise and tenon joint 7. The X-axis is along the longitudinal direction of the component, the Y-axis is along the transverse direction, and the Z-axis is along the vertical direction, so as to realize the acquisition of three-dimensional acceleration signals.

[0079] S1.2 Mesoscopic Scale: 60GHz Millimeter Wave Radar 2.

[0080] The 60GHz millimeter-wave radar 2 (size 45×45mm, accuracy ±0.1mm, penetration depth ≤20mm) is selected as the core device for mesoscopic relative displacement acquisition. It can achieve relative displacement measurement without direct alignment with the mortise and tenon joints. The working frequency band is 60-64GHz, and it is not affected by environmental factors such as light, dust, and water vapor, making it suitable for the complex environment of ancient buildings.

[0081] like Figure 2 As shown, the 60GHz millimeter-wave radar 2 is installed under the wooden beam 6 at the tenon-and-mortise node 7 using a detachable aluminum alloy bracket and is fixed with expansion screws or reversible silicone to ensure that the sensor beam is vertically pointed to the area where the tenon-and-mortise node 7 is located.

[0082] The 60GHz millimeter-wave radar 2 uses radar waves to penetrate the surface of the wood and illuminate the interface between the tenon and mortise, calculating the relative displacement between them using the time-of-flight (ToF) principle. This deployment method can be concealed under the rafters or behind the columns, without affecting the appearance of the ancient building, and does not require line-of-sight alignment, solving the deployment problem in complex roof truss environments.

[0083] S2. Synchronous data acquisition and preprocessing at the edge.

[0084] An edge computing unit 3 is deployed and connected to the MEMS accelerometer 1 and the 60GHz millimeter-wave radar 2. The edge computing unit 3 realizes the synchronous acquisition and preprocessing of data between the MEMS accelerometer 1 and the 60GHz millimeter-wave radar 2 through a custom synchronization protocol, and obtains a vibration-displacement time-series synchronous dataset.

[0085] This step completes the synchronous acquisition and fusion processing of dual-scale data on edge computing unit 3, providing data support for subsequent intelligent diagnosis.

[0086] S2.1 Edge Computing Unit 3 Selection and Deployment.

[0087] The Raspberry Pi CM4 paired with the Coral USB Accelerator AI accelerator stick was selected as the edge computing unit 3. It is not only small in size (Raspberry Pi CM4 size 22×45 mm) and low in power consumption (operating power consumption ≤5W), but also has powerful AI computing capabilities, enabling local deployment and operation of lightweight intelligent models.

[0088] like Figure 2 As shown, the edge computing unit 3 is encapsulated in a waterproof box and hung on the side of the wooden column 5. It is wired to the MEMS accelerometer 1 and the 60GHz millimeter-wave radar 2 via a USB interface to ensure the stability of data transmission.

[0089] S2.2 Data Synchronization Acquisition.

[0090] Edge computing unit 3 achieves synchronized signal acquisition between MEMS accelerometer 1 and 60GHz millimeter-wave radar 2 through a custom synchronization protocol, with a synchronization error ≤1ms. Specifically, edge computing unit 3 simultaneously sends trigger signals to both MEMS accelerometer 1 and 60GHz millimeter-wave radar 2, initiating data acquisition. The acquired data is then transmitted in real-time via a USB interface to the local storage module of edge computing unit 3 (configured with 8GB eMMC flash memory), ensuring no data loss.

[0091] S2.3 Data Preprocessing and Fusion.

[0092] High-frequency noise was removed from the MEMS acceleration signal by using a moving average filter (window size 50).

[0093] Median filtering (window size 30) was used to remove abnormal flying points from the displacement signal of the 60GHz millimeter-wave radar.

[0094] The processed data is stored in a local buffer to form a vibration-displacement timing synchronization dataset.

[0095] S3, edge environment adaptive compensation and core feature extraction.

[0096] This step extracts features from the preprocessed data and uses environmental parameters to correct them, eliminating interference from non-destructive factors (mainly temperature) to form a 6-dimensional feature vector for input into the intelligent diagnostic model.

[0097] S3.1 Environmental parameter acquisition and adaptive compensation.

[0098] like Figure 2 As shown, a temperature and humidity sensor 9 (measurement range: temperature 0-50℃, humidity 20%-90%RH) is integrated on the edge computing unit 3 to collect and monitor the temperature and humidity data of the environment in real time. The temperature and humidity sensor 9 is attached to the side of the wooden column 5, near the mortise and tenon joint 7.

[0099] Reference frequency calibration: During the initial installation phase, vibration signals were collected when the ancient building was in good condition and the ambient temperature was T0. The reference natural frequency of the node was determined by FFT analysis. .

[0100] Temperature compensation coefficient calibration: Through continuous monitoring, a linear regression model of frequency offset versus temperature is established to determine the temperature compensation coefficient for that node. k (k=0.002 Hz / ℃).

[0101] S3.2 Core Feature Extraction and Calculation.

[0102] (1) Calculate the corrected frequency offset rate .

[0103] Calculate the dominant frequency of the power spectral density of MEMS acceleration signals .

[0104] For the main frequency To correct and eliminate temperature drift, the formula is as follows: .

[0105] in, The frequency after temperature compensation. T For real-time temperature, T 0 is the reference temperature.

[0106] Calculate the corrected frequency offset The formula is used to characterize the degree of stiffness degradation caused by damage, as follows: .

[0107] in, As the reference frequency, k This is the temperature compensation coefficient.

[0108] (2) Calculate the damping ratio .

[0109] A fast Fourier transform was performed on the MEMS acceleration signal to obtain the power spectral density curve.

[0110] Find the point on the power spectral density curve where the power value or amplitude is the largest, and denote the frequency value corresponding to this point on the frequency axis as the resonant frequency. f n For the mortise and tenon joint 7 of ancient buildings, the resonance frequency f n It is usually the natural frequency of its first (or primary) vibrational mode.

[0111] Find the power value or amplitude at the resonance peak on the power spectral density curve. Two points, each a factor of 1, are denoted by their corresponding frequency values ​​on the frequency axis. and ,in, .

[0112] The damping ratio is calculated using the half-power bandwidth method, as shown in the following formula: .

[0113] (3) Calculate the energy spectral entropy .

[0114] The probability distribution is obtained by normalizing the spectrum of the MEMS acceleration signal after performing a fast Fourier transform. p i .

[0115] Calculate the energy spectral entropy using the Shannon entropy formula : .

[0116] Where N is the number of frequency sampling points, the signal complexity increases and the entropy value increases significantly when damage occurs.

[0117] (4) Calculate the maximum slip. .

[0118] The difference between the extreme values ​​of the relative displacement data sequence d(t) measured by the 60GHz millimeter-wave radar is calculated using the following formula: .

[0119] (5) Calculate the residual displacement .

[0120] The formula for calculating the magnitude of displacement failure after the monitoring period ends is as follows: .

[0121] in, For the initial displacement of the window, This is the window end displacement, which directly reflects the degree of tenon pull-out.

[0122] (6) Calculate the area of ​​the hysteresis loop. .

[0123] A virtual force F(t) = M is constructed by combining the vibration acceleration a(t) with the pre-calibrated equivalent mass M of the tenon and mortise joint 7. a(t).

[0124] Construct a displacement-force coordinate system, normalize d(t) and F(t) to the same time axis, and plot the hysteresis curve.

[0125] The area enclosed by the curve can be calculated using numerical integration, as shown in the following formula: .

[0126] The six feature parameters obtained from the above calculations are combined to form a six-dimensional feature vector. .

[0127] S4, Intelligent diagnosis of lightweight damage at the edge.

[0128] This step utilizes an edge AI model to perform real-time analysis of the feature vectors and output the damage status.

[0129] S4.1 Intelligent Model Selection and Construction.

[0130] An LSTM-CNN hybrid network was selected as the lightweight intelligent diagnostic model. This model combines the ability of LSTM to capture time-series data with the feature extraction capability of CNN. Furthermore, the model has a simple structure, low computational cost, and is compatible with the operational capabilities of edge computing unit 3. The model structure is as follows: Input layer with 6 nodes (corresponding to 6-dimensional feature vectors); CNN feature extraction layer (1 convolutional layer with 3×1 kernel size, 16 kernels, stride 1; 1 pooling layer with 2×1 kernel size); LSTM layer (32 hidden nodes); fully connected layer (16 nodes); Output layer with 4 nodes (corresponding to 4 damage states).

[0131] It should be noted that the four damage states strictly correspond to the damage severity levels in the "Technical Standard for Maintenance and Reinforcement of Ancient Wooden Structures" (GB / T50165-2020), and have been quantitatively defined: Level-0 (In good condition): Corresponds to national standard level a, with tight and unwavering joints.

[0132] Level-1 (Minor Damage): Corresponds to the initial stage of national standard level b, with slight loosening or minor tenon pull-out at the joint.

[0133] Level-2 (Moderate Damage): Corresponds to the late stage of Level B to the early stage of Level C in the national standard. The joint is obviously pulled out, and the length of the pull-out is less than 1 / 4 of the tenon length.

[0134] Level-3 (Severe Damage): Corresponds to the later stages of Level C to Level D in the national standard, with severe tenon pull-out (extrusion length ≥ 1 / 4 of the tenon length) or failure of the joint, and the frame tilting.

[0135] S4.2 Model Training and Transfer Learning.

[0136] The model training dataset comes from laboratory low-cycle repeated loading tests and field measurement data of mortise and tenon joint 7. It collects dual-scale monitoring data of mortise and tenon joint 7 in four states: "intact Level-0, slightly loose Level-1, obvious tenon pull-out Level-2 and joint failure Level-3" for different tree species such as Phoebe zhennan, pine, and fir and different joint forms such as straight tenon and dovetail tenon. A total of 10,000 samples are constructed (2,500 samples for each damage state), which are divided into training set (8,000 samples) and test set (2,000 samples) in an 8:2 ratio.

[0137] Transfer learning was employed to optimize model adaptability: a model trained on samples of Phoebe zhennan straight tenons was used as a pre-trained model, and fine-tuning was performed on sample data of other tree species and node types to reduce the sample requirements and training time for model training. After training, the model achieved a test set accuracy of ≥95% and a single diagnosis time of ≤0.2s, meeting the requirements for real-time diagnosis.

[0138] S4.3 Damage status output.

[0139] The trained lightweight intelligent diagnostic model is deployed to edge computing unit 3. The lightweight intelligent diagnostic model receives a 6-dimensional feature vector. Output the damage status label and confidence level of the tenon and mortise node 7. The specific mapping logic is shown in Table 1.

[0140] The model simultaneously outputs the confidence level (range 0-1) for each damage state. When the confidence level is ≥0.8, it is determined as a valid diagnosis result, and the corresponding level of early warning process is directly triggered; when the confidence level is <0.8, the system marks it as "suspected damage", triggers the re-collection and secondary diagnosis process, and pushes a manual review suggestion.

[0141] S5, Visualized Early Warning.

[0142] This step uploads the location of each tenon and mortise node 7, the damage diagnosis results, temperature and humidity data, and the system operating status to the management platform. The management platform visualizes the received data and provides graded warnings based on the damage diagnosis results.

[0143] The remote management platform adopts a B / S architecture, supporting access via web and mobile devices (Android / iOS systems). The platform displays the real-time location distribution of each mortise and tenon joint (based on ancient building CAD drawings), current damage level, historical damage evolution trend curves, temperature and humidity data, and system operating status (battery level, communication signal strength). A tiered early warning mechanism is implemented: Level-1 (Yellow Alert): a pop-up notification and SMS notification to administrators; Level-2 (Orange Alert): in addition to pop-up and SMS notifications, preliminary reinforcement suggestions are generated; Level-3 (Red Alert): administrators are immediately notified to conduct on-site verification and take emergency reinforcement measures.

[0144] In addition, to ensure the long-term stable operation of the system, the methods also include: (1) Event triggering mechanism.

[0145] Under normal circumstances, the system is in sleep monitoring mode, maintaining only low-frequency monitoring by MEMS accelerometer 1 (low sampling rate, operating current ≤10μA). When a vibration peak value >0.01g (earthquake, strong wind or impact) is detected, the system starts 60GHz millimeter-wave radar 2 for high-frequency acquisition (sampling rate ≥1kHz) and executes a complete diagnostic process. When the vibration peak value is ≤0.01g for 30s, the system automatically switches back to sleep monitoring mode.

[0146] (2) Power supply scheme design.

[0147] It adopts a combination of 8 10000mAh lithium batteries and 4 5W flexible solar charging panels, and has overcharge and over-discharge protection to ensure long-term unattended operation of the system.

[0148] like Figure 2 As shown, the solar charging panel 4 is fixed to the sunlit area above the wooden beam 6, and the lithium battery 8 and the edge computing unit 3 are integrated in the same waterproof box.

[0149] (3) Data upload strategy.

[0150] It integrates a LoRa / NB-IoT dual-mode communication module. The communication mode automatically switches based on network coverage in the area where the ancient building is located: NB-IoT mode is used in areas with good network coverage, with a transmission rate ≥100 kbps; LoRa mode is used in remote areas, with a transmission distance ≥3km.

[0151] The system transmits data through scheduled uploads and abnormal trigger uploads: Under normal circumstances, damage diagnosis results and key monitoring data are uploaded once per hour; when the diagnosis result is one of the following: slight loosening Level-1, obvious tenon pull-out Level-2, and node failure Level-3, the upload process is immediately triggered, and the damage status, node location, and confidence level are uploaded to the management platform.

[0152] The present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for monitoring seismic damage to mortise and tenon joints in ancient buildings, characterized in that, include: S1. Dual-scale lossless sensor network deployment: MEMS accelerometers and 60GHz millimeter-wave radar are deployed at the mortise and tenon joints of ancient buildings according to the preset deployment plan. S2. Edge Data Synchronization Acquisition and Preprocessing: Deploy edge computing units and connect them to MEMS accelerometers and 60GHz millimeter-wave radar. The edge computing units use a custom synchronization protocol to achieve synchronous data acquisition and preprocessing between the MEMS accelerometers and the 60GHz millimeter-wave radar, resulting in a vibration-displacement time-series synchronized dataset. S3. Edge Environment Adaptive Compensation and Core Feature Extraction: Temperature and humidity sensors are integrated into the edge computing unit to collect and monitor environmental temperature and humidity data in real time. The edge computing unit uses a pre-calibrated reference frequency... Six core feature parameters characterizing the damage state of mortise and tenon joints were extracted from the vibration-displacement time-series synchronization dataset, along with the temperature compensation coefficient k, to construct a 6-dimensional feature vector. ,in, To correct the frequency offset, For damping ratio, For energy spectrum entropy, For maximum slip, For residual displacement, The area of ​​the hysteresis loop; S4. Lightweight Intelligent Damage Diagnosis at the Edge: A lightweight intelligent diagnosis model is constructed and trained based on an LSTM-CNN hybrid network. The trained lightweight intelligent diagnosis model is then deployed to the edge computing unit. The lightweight intelligent diagnosis model receives a 6-dimensional feature vector. Output the damage status and confidence level of the mortise and tenon joint. The damage status includes intact (Level-0), slightly loose (Level-1), obviously pulled out (Level-2), and joint failure (Level-3). S5. Visualized Early Warning: The location of each mortise and tenon joint, damage diagnosis results, temperature and humidity data, and system operating status are uploaded to the management platform. The management platform visualizes the received data and provides graded early warnings based on the damage diagnosis results.

2. The method for monitoring seismic damage to mortise and tenon joints in ancient buildings according to claim 1, characterized in that, Step S1 specifically includes: A reversible silicone base is used to attach the MEMS accelerometer to the top surface of the beam or the side surface of the column adjacent to the mortise and tenon joint. The X-axis is along the longitudinal direction of the component, the Y-axis is along the transverse direction, and the Z-axis is along the vertical direction, so as to realize the acquisition of three-dimensional acceleration signals. The 60GHz millimeter-wave radar is mounted on the side of the beam or column where the mortise and tenon joint is located using a detachable aluminum alloy bracket. It is then fixed with expansion screws or reversible silicone to ensure that the sensor beam is perpendicularly pointed to the area where the mortise and tenon joint is located.

3. The method for monitoring seismic damage to mortise and tenon joints in ancient buildings according to claim 1, characterized in that, The edge computing unit uses a Raspberry Pi CM4 paired with a Coral USB Accelerator AI accelerator stick; Accordingly, in step S2, the edge computing unit achieves synchronized data acquisition and preprocessing between the MEMS accelerometer and the 60GHz millimeter-wave radar through a custom synchronization protocol, including: The edge computing unit simultaneously sends trigger signals to the MEMS accelerometer and the 60GHz millimeter-wave radar, triggering the MEMS accelerometer and the 60GHz millimeter-wave radar to start data acquisition. The collected data is transmitted in real time to the local storage module of the edge computing unit via a USB interface; High-frequency noise was removed from the MEMS acceleration signal by using a moving average filter. Median filtering was used to remove outliers from the 60GHz millimeter-wave radar displacement signal.

4. The method for monitoring seismic damage to mortise and tenon joints in ancient buildings according to claim 1, characterized in that, In step S3, the reference frequency is pre-calibrated. and temperature compensation coefficient k Six core feature parameters characterizing the damage state of mortise and tenon joints were extracted from the vibration-displacement time-series synchronization dataset, including: (1) Calculate the corrected frequency offset rate ; Calculate the dominant frequency of the power spectral density of MEMS acceleration signals ; For the main frequency To correct and eliminate temperature drift, the formula is as follows: ； in, The frequency after temperature compensation. T For real-time temperature, T 0 is the reference temperature; Calculate the corrected frequency offset The formula is used to characterize the degree of stiffness degradation caused by damage, as follows: ； in, As the reference frequency, k This is the temperature compensation coefficient; (2) Calculate the damping ratio ; A fast Fourier transform was performed on the MEMS acceleration signal to obtain the power spectral density curve; Find the point on the power spectral density curve where the power value or amplitude is the largest, and denote the frequency value corresponding to this point on the frequency axis as the resonant frequency. f n ; Find the power value or amplitude at the resonance peak on the power spectral density curve. Two points, each a factor of 1, are denoted by their corresponding frequency values ​​on the frequency axis. and ,in, ; The damping ratio is calculated using the half-power bandwidth method, as shown in the following formula: ； (3) Calculate the energy spectral entropy ; The probability distribution is obtained by normalizing the spectrum of the MEMS acceleration signal after performing a fast Fourier transform. p i ; Calculate the energy spectral entropy using the Shannon entropy formula : ； Where N is the number of frequency sampling points; (4) Calculate the maximum slip. ; The formula for calculating the difference of extreme values ​​of the relative displacement data sequence d(t) measured by a 60 GHz millimeter-wave radar is as follows: ； (5) Calculate the residual displacement ; The formula for calculating the magnitude of displacement failure after the monitoring period ends is as follows: ； in, For the initial displacement of the window, This is the window end displacement; (6) Calculate the area of ​​the hysteresis loop. ; A virtual force F(t) = M is constructed by combining the vibration acceleration a(t) with the pre-calibrated equivalent mass M of the mortise and tenon joint. a(t); Construct a displacement-force coordinate system, normalize d(t) and F(t) to the same time axis, and plot the hysteresis curve; The area enclosed by the curve can be calculated using numerical integration, as shown in the following formula: 。 5. The method for monitoring seismic damage to mortise and tenon joints in ancient buildings according to claim 1, characterized in that, In step S4, a lightweight intelligent diagnostic model is constructed and trained based on an LSTM-CNN hybrid network, including: A lightweight intelligent diagnostic model was constructed using an LSTM-CNN hybrid network. The model structure is as follows: the input layer has 6 nodes; the CNN feature extraction layer has 1 convolutional layer and 1 pooling layer, with 16 convolutional kernels of size 3×1 and stride 1, and 2×1 pooling kernels; the LSTM layer has 32 nodes; the fully connected layer has 16 nodes; and the output layer has 4 nodes. Through laboratory low-cycle repeated loading tests on mortise and tenon joints and field measurements, dual-scale monitoring data were collected on mortise and tenon joints of different tree species and joint types under four damage states: intact (Level-0), slightly loose (Level-1), obviously pulled tenon (Level-2), and joint failure (Level-3). A total of 10,000 samples were constructed, with 2,500 samples for each damage state, and the samples were divided into training and testing sets in an 8:2 ratio. The model's adaptability was optimized using transfer learning: the model trained on samples of Phoebe zhennan wood was used as a pre-trained model, and fine-tuning was performed on sample data of other tree species and node types.

6. The method for monitoring seismic damage to mortise and tenon joints in ancient buildings according to any one of claims 1-5, characterized in that, The method further includes: (1) Event triggering mechanism; Under normal circumstances, the system is in sleep monitoring mode, maintaining only low-frequency monitoring by the MEMS accelerometer. When the detected vibration peak value is greater than the vibration threshold, the system starts the 60GHz millimeter-wave radar for high-frequency acquisition and executes the complete diagnostic process. When the vibration peak value is consistently less than the vibration threshold for a preset duration, the system automatically switches back to sleep monitoring mode. (2) Power supply scheme design; It uses a combination of a 10000mAh lithium battery and a 5W flexible solar charging panel; (3) Data upload strategy; It integrates a LoRa / NB-IoT dual-mode communication module, which automatically switches the communication mode according to the network coverage of the area where the ancient building is located: NB-IoT mode is used in areas with good network coverage, and LoRa mode is used in remote areas; The system transmits data through scheduled uploads and abnormal trigger uploads: Under normal circumstances, damage diagnosis results and key monitoring data are uploaded once per hour; when the diagnosis result is one of the following: slight loosening Level-1, obvious tenon pull-out Level-2, and node failure Level-3, the upload process is immediately triggered, and the damage status, node location, and confidence level are uploaded to the management platform.

Images