Acoustic emission deep learning real-time diagnosis method for node damage of metal structure

Abstract

The invention relates to the technical field of metal structure health monitoring, and discloses an acoustic emission deep learning real-time diagnosis method for metal structure node damage, which comprises the following steps of: continuously sensing evolution of a physical state of a structure by calculating an acoustic transfer function fingerprint under environmental noise and generating a perturbation time sequence of the acoustic transfer function fingerprint relative to a health benchmark; and when the sequence presents a continuous degradation trend, active sound wave challenge is triggered to identify the physical property of the damage, sensor network self-checking is periodically carried out, finally, the perturbation time sequence, the damage property and the sensor state are jointly input into a deep learning model, and diagnosis information is output. According to the invention, a diagnosis focus is changed from a sparse event to continuous state monitoring, and the problem of monitoring blank of a traditional acoustic emission technology in an event intermission period is solved.

Description

Technical Field

[0001] This invention relates to a deep learning-based real-time diagnostic method for acoustic emission of node damage in metal structures, belonging to the field of metal structure health monitoring technology. Background Technology

[0002] Currently, the use of acoustic emission technology to capture and analyze transient elastic waves generated inside materials due to crack initiation or propagation is a widely adopted technique. It can effectively respond to local damage events that have already occurred and released sufficient energy, providing a direct basis for structural safety assessment.

[0003] However, for critical infrastructure that needs to serve for a long time, structural safety not only needs to cope with occasional rapid damage, but also needs to effectively manage the long-term slow deterioration of material properties caused by fatigue corrosion or creep. In this context, acoustic emission technology, which operates based on an event-driven approach, may have a long monitoring information gap between two identifiable high-energy acoustic emission events. During this period, the health of the structure may be continuously deteriorating in an acoustically silent manner, and existing technologies are unable to perceive this risk accumulation process.

[0004] To improve the detection capability of early-stage, weak damage, a direct technical approach is to increase the receiving sensitivity of sensors or increase the deployment density of sensors. However, the energy of environmental and operational noise in real structures is usually much higher than the weak elastic waves released by early microcracks. Simply increasing the system sensitivity will cause the weak damage signal to be more thoroughly submerged in synchronously amplified background noise, leading to an increase in the false alarm rate and affecting the reliability and practicality of the monitoring system. Specifically, existing technologies mainly have the following two technical problems: 1. The phenomenon that weak damage signals are easily submerged by high-energy background noise affects the sensitivity and accuracy of monitoring; 2. Event-based monitoring methods have monitoring gaps during event intervals, making it impossible to continuously track the slow degradation process of material physical properties. Therefore, how to establish a method that can continuously measure the gradual changes in the physical properties of structural node materials without relying on the occurrence of high-energy discrete damage events becomes the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a deep learning-based real-time diagnostic method for acoustic emission of metal structure node damage. Its main purpose is to solve the problem in the prior art that it is impossible to continuously measure the gradual changes in the physical properties of structural materials in an environment full of background noise.

[0006] To achieve the above objectives, this invention provides a real-time acoustic emission deep learning-based diagnostic method for node damage in metal structures. This method establishes and executes the following monitoring and diagnostic procedures: Based on the ambient background noise collected by the reference sensor and the monitoring sensor, the baseline acoustic transfer function fingerprint characterizing the health status of the metal structure node is calculated and stored. The real-time acoustic transfer function fingerprint is periodically acquired, and the perturbation between the real-time acoustic transfer function fingerprint and the reference acoustic transfer function fingerprint is calculated to generate a perturbation time series. If the value of the perturbation time series is continuously increasing within a predetermined number of calculation periods and its growth slope exceeds the degradation rate threshold, an active challenge is triggered. This active challenge controls one of the monitoring sensors to switch to acoustic emission mode to emit a challenge acoustic wave, and the other monitoring sensors receive the response acoustic wave. By analyzing the harmonic nonlinear acoustic characteristics generated in the response acoustic wave, the physical properties of the damage are determined. At least one pair of monitoring sensors are periodically selected, acting as the transmitter and receiver of sound waves to each other. Based on the principle of acoustic reciprocity, the forward and reverse acoustic transfer functions are compared to diagnose the coupling state between the sensors themselves and the metal structure nodes. Finally, the perturbation time series is input into the deep learning model, and based on the physical properties of the damage and the state of the sensor and coupling, diagnostic information about the damage state of the metal structure nodes is output.

[0007] Preferably, the analysis of the harmonic nonlinear acoustic characteristics generated in the response sound wave is achieved by calculating a nonlinear response index R_nl(t), where: R_nl(t) = E_harmonic / E_fundamental, where E_fundamental is the energy of the response sound wave at the reference frequency of the interrogating sound wave, and E_harmonic is the energy of the response sound wave at the newly generated harmonic frequency; and the nonlinear response index R_nl(t) is compared with a nonlinear damage threshold pre-calibrated according to the fracture mechanical properties of the metallic structural material to distinguish the physical nature of the damage.

[0008] Preferably, the method also establishes and executes the following rule: continuously monitors the energy level of the ambient background noise signal, and automatically activates an autonomous heartbeat mode when the energy level is lower than a silence threshold determined by the average energy level over a period of time based on the fingerprint of the reference acoustic transfer function; the autonomous heartbeat mode includes: controlling one of the monitoring sensors to emit an internally generated, energy-stable heartbeat sound wave, and using the heartbeat sound wave as an alternative excitation source to maintain the continuous generation of the perturbation time series.

[0009] Preferably, the method also establishes and executes the following rules: during the calculation of the acoustic transfer function fingerprint, components coherent with the signal acquired by the reference sensor are removed from the signal acquired by the monitoring sensor to obtain the residual signal; the energy and spectral kurtosis of the residual signal are analyzed to generate an index characterizing the coupling quality between the monitoring sensor and the metal structure node; and when outputting diagnostic information, the monitoring data are weighted according to the coupling quality index.

[0010] Preferably, the method also establishes and executes the following rules: when generating the micro-perturbation time series, the power spectrum of the environmental background noise signal collected by the reference sensor is calculated, and a spectral entropy index characterizing the quality of the signal as an excitation source is generated based on the power spectrum; and the weight of the micro-perturbation time series in outputting the final diagnostic information is dynamically adjusted according to the spectral entropy index.

[0011] Preferably, the rules for diagnosing the sensor itself and its coupling state with the metal structure node further include: when the difference between the forward and reverse acoustic transfer functions exceeds a reciprocity error threshold set according to the sensor's factory calibration data, it is determined that the corresponding sensor or its coupling state has failed, and when outputting diagnostic information, the data weight from the corresponding sensor is automatically reduced or logically isolated.

[0012] Preferably, the perturbation is determined by calculating the integral of the difference between the logarithmic magnitude of the real-time acoustic transfer function fingerprint and the logarithmic magnitude of the reference acoustic transfer function fingerprint over the entire frequency band.

[0013] Preferably, the degradation rate threshold is preset based on the design safety level and material fatigue properties of the metal structure node.

[0014] Preferably, the method also establishes and executes the following rules: acquiring in parallel a continuous state sequence composed of perturbation time series and a discrete event sequence composed of monitored transient high-energy acoustic emission events; using a transfer entropy algorithm to calculate the information flow from the discrete event sequence to the continuous state sequence to generate an index characterizing the temporal causal correlation between the two; and based on this index, identifying and suppressing false alarms of external false source storm interference conditions.

[0015] Preferably, the deep learning model is an autoencoder model. This model determines whether there is an abnormal evolution pattern in the time series by identifying whether the reconstruction error of the perturbation time series exceeds a dynamic baseline determined by the statistical distribution of the reconstruction error generated by the model when processing perturbation time series in a healthy state, thereby outputting diagnostic information.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This method establishes a working mode for continuous monitoring of structural physical properties. It uses the ever-present environmental vibration as the detection medium and obtains a stable basis for characterizing the current physical state of the node by calculating the acoustic transfer function generated when the vibration propagates in the structural node. When a gradual physical change occurs inside the material, its sound wave transmission characteristics will also change accordingly. This change is reflected in the continuous drift of the acoustic transfer function. By analyzing the time series of the difference in this function, this method shifts the focus of diagnosis from capturing rare high-energy sudden damage events to observing the continuous evolution of the structural physical state. Thus, whether there is a continuous and slight deterioration process in the structural health state between two sudden events becomes a directly observable fact, avoiding the monitoring information gap caused by the absence of events in traditional event monitoring methods.

[0017] 2. This method constructs a diagnostic process triggered by state monitoring and property identification. After confirming that the acoustic transfer function of the structural node exhibits an irreversible and continuous evolution trend through the above methods, the system uses this trend as a trigger condition to temporarily change the working mode of the monitoring network. It controls one of the monitoring sensors to emit a preset interrogation sound wave, and the other sensors receive the response sound wave passing through the node. Due to the physical characteristics of nonlinear damage such as microcracks, the passing sound wave will be modulated and new frequency components will be generated. Linear physical changes such as bolt loosening do not have this characteristic. By analyzing the difference in the spectral composition between the response sound wave and the interrogation sound wave, the system can qualitatively distinguish the physical causes that lead to the change in the transfer function. This workflow, which combines continuous state trend monitoring with discrete physical property identification, ensures that the final output diagnostic information not only includes the conclusion that the state is changing, but also includes physical evidence for why it is changing, thus improving the completeness and credibility of the diagnostic conclusion.

[0018] 3. This method integrates a self-verification mechanism for verifying diagnostic prerequisites. In the system performing structural damage diagnosis, information from the acoustic transfer function calculation process is used to assess the coupling state. The residual signal in the monitoring signal that cannot be interpreted by the reference signal directly reflects the physical coupling quality between the monitoring sensor and the structure in terms of its energy and spectral distribution characteristics. Poor coupling will cause the energy of the residual signal to increase and the spectrum to become abnormal. By periodically controlling paired sensors to transmit and receive from each other and comparing the acoustic transfer functions in both directions, the principle of acoustic reciprocity can be used to determine whether the sensor's performance has drifted. These two parallel self-verification processes ensure that all data sources used for structural diagnosis are reliable, allowing the final diagnostic conclusion to be based on a stable and reliable measurement foundation that has been internally verified. Attached Figure Description

[0019] Figure 1This is a flowchart of the closed-loop diagnostic process driven by the state monitoring of the present invention; Figure 2 This is a diagram showing the mapping relationship between the excitation source spectral entropy and the diagnostic weights in this invention. Figure 3 This is a state machine diagram of the multi-mode operation logic of the monitoring system of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are only for explaining this invention and are not intended to limit the scope of protection of this invention.

[0021] This invention discloses a deep learning-based real-time diagnostic method for acoustic emission of metal structure nodes. Its overall architecture is built upon a monitoring procedure consisting of continuous state perception, discrete event identification, and system self-verification. This procedure first establishes a benchmark acoustic transfer function fingerprint, characterizing the elastic wave propagation physical properties of the metal structure node in a healthy state, using a constant ambient background noise during a calibration phase. Then, in a continuous monitoring phase, the real-time acquired acoustic transfer function fingerprint is continuously compared with this benchmark to generate a perturbation time series that quantifies the gradual changes in this physical property. Finally, the procedure uses this time series as the core diagnostic basis, supplemented by an active acoustic wave interrogation mechanism triggered by the abnormal evolution trend of the series to identify the physical properties of the damage, and a parallel sensor network self-verification mechanism to ensure the reliability of the measurement premises. The system jointly outputs diagnostic information about the damage status of nodes. In a typical application scenario, such as long-term health monitoring of the weld area of ​​a critical node in a long-span metal truss bridge (e.g., a steel truss or aluminum alloy truss bridge), this area is continuously subjected to complex stresses caused by traffic wind loads and changes in ambient temperature, posing a risk of fatigue microcrack initiation. To achieve continuous measurement of the gradual changes in the physical properties of materials in such high-energy background noise engineering environments, the monitoring system using the method of this invention is configured to perform the following operations: The monitoring system is deployed and its initial state is defined. A monitoring network consisting of multiple piezoelectric acoustic emission sensors is arranged in the critical node area, wherein at least one sensor is defined as a reference sensor, which is installed on the base material with a relatively stable structural response, far from the heat-affected zone of the weld, to collect the original environmental background noise signal S_ref as the excitation source. (t), the remaining sensors, as monitoring sensors, are directly arranged in the weld and surrounding area to collect the response signal S_(mon_i) (t) after propagation through the node path. All sensors are connected to a data acquisition card with multi-channel synchronous data acquisition function through a low-noise cable. The sampling frequency of the data acquisition card is set to no less than 2MHz to cover the main energy frequency band of the acoustic emission signal.

[0022] After the monitoring system is deployed and enters the calibration phase, it is necessary to extract an objective basis that can stably characterize the physical properties of the structure from the random environmental noise. To this end, this method performs the calculation and storage step of the reference acoustic transfer function fingerprint. This step is performed after the structure is confirmed to be in a healthy state. The system continuously acquires signals from the reference sensor and each monitoring sensor for 24 hours without interruption, and divides the acquired long-term signals into several data frames. The length of each frame is, for example, 1024 sampling points. For each data frame, the data processing unit performs Fourier transforms on the reference signal and the monitoring signal, and calculates the cross-power spectral density between them and the auto-power spectral density of the reference signal, ultimately obtaining the fingerprint from the reference point to the... The acoustic transfer function H_i(f) between monitoring points is obtained by averaging all H_i(f) calculated over 24 consecutive hours to eliminate the influence of short-term environmental noise fluctuations. This results in a statistically stable set of baseline acoustic transfer function fingerprints {H_1(f), H_2(f), ...} that describes the broadband acoustic wave conduction characteristics of healthy nodes. This set of acoustic transfer function fingerprints is stored in local memory as the benchmark for all subsequent diagnostic analyses. After calibration and transition to long-term continuous monitoring, it is necessary to continuously quantify the minute changes that may occur in the node's physical characteristics. To achieve this, the system is configured to periodically generate a perturbation time series. The system repeats the operation of acquiring the acoustic transfer function at a fixed time interval, such as every 60 seconds, to obtain the real-time acoustic transfer function fingerprints {H_1_'(f), H_2_'(f), ...} at the current moment. (f), ...}, Subsequently, the processor compares the real-time acoustic transfer function fingerprint with the stored reference acoustic transfer function fingerprint, and generates a single scalar value, namely the perturbation D(t), through a deterministic calculation procedure. Specifically, the calculation procedure involves calculating the difference between the logarithmic magnitude of the real-time acoustic transfer function fingerprint and the logarithmic magnitude of the reference acoustic transfer function fingerprint, and integrating it over the entire effective frequency band, for example, from 50kHz to 400kHz, to obtain a quantitative index that reflects minute differences. For example, in a certain calculation period t, if the logarithmic magnitudes of the reference acoustic transfer function fingerprint at three discrete frequency points f_1, f_2, f_3 are respectively... [10.5, 12.1, 9.8], while the logarithmic magnitude of the real-time acoustic transfer function fingerprint at the corresponding frequency is [10.6, 12.0, 9.9]. Therefore, the perturbation D(t) at that moment is proportional to the sum of the differences between the two sets of values, which in this example is |10.6-10.5|+|12.0-12.1|+|9.9-9.8|=0.3. By continuously performing this operation, the system transforms the complex multidimensional function changes into a single variable perturbation time series D(t). Any irreversible and continuous growth trend in this series directly reflects the continuous and irreversible change in the physical properties of the material inside the node that affect the propagation of sound waves.

[0023] When the perturbation time series D(t) is observed to show a continuous increasing trend, it is necessary to further distinguish whether this change is caused by crack initiation or by other physical changes such as bolt preload relaxation. Therefore, the system establishes and executes an active challenge procedure for damage physics properties triggered by condition monitoring. The triggering condition of this procedure is set as follows: when the value of the perturbation time series D(t) continuously increases monotonically within a predetermined number of calculation cycles, such as 10 consecutive calculation cycles, and the growth slope of this segment of the series calculated by the least squares method exceeds a preset degradation rate threshold, an active challenge is automatically triggered. The degradation rate threshold is set based on a pre-calibrated design safety level of the monitored metal structural node and the fatigue crack propagation rate curve of the metal material used; a higher safety level corresponds to a lower threshold. Degradation rate threshold; Once a challenge is triggered, the system control logic temporarily changes the operating mode of the sensor network, selecting one of the monitoring sensors, such as sensor A, and switching its operating mode from signal reception to acoustic wave emission. A standardized electrical pulse with controlled energy, such as a 5-cycle sine wave packet with a center frequency of 150kHz, is applied to it via a digital-to-analog converter, causing it to emit a beam of challenge acoustic waves. The remaining monitoring sensors in the area, such as sensors B and C, synchronously capture the response acoustic waves propagating through the nodes. Because microcracks open and close their crack interfaces under acoustic stress, this nonlinear physical behavior modulates the passing acoustic waves and generates new high-frequency harmonic components, a characteristic not found in other linear physical changes. Therefore, by performing spectral analysis on the response acoustic waves and calculating a nonlinear response index... This allows for the identification of the physical properties of the damage. The index is defined as: R_nl(t) = E_harmonic / E_fundamental, where E_fundamental is the energy of the response sound wave at the reference frequency of the interrogating sound wave, and E_harmonic is the energy of the response sound wave at the newly generated second or third harmonic frequency. If the calculated R_nl(t) is higher than a nonlinear damage threshold pre-calibrated based on the fracture mechanics properties of the metal structure material, the diagnostic conclusion is updated to indicate the presence of nonlinear damage, thus providing physical evidence for the continuous increase of D(t).

[0024] It should be noted that the effectiveness of all the diagnostic logic described above is based on the premise that the sensor itself and its acoustic coupling with the structure are stable. To eliminate interference from sensor performance drift or coupling degradation on the diagnostic results, this method also establishes a parallel, periodically executed sensor network state self-verification procedure, which utilizes the principle of acoustic reciprocity. Specifically, the system selects at least one pair of monitoring sensors, such as sensor A and sensor B, at relatively long time intervals, such as once a week. Sensor A is switched to transmit mode and emits a standardized probe pulse, which is received by sensor B and the forward transfer function H_ab is calculated. Immediately afterward, sensor B is switched to transmit mode and emits the exact same probe pulse, which is received by sensor A and the reverse transfer function H_ba is calculated. The processor calculates the difference between the two to obtain a reciprocity error E_ab = |H_ab - H_ba|. If the error exceeds a reciprocity error threshold set based on the sensor's factory calibration data and the acoustic characteristics of the coupling agent, it directly indicates that the performance of sensor A or B or its coupling state has undergone an asymmetric change. The system can locate the specific sensor that has failed or experienced performance drift by polling different sensor pairs in the test network, and automatically reduce the data weight from the faulty sensor or logically isolate it when finally outputting diagnostic information.

[0025] Furthermore, to address boundary conditions in different application scenarios, this method integrates several adaptive adjustment mechanisms. For example, when the monitored object enters a low-energy environment at night, it may be impossible to calculate an effective acoustic transfer function. Therefore, the system is configured to continuously monitor the real-time energy level of the signal acquired by the reference sensor. When this energy level remains below a silence threshold determined based on the average energy level during the calibration phase, the system automatically activates an autonomous heartbeat mode. This mode controls one of the monitoring sensors to periodically emit an internally generated and energy-stable heartbeat sound wave, using this internal sound source as an alternative excitation source to maintain the continuous generation of the perturbation time series. Once the environmental noise energy recovers, the heartbeat mode automatically stops and switches back to passive monitoring mode. In addition, the system also evaluates the quality of the environmental noise signal itself, which is the excitation source. In parallel with the system calculating the perturbation time series, the system performs power spectrum analysis on the reference sensor signal and calculates a normalized spectral entropy index characterizing its spectral complexity. An excitation source with a flat and broad spectrum has a high spectral entropy, while an excitation source with a spectrum concentrated in a few narrow peaks has a lower quality. When making diagnostic decisions, the system dynamically adjusts the weight of the currently calculated perturbation in the time series analysis based on this spectral entropy index. Higher weights are assigned to perturbations generated by higher quality excitation sources, and lower weights are assigned to lower quality sources. Finally, a perturbation time series D(t) corrected by the aforementioned self-calibration procedure and adaptive adjustment mechanism is input into a deep learning model. This model is an autoencoder model, specifically a fully connected neural network containing one input layer, two encoding layers, one bottleneck layer, two decoding layers, and one output layer. It is trained using the ReLU activation function and the Adam optimizer, with the goal of minimizing the mean squared error (MSE). During the calibration phase, this model has established a baseline model capable of reconstructing such normal sequences by learning perturbation time series in a healthy state, as well as a statistical distribution of reconstruction error generated by the model when processing healthy sequences. During monitoring... In this phase, the autoencoder model continuously reconstructs the input real-time D(t) sequence and calculates its reconstruction error. When the D(t) sequence exhibits an irreversible, monotonically increasing abnormal evolution pattern due to the accumulation of structural damage, its shape will deviate from the healthy pattern learned by the model, leading to an increase in reconstruction error. When the reconstruction error exceeds the dynamic baseline determined by the statistical distribution of the healthy state, such as the mean plus three times the standard deviation, the model judges that the time series has an abnormal evolution pattern. Combined with the qualitative conclusions on the physical properties of the damage output by the active questioning procedure and the state information on the sensors and coupling output by the self-verification procedure, the model jointly generates and outputs the final diagnostic information on the damage state of the metal structure nodes.

[0026] Example 1: This example aims to illustrate in detail how the method claimed in this invention can systematically and collaboratively overcome the inherent limitations of existing technologies at the application level by combining a specific and challenging engineering application scenario. The scenario involves long-term health monitoring of the butt weld joint of the web plate of a critical steel box girder of a cross-sea bridge. This joint not only bears enormous temperature stress caused by diurnal temperature differences but also needs to withstand random impact loads generated by tens of thousands of heavy trucks passing through during peak hours. Simultaneously, it enters a quasi-quiet condition with sparse traffic and extremely weak environmental excitation during midnight to early morning. The complexity and extreme nature of this condition place stringent demands on the sensitivity and all-weather adaptability of the monitoring method. In the initial stage of this monitoring task, a high-sensitivity acoustic emission monitoring system based on traditional event monitoring methods was deployed. However, during a three-month trial run, this system revealed two insurmountable application obstacles: firstly, during peak daytime traffic hours, the... The structural vibrations caused by vehicles passing by and various non-structural frictional noises inside the steel box girder have energy far exceeding the weak acoustic emission signals that may be released by the propagation of early fatigue microcracks. To avoid triggering false alarms with massive noise signals, the monitoring threshold of the system must be set at a high level, but this directly leads to the system losing its ability to detect truly weak damage signals. Secondly, during the quasi-quiet working conditions from 2:00 AM to 4:00 AM, due to the lack of sufficient external excitation, the entire monitoring system completely fails, unable to effectively monitor crack propagation that may be caused by residual stress or temperature stress changes during this period, thus forming a monitoring information gap of several hours. To address the above challenges, the monitoring deployment was subsequently carried out using the method claimed in this invention. After completing the sensor deployment and connection, the system performed a 48-hour calibration process, using complete traffic and environmental noise samples during this period to calculate and store the reference acoustic transfer function fingerprint set {H_1} representing the weld node in a healthy state. (f), H_2 (f), ...}; After transitioning to long-term monitoring, the system continuously calculates the real-time acoustic transfer function fingerprint at a frequency of once per minute and compares it with the benchmark, thereby generating a continuous perturbation time series D(t); On the 45th day of monitoring, the data shows that the D(t) sequence begins to deviate from its initial baseline level of random fluctuations around 0.05, showing a slow but continuous and irreversible growth trend. By the 90th day, its value has stabilized at 0.82. During this period, the parallel-running traditional event monitoring system did not record any high-energy acoustic emission events exceeding the alarm threshold.

[0027] The clear, irreversible evolutionary trend exhibited by the D(t) sequence indicates that the material physical properties within the node are undergoing continuous changes. When the growth slope of this sequence exceeds the deterioration rate threshold specified according to the bridge's design safety level for 20 consecutive calculation cycles, the system automatically triggers an active interrogation procedure. One sensor in the monitoring network is temporarily switched to acoustic emission mode, emitting a standardized interrogation acoustic wave towards the node region. The nonlinear response exponent R_nl is calculated after spectral analysis of the response acoustic waves received by the other sensors. The value of D(t) was 0.35, exceeding the nonlinear damage threshold of 0.1 determined based on the material properties of the Q345qDNH weathering bridge steel used in the bridge. This result provides direct physical evidence that the increasing trend of the D(t) sequence was caused by a damage source with nonlinear physical characteristics, namely a suspected microcrack. It should be noted that during several nighttime quasi-quiet periods throughout the monitoring cycle, the system automatically activated the autonomous heartbeat mode, using internally generated weak heartbeat sound waves as an alternative excitation source to maintain the continuous generation of the D(t) sequence, ensuring that the monitoring process of this increasing trend was never interrupted. Based on the diagnostic information that included the continuous deterioration trend and the confirmation of the nonlinear physical properties of the damage, the maintenance team conducted an ultrasonic flaw detection re-examination of the node, and finally found a crack with a length of about 3mm at the root of the weld. The surface microcracks; the results of this embodiment show that the core value of the method claimed in this invention lies not merely in the simple combination of various technologies, but in constructing a diagnostic working method driven by condition monitoring; it utilizes perturbation analysis of acoustic transfer function fingerprints to shift the focus of monitoring from searching for sparse signal needles in the ocean to observing the continuous changes of the entire ocean, i.e., the structural medium itself, thereby avoiding the limitations of insufficient signal-to-noise ratio in traditional methods; by combining this continuous state perception capability with a discrete diagnostic tool, i.e., an active interrogation procedure, which is intelligently triggered by this capability for identifying physical properties, a logically self-consistent diagnostic closed loop is formed from the discovery of anomalies to the confirmation of their nature. Ultimately, this shift in working method makes it possible to continuously and quantitatively track the structural health status over a long period between two high-energy damage events.

[0028] Example 2: This example verifies the ability of the monitoring system using the method of this invention to continuously monitor the early fatigue crack propagation process of metal structure nodes under controlled noise conditions through an accelerated fatigue test conducted in a laboratory environment, and compares it with traditional acoustic emission event monitoring methods. The test is conducted on an MTS electro-hydraulic servo fatigue testing machine platform, which can apply controlled cyclic loads. The test object is a metal specimen, such as a Q345B steel plate specimen with dimensions of 300mm × 100mm × 10mm, which is welded in the middle by a V-groove butt weld process, and an initial notch with a depth of 0.5mm is pre-fabricated at the weld toe on one side of the weld by electrical discharge machining as the initiation point of fatigue cracks. During the test, an axial tensile sinusoidal load is applied to the specimen, and the load frequency is set to 10Hz, which is intended to balance the test effect. To avoid the thermal effects that may be generated by high-frequency loading, the maximum nominal stress was set to 180 MPa, and the stress ratio R was 0.1. To simulate background noise interference in the engineering environment, an industrial sound source was placed 1 meter away from the specimen. The sound source played broadband white noise, and the noise level measured by the sound level meter on the specimen surface was maintained at 75 dB. The initial state of the specimen was defined as no load was applied, and its physical characteristics were no macroscopic cracks. The temperature of the entire test environment was maintained at 25°C ± 2°C. The monitoring system using the method of this invention (hereinafter referred to as the sample group of this invention) and a standard acoustic emission event monitoring system (hereinafter referred to as the control group) were simultaneously deployed in the weld joint area of ​​the specimen. The event discrimination threshold of the control group was set to 45 dB according to the background noise level, while the sample group of this invention first completed the establishment of the reference acoustic transfer function fingerprint of the specimen in a silent state without load.

[0029] After the experiment began, the fatigue testing machine was started and cyclic loads were applied, while the noise source was simultaneously activated. The data acquisition systems for the experimental and control groups of this invention started working synchronously. The experimental group continuously calculated and recorded the perturbation D(t) value with a 60-second cycle, while the control group recorded the cumulative number of acoustic emission events with amplitudes exceeding 45 dB. To obtain the true physical dimensions of crack propagation, the loading was paused every 50,000 cycles, and the crack length at the pre-cut tip was measured using a mobile metallurgical microscope equipped with a micrometer eyepiece, with a measurement accuracy of 0.01 mm. This measurement result served as an objective benchmark for evaluating the two monitoring methods. Throughout the experiment, the operating parameters of each system remained constant until the crack propagated to the predetermined length, at which point the experiment terminated. Within the initial 100,000 cycles of fatigue loading, neither monitoring method showed a significant response; at this point, the microscope... The observed crack length was 0.55 mm, showing only a slight expansion compared to the initial incision. The calculated D(t) value of the sample group fluctuated around the baseline level of 0.04, while the cumulative number of events in the control group was 0. When the loading reached 250,000 cycles, the crack length increased to 1.21 mm. The D(t) value of the sample group showed a continuous upward trend, reaching 0.28, while the control group only recorded 3 instantaneous acoustic emission events during this period. As the loading cycle continued, when the loading reached 500,000 cycles, the actual crack length was 3.85 mm. The D(t) value of the sample group had increased to 0.95, and its growth curve showed a positive correlation with the crack length expansion curve. The cumulative number of events in the control group was only 17, and its data points were sparsely distributed on the time axis, failing to form a continuous sequence that could be used for trend judgment. For specific experimental data records, please refer to Table 1.

[0030] Table 1: A comparison of monitoring data of the sample group and the control group at different fatigue stages of the present invention.

[0031]

[0032] Experimental data show that the perturbation D(t) generated by the monitoring system using the method of this invention exhibits a direct positive correlation with the cumulative damage length of microcracks inside the metallic material. Under the same noise interference conditions, compared with the traditional acoustic emission monitoring method that relies on high-energy damage events, the method of this invention provides a quantitative index that can continuously track the gradual changes in the physical properties of materials.

[0033] To further verify the method claimed in this invention, and to illustrate its substantial differences and significant beneficial effects compared to the conventional technical solutions mentioned in the background art, the following comparative examples are provided.

[0034] Comparative Example 1: This comparative example aims to simulate the actual process and results of health monitoring of the butt weld joint of the web plate of the key steel box girder of the same cross-sea bridge under the same engineering background and hardware conditions as Example 1, using the traditional event-driven acoustic emission monitoring technology disclosed in the background section. The monitoring system used in this comparative example has sensor types, quantities, deployment locations, and data acquisition hardware specifications that are strictly consistent with those described in Example 1. The only difference between this comparative example and Example 1 is that the data processing and diagnostic logic of this comparative example does not use the method based on acoustic transfer function fingerprint perturbation analysis claimed in this invention, but instead uses the traditional acoustic emission event threshold monitoring method. Specifically, the data processing unit of the monitoring system performs real-time energy amplitude analysis on the acoustic signals collected by each monitoring sensor. Only when the signal amplitude exceeds a preset alarm threshold is it recorded as a valid acoustic emission event and accumulated.

[0035] In the early stages of system deployment, to verify the feasibility of this traditional technical solution under real working conditions, a 24-hour on-site noise floor test was first conducted. Test results show that during peak daytime traffic hours, the signal amplitude of structural vibration noise caused by heavy vehicles is generally between 40dB and 50dB, with an instantaneous peak value reaching 55dB. To avoid the massive traffic noise continuously triggering false alarms and causing the system to malfunction, following industry practice, a 5dB margin was reserved based on the measured noise peak value, and the alarm threshold for acoustic emission events was ultimately set at 60dB. This setting aims to ensure that only signals with energy significantly higher than general background noise are counted as damage events. After setting the threshold, the traditional monitoring system was put into continuous operation for 90 days. During this period, the maintenance team tracked the physical state of the node in parallel using other non-destructive testing methods (such as periodic ultrasonic testing). The specific correspondence between monitoring results and physical state is shown in Table 2.

[0036] Table 2: Comparison record of traditional acoustic emission monitoring method and actual node status in Comparative Example 1.

[0037]

[0038] Experimental results show that the acoustic emission event monitoring method based on a fixed threshold has inherent technical limitations when dealing with early structural damage monitoring tasks in high-noise backgrounds. Because the alarm threshold must be set at a relatively high level (60dB) sufficient to avoid environmental noise, the system is completely unable to detect the low-energy acoustic emission signals released by fatigue microcracks during the slow propagation stage. As shown in Table 2, during a monitoring period of up to 90 days, although the physical damage (crack length) of the node continued to expand from 0.51mm to 3.15mm, the traditional monitoring system only recorded two discrete, irregular instantaneous events, and its output diagnosis conclusion was always that the node was in normal condition. This result confirms that this monitoring paradigm, which relies on high-energy discrete events, has a huge information gap in the gradual evolution of damage, and cannot achieve continuous tracking and early warning of the structural health status. The fundamental reason for its monitoring failure is that its technical principle itself cannot effectively separate the weak damage signal from the high-energy background noise.

[0039] Example 3: This example combines Figures 1 to 3 A deep learning-based real-time diagnostic method for acoustic emission damage at metal structure nodes is described, as follows: Figure 1 As shown, the overall diagnostic process begins by using the persistent ambient background noise as a passive excitation source. First, it establishes a baseline acoustic transfer function fingerprint, calculating and storing the acoustic transfer function for a healthy state using ambient noise. Then, it enters the core stage of continuous monitoring and comparison. This stage periodically acquires the real-time acoustic transfer function fingerprint and compares it with the baseline to generate a perturbation time series. In this stage, a parallel sensor network self-calibration mechanism, based on the principle of acoustic reciprocity, periodically diagnoses the sensor and coupling state to ensure the reliability of the monitoring premise. An autonomous heartbeat mode to maintain monitoring continuity is set to automatically activate when the ambient noise is below the silence threshold, emitting... Internally generated heartbeat sound waves serve as an alternative excitation source. During continuous monitoring, the system performs trend analysis and decision-making on the perturbation sequence, determining whether it continues to grow and whether the growth slope exceeds a threshold. If the determination is negative, it returns to the continuous monitoring and comparison stage. If the determination is positive, it triggers an active sound wave challenge. This challenge emits challenge sound waves, which analyze harmonics to identify the physical properties of the damage. The obtained physical property corroborating information, along with the perturbation sequence, is input into the deep learning diagnostic module. This module uses an autoencoder model to analyze the abnormal evolution pattern of the perturbation sequence to determine the damage. Finally, the system integrates the state evolution, damage properties, and system self-verification results to output diagnostic information and form a final conclusion.

[0040] like Figure 2As shown, the horizontal axis represents five different types of excitation sources, from left to right: narrowband vibration 50Hz, mixed noise low quality, mixed noise medium quality, broadband wind load high quality, and white noise ideal. The left vertical axis represents the normalized spectral entropy H_s(t), with a value range of 0 to 1.0, and the right vertical axis represents the diagnostic weight W(t), also with a value range of 0 to 1.0. The figure clearly shows that as the excitation source quality changes from low to high, the corresponding spectral entropy H_s(t), represented by a grid-filled bar, and the diagnostic weight W(t), represented by a diagonally filled bar, both exhibit a monotonically increasing relationship. That is, the more complex and higher the quality of the excitation source spectrum, the larger its spectral entropy H_s(t) value, and the higher the diagnostic weight W(t) assigned in the final diagnosis.

[0041] like Figure 3 As shown, after system startup, it first enters the core passive monitoring state. In this state, the system continuously generates a micro-perturbation time series using environmental noise and analyzes its evolution trend. Depending on different triggering conditions, the system can switch between multiple states: when the environmental noise is below the silence threshold, the system switches from passive monitoring to autonomous heartbeat state, emitting internal heartbeat sound waves to replace the excitation source, thereby maintaining the continuity of monitoring data. When the environmental noise recovers to above the silence threshold, the system automatically returns to passive monitoring state. When the periodic self-calibration time is reached, the system enters the system self-calibration state, diagnosing the sensor itself and its coupling state based on the acoustic reciprocity principle, and assessing the reliability of the data source. After self-verification, the system returns to passive monitoring. When the perturbation continuously increases and its slope exceeds the degradation threshold in passive monitoring, the system switches to active interrogation. In this state, the sensor's operating mode is temporarily switched, interrogation sound waves are emitted, and the response sound waves are analyzed to identify the nature of the damage. If the analysis results show that the nonlinear response is lower than the damage threshold, it is considered that the physical evidence is insufficient, and the system returns to passive monitoring. If the nonlinear response is higher than the damage threshold, the system enters damage warning state. In this state, an abnormal evolution pattern is confirmed, diagnostic information containing physical properties is output, and an alarm is triggered to notify the operation and maintenance personnel. The system only returns to passive monitoring after receiving manual confirmation and a reset command.

[0042] Example 4: Before applying the monitoring system using the method of this invention to a critical cast steel node of a large metal structure stadium, a standardized offline calibration and construction procedure needs to be executed to eliminate uncertainties in the setting of key parameters and model training. This procedure aims to ensure that all judgments of the system are based on the physical characteristics of the monitored object itself. Its initial state is defined as obtaining a metal material sample of the same batch, grade, and welding process as the monitored node. The physical characteristics of this sample are confirmed by non-destructive testing to be free of initial defects. A laboratory with the capability of microcrack preparation and high-precision measurement is prepared. The testing platform is designed to withstand fatigue loads of at least 200 kN and is equipped with a mobile microscope with a measurement accuracy of 0.01 mm. To calibrate the nonlinear damage threshold offline, the first step involves creating a 0.5 mm long surface crack on the sample using a fatigue pre-fabrication method, which is then confirmed using a microscope. Subsequently, an active challenge operation is performed on the sample with the known microcrack. This involves emitting a standardized challenge acoustic wave using a sensor, receiving the response acoustic wave using another sensor, and calculating the nonlinear response exponent R_nl(t) through spectral analysis of the response acoustic wave. This operation is repeated 10 times, and the average value is taken to obtain a stable response value of 0.42 corresponding to the 0.5 mm crack length. To achieve a balance between monitoring sensitivity and anti-interference capability, the nonlinear damage threshold of the monitoring system is set to 50% of this measured average value, i.e., 0.21.

[0043] A calibration procedure for the degradation rate threshold was implemented, aiming to establish a quantitative correlation between the physical damage propagation rate and the perturbation D(t) growth rate. To this end, a healthy, defect-free metallic material sample from the same batch was selected, mounted on a fatigue testing machine, and a monitoring system was deployed simultaneously. During the application of a constant amplitude accelerated fatigue load to the sample, the system continuously recorded the value of D(t) at 60-second intervals. The loading was paused after every 10,000 cycles, and the actual crack propagation length was measured using a microscope. By processing the data from the entire test process, a curve showing the relationship between the physical crack propagation rate (crack length increment da / dN per unit cycle) and the perturbation growth rate dD / dN was obtained. According to the design safety specifications for this stadium structure, its maximum permissible crack propagation rate is... mm / cycle, by searching on the corresponding curve, the micro-disturbance growth rate corresponding to this physical rate is obtained as follows: / minute, therefore, the degradation rate threshold used to trigger proactive questioning is set to this value; finally, the training and construction procedure of the autoencoder deep learning model for final diagnosis is executed; the input data of this procedure comes from the monitoring system that has been installed on the actual nodes of the venue and the baseline acoustic transfer function fingerprint has been established; the system continuously collected 72 hours of environmental noise data under the unloaded state after the venue was built and before its official operation, as well as under different construction disturbance conditions such as the installation of seats in different areas and the opening and closing of the roof, and generated a large-scale health state micro-perturbation time series D covering various normal working conditions. (t) Sample set; This sample set is used as training data for the autoencoder model, and the training objective of the model is set to minimize the reconstruction error of these healthy sequences; After training, the same set of healthy samples is input into the trained model again, and the reconstruction error of each time series sample is calculated to obtain a statistical distribution of reconstruction error representing the health status; Finally, the decision threshold used by the model to judge anomalies is set as the sum of the mean of this statistical distribution and three times the standard deviation; By executing the above process, all key judgment thresholds and data models in the monitoring system are set based on the physical and statistical characteristics of the monitored object.

[0044] Example 5: After a monitoring system using the method of the present invention has been in continuous service for two years on a critical welding node of a large port crane, the system simultaneously detected two phenomena. First, the perturbation time series D(t) corresponding to a specific monitoring sensor channel began to deviate from the stable baseline and showed continuous growth. Second, a large hydraulic pump near the node began to generate high-frequency periodic vibrations due to wear, causing the monitoring system to capture a large number of transient high-energy acoustic emission events. Under this condition, it is necessary to determine the source of the growth of the D(t) sequence, that is, whether it is caused by actual structural damage.

[0045] To address this situation, the system executes an online sensor coupling status evaluation procedure based on residual signal analysis in parallel. While the system periodically calculates the acoustic transfer function fingerprint of each channel, the processor synchronously acquires residual signals from the monitoring signals that cannot be interpreted by the reference sensor signals. A well-coupled sensor produces low-energy random noise as its residual signal, while a sensor with deteriorated physical coupling will experience friction and resonance with its mounting base. These unstructured propagated signals will manifest as increased energy and changes in spectral kurtosis in the residual signal. The system calculates the residual signal energy and spectral kurtosis of the abnormal channel to generate an index characterizing the coupling quality between it and the structural nodes. The calculation results show that this index exceeds a preset threshold, indicating that the sensor and structure... The physical coupling state between the sensors changed; the system also executed a time-series causal correlation analysis procedure, which acquired in parallel a continuous state sequence composed of perturbation time series and a discrete event sequence composed of monitored transient high-energy acoustic emission events; based on the transfer entropy algorithm, the system calculated the information flow from the discrete event sequence to the continuous state sequence to generate an index characterizing the time-series causal correlation between the two; the calculation results showed that the index was close to zero, indicating that there was no causal correlation between the growth of the D(t) sequence and the high-energy events generated by the hydraulic pump; combining the outputs of the above two parallel procedures, the system classified the anomaly as sensor coupling state deterioration accompanied by external environmental interference, and then suppressed the structural damage alarm and issued a maintenance instruction to the operation and maintenance personnel regarding the physical inspection of the specified sensor.

[0046] Example 6: To eliminate the influence of changes in the spectral characteristics of environmental excitation sources on the calculation results of perturbation time series, before deploying the monitoring system using the method of this invention on a tall metal structure tower in an industrial area subjected to intermittent narrowband mechanical vibration and broadband wind load, a standardized pre-calibration procedure needs to be performed to establish a quantitative relationship between the quality of the excitation source and the weight of diagnostic information. This procedure provides an engineering calibration method for the dynamic adjustment logic of diagnostic weights based on the spectral entropy of the reference signal in specific implementations. To quantitatively assess the quality of environmental background noise as an excitation source, this procedure first introduces a normalized spectral entropy index. The calculation steps utilize information theory principles to quantify the complexity of the signal spectrum; specifically, within each monitoring cycle, the system analyzes the environmental background noise signal collected by the reference sensor. The processing involves three steps: First, calculating the power spectral density P(f); second, normalizing the power spectral density within the effective frequency band so that its integral equals 1, yielding the probability density function p(f); third, calculating the entropy of this probability density function using the following formula: In the formula, p(f) is the normalized power spectral density of the reference sensor signal; a broadband noise with a flat spectrum has a uniform p(f) distribution, and the calculated H_s value is close to 1, while a narrowband noise with energy concentrated in a few frequencies has a concentrated p(f) distribution, and the calculated H_s value is close to 1. The value is close to 0.

[0047] After defining the quantitative method for the quality of the excitation source, the next step in the procedure is to establish the quality index H_s(t) and diagnostic weights. The mapping relationship between them is established; this process is completed through an offline learning and fitting step; the system first collects long-term reference signals of the tower under healthy conditions and different excitation conditions, including the case of only broadband wind load and the case of only 50Hz narrowband vibration generated by nearby equipment, and calculates the statistical distribution of H_s(t) corresponding to these two cases respectively. The results show that H_s(t) under broadband wind load is mainly distributed around 0.85, while H_s(t) under narrowband vibration is mainly distributed around 0.85. The values ​​are mainly distributed around 0.20; based on this, the following piecewise linear weighting function is established. When H_s(t) is below 0.3, W(t) is set to 0.1 to reduce the impact of perturbation data generated by this type of excitation source on the final diagnostic results; when When the value is higher than 0.8, W(t) is set to 1.0; when H_s(t) is between 0.3 and 0.8, W(t) is determined by linear interpolation; after the system enters long-term monitoring, for the perturbation D(t) generated in each calculation cycle, the processor will synchronously calculate the H_s(t) of the excitation source for the corresponding time period, and use the weight function to calculate the corresponding weight W(t). Finally, the time series that has been dynamically adjusted by the weight is input into the deep learning model for trend analysis, thereby reducing the fluctuation of diagnostic results caused by changes in the characteristics of the excitation source.

[0048] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A real-time acoustic emission deep learning method for diagnosing node damage in metal structures, characterized in that, Includes the following steps: Step S1: Based on the ambient background noise collected by the reference sensor and the monitoring sensor, calculate and store the reference acoustic transfer function fingerprint that characterizes the health status of the metal structure node. Step S2: Periodically acquire the real-time acoustic transfer function fingerprint and calculate the perturbation between the real-time acoustic transfer function fingerprint and the reference acoustic transfer function fingerprint to generate a perturbation time series. Step S3: Determine whether the value of the perturbation time series continues to increase continuously within a predetermined number of calculation periods and whether its growth slope exceeds the degradation rate threshold. Step S4: If the judgment in step S3 is yes, an active challenge is triggered. The active challenge controls one of the monitoring sensors to switch to the sound wave emission mode to emit a challenge sound wave, and the other monitoring sensors receive the response sound wave. The physical properties of the damage are determined by analyzing the harmonic nonlinear acoustic characteristics generated in the response sound wave. Step S5: Periodically select at least one pair of monitoring sensors, which are sound wave emitters and receivers, and compare the forward and reverse acoustic transfer functions based on the principle of acoustic reciprocity to diagnose the coupling state between the sensor itself and the metal structure node. Step S6: Input the perturbation time series into the deep learning model, and output diagnostic information about the damage state of the metal structure node based on the physical properties of the damage and the state of the sensor and coupling.

2. The method for real-time acoustic emission deep learning diagnosis of node damage in metal structures according to claim 1, characterized in that, The analysis of the harmonic nonlinear acoustic characteristics generated in the response sound wave is achieved by calculating a nonlinear response index R_nl(t), where: R_nl(t) = E_harmonic / E_fundamental, where E_fundamental is the energy of the response sound wave at the reference frequency of the interrogating sound wave, and E_harmonic is the energy of the response sound wave at the newly generated harmonic frequency; and the nonlinear response index R_nl(t) is compared with a nonlinear damage threshold pre-calibrated based on the fracture mechanics properties of the metallic structural material.

3. The method for real-time acoustic emission deep learning diagnosis of node damage in metal structures according to claim 1, characterized in that, The method also establishes and enforces the following rule: continuously monitor the energy level of the ambient background noise signal, and automatically activate the autonomous heartbeat mode when the energy level is lower than a silence threshold determined by the average energy level over a period of time based on the fingerprint of the reference acoustic transfer function. The autonomous heartbeat mode includes: controlling one of the monitoring sensors to emit an internally generated, energy-stable heartbeat sound wave, and using this heartbeat sound wave as an alternative excitation source.

4. The method for real-time acoustic emission deep learning diagnosis of node damage in metal structures according to claim 1, characterized in that, The method also establishes and executes the following rules: during the calculation of the acoustic transfer function fingerprint, components coherent with the signal collected by the reference sensor are removed from the signal collected by the monitoring sensor to obtain the residual signal; the energy and spectral kurtosis of the residual signal are analyzed to generate an index characterizing the coupling quality between the monitoring sensor and the metal structure node; and when outputting diagnostic information, the monitoring data are weighted according to the coupling quality index.

5. The method for real-time acoustic emission deep learning diagnosis of node damage in metal structures according to claim 1, characterized in that, The method also establishes and executes the following rules: when generating the perturbation time series, the power spectrum of the environmental background noise signal collected by the reference sensor is calculated, and a spectral entropy index characterizing the quality of the signal as an excitation source is generated based on the power spectrum; and the weight of the perturbation time series in the output of the final diagnostic information is dynamically adjusted according to the spectral entropy index.

6. The method for real-time acoustic emission deep learning diagnosis of node damage in metal structures according to claim 1, characterized in that, The rules for diagnosing the sensor itself and its coupling state with the metal structure nodes also include: when the difference between the forward and reverse acoustic transfer functions exceeds a reciprocity error threshold set according to the sensor's factory calibration data, the corresponding sensor or its coupling state is determined to be faulty, and when outputting diagnostic information, the data weight from the corresponding sensor is automatically reduced or logically isolated.

7. The method for real-time acoustic emission deep learning diagnosis of node damage in metal structures according to claim 1, characterized in that, The perturbation is determined by integrating the difference between the logarithmic magnitude of the real-time acoustic transfer function fingerprint and the logarithmic magnitude of the reference acoustic transfer function fingerprint over the entire frequency band.

8. The method for real-time acoustic emission deep learning diagnosis of node damage in metal structures according to claim 1, characterized in that, The degradation rate threshold is preset based on the design safety level and material fatigue properties of the metal structure node.

9. The method for real-time acoustic emission deep learning diagnosis of node damage in metal structures according to claim 1, characterized in that, The method also establishes and executes the following rules: acquiring in parallel a continuous state sequence consisting of perturbation time series and a discrete event sequence consisting of monitored transient high-energy acoustic emission events; using a transfer entropy algorithm to calculate the information flow from the discrete event sequence to the continuous state sequence to generate an index characterizing the temporal causal correlation between the two; and based on this index, identifying and suppressing false alarms of external false source storm interference conditions.

10. The method for real-time acoustic emission deep learning diagnosis of node damage in metal structures according to claim 1, characterized in that, The deep learning model is an autoencoder model. This model determines whether there is an abnormal evolution pattern in the time series by identifying whether the reconstruction error of the perturbation time series exceeds a dynamic baseline determined by the statistical distribution of the reconstruction error generated by the model when processing perturbation time series in a healthy state, and thus outputs diagnostic information.

Images