Bridge temperature effect compensation method based on digital twinning

Abstract

The invention relates to the technical field of bridge construction, in particular to a bridge temperature effect compensation method based on digital twinning, comprising the following steps: S1, acquiring temperature data and strain data of a bridge monitoring area; s2, generating a corresponding temperature pulse sequence and a strain pulse sequence; s3, outputting an indication signal based on a time sequence analysis result; s4, performing temperature effect compensation on the strain data through a pre-constructed bridge digital twin model; a minute-level temperature and strain slow change signal is converted into a pulse sequence decoupled from physical time through asynchronous event driving coding, the contradiction of sampling frequency and crack propagation time scale mismatch is fundamentally solved, and normal thermal strain and damage abnormal strain are intelligently distinguished through timing sequence analysis of a pulse neural network. And a digital twin model is driven to dynamically switch a compensation strategy, and a micro fatigue crack transient signal is accurately captured and separated under the background of strong thermal shock noise.

Description

Technical Field

[0001] This invention relates to the technical field of bridge construction, and in particular to a bridge temperature effect compensation method based on digital twins. Background Technology

[0002] Paving bridge decks with conductive concrete is an active snow melting and de-icing technology. It involves incorporating conductive materials (such as steel fibers, carbon fibers, and graphite) into the concrete to form a conductive network. When electricity is applied, electrical energy is converted into heat energy, raising the road surface temperature above 0°C to melt snow and ice. However, existing bridges equipped with this system experience severe thermal shock upon power failure due to the disappearance of the Joule heating effect, resulting in a rapid cooling of the concrete structure. Under these conditions, the concentrated thermal stress within the material easily leads to the initiation and propagation of micro-fatigue cracks, posing a potential threat to the long-term structural safety.

[0003] In existing technologies, current structural health monitoring methods based on digital twins face fundamental technical challenges in such scenarios. The initiation and unstable propagation of micro-fatigue cracks are transient events on microsecond or even shorter timescales, while the heat conduction and overall temperature field changes of bridge components are slow processes on the order of minutes to hours. If high-frequency sampling is used to capture transient crack signals, massive amounts of redundant data will be generated, placing a huge burden on transmission, storage, and real-time processing. If low-frequency sampling adapted to the thermal process is used, transient signals that reveal the key to damage will be completely missed, leading to monitoring failure. Summary of the Invention

[0004] To address the technical problems existing in the background art, this invention proposes a bridge temperature effect compensation method based on digital twins, the specific scheme of which is as follows: A bridge temperature effect compensation method based on digital twins includes the following steps: S1. Obtain temperature and strain data for the bridge monitoring area; S2. Based on the temperature data and the strain data, generate corresponding temperature pulse sequences and strain pulse sequences respectively through asynchronous event-driven differential modulation; S3. Input the temperature pulse sequence and the strain pulse sequence into a pre-trained pulse neural network model for time series analysis, and output an indication signal based on the time series analysis results; S4. Based on the indication signal output by the spiking neural network model, temperature effect compensation is performed on the strain data through a pre-constructed bridge digital twin model; wherein, the bridge digital twin model includes the thermo-mechanical coupling constitutive relationship of the bridge monitoring area.

[0005] Furthermore, in S2, based on the temperature data and the strain data, corresponding temperature pulse sequences and strain pulse sequences are generated respectively through asynchronous event-driven differential modulation, as follows: Asynchronous event-driven differential modulation is performed on the temperature data and the strain data respectively to generate corresponding temperature pulse sequences and strain pulse sequences; The asynchronous event-driven differential modulation process is as follows: continuously compare the current value of the monitoring data with a dynamically updated reference value. When the difference between the current value and the dynamically updated reference value is greater than or equal to a preset change threshold, trigger a pulse event, generate a pulse that represents the change direction of this pulse event, and adjust the reference value to the current value. The monitoring data is temperature data or strain data. By continuously triggering the pulse events, a temperature pulse sequence and a strain pulse sequence are formed, which are composed of pulses in the triggering order.

[0006] Furthermore, the preset change threshold includes a temperature threshold set for changes in temperature data and a strain threshold set for changes in strain data. In the pulse sequence, each pulse event is characterized by the polarity of the pulse, and the sequential relationship between pulse events is characterized by the trigger interval.

[0007] Furthermore, in S3, the temperature pulse sequence and the strain pulse sequence are input into a pre-trained spiking neural network model for time series analysis, as follows: Using the aforementioned pulsation neural network model, a correlation time sequence pattern is established between the temperature pulse sequence and the strain pulse sequence under normal thermodynamic response; wherein, the correlation time sequence pattern is formed based on a pulse time sequence dependent plasticity mechanism. In real-time monitoring, the strain pulse sequence input in real time is compared with the established associated timing pattern, and the timing deviation of the strain pulse sequence is determined based on the comparison result.

[0008] Furthermore, the pre-trained spiking neural network model is constructed through the following steps: Acquire historical temperature and strain data of the bridge monitoring area under structural health conditions; The historical temperature data and historical strain data are converted into historical temperature pulse sequences and historical strain pulse sequences, respectively, using an asynchronous event-driven differential modulation method. The historical temperature pulse sequence and the historical strain pulse sequence are used as training sample pairs, input into an initial spiking neural network model, and trained using a pulse time-dependent plasticity mechanism. The training process enables the spiking neural network model to establish a temporal correlation pattern between the historical temperature pulse sequence and the historical strain pulse sequence under normal thermodynamic response.

[0009] Furthermore, based on the time series analysis results, an indication signal is output as follows: When the strain pulse sequence is detected to deviate from the associated timing pattern, an abnormality indication signal is output; otherwise, a normal indication signal is output.

[0010] Furthermore, the strain pulse sequence deviates from the associated timing pattern as follows: The strain pulse is missing within the preset timing window for the strain pulse to appear, or the strain pulse bursts out when there is no timing correlation with the temperature pulse.

[0011] Furthermore, in S4, a digital twin model of the bridge is constructed: Obtain the geometric information and material parameters of the bridge monitoring area; Based on the geometric information and material parameters, a thermo-mechanical coupling constitutive relation is generated, and a digital twin model of the bridge is constructed using the finite element analysis method. The bridge digital twin model is used to simulate the theoretical strain response under temperature data changes and to compensate for temperature effects on the measured strain data.

[0012] Furthermore, in S4, based on the indication signal output by the spiking neural network model, temperature effect compensation is performed on the strain data using a pre-constructed bridge digital twin model, as follows: When the indicator signal output by the spiking neural network model is a normal indicator signal, the bridge digital twin model uses a first compensation strategy based on the linear thermal expansion theory to compensate for the strain data. When the indicator signal output by the spiking neural network model is an abnormal indicator signal, the bridge digital twin model switches to a second compensation strategy based on nonlinear fracture mechanics to compensate for the strain data.

[0013] Furthermore, the first compensation strategy is as follows: based on the linear thermal expansion theory, the theoretical thermal strain value is calculated according to the change of the temperature data, and the theoretical thermal strain value is subtracted from the measured strain data; The second compensation strategy is as follows: taking the moment when the abnormal indication signal is triggered as the reference point, freezing the temperature compensation calculation based on the first compensation strategy, and identifying the residual between the measured strain data and the frozen theoretical thermal strain value after the reference point as a component related to structural damage for output or recording.

[0014] Compared with the prior art, the present invention can achieve at least the following beneficial effects: 1. This invention, by setting asynchronous event-driven differential modulation, abandons the traditional fixed-frequency sampling clock. Pulse events are generated only when the change in temperature or strain data exceeds a preset threshold, converting continuous, slowly changing physical signals into discrete pulse sequences. This shifts the focus from absolute time-based signal values ​​to event-driven relative changes, thus removing the time scale of the physical process. It eliminates the need for global high-frequency sampling to capture microsecond-level cracks, avoiding massive data redundancy. Simultaneously, its sensitivity to the rate of change allows for denser pulse generation and automatically improved resolution during the rapid transient phase of thermal shock, ensuring that rapid events such as crack propagation are not missed. This expands the time scale from minutes to hours. The thermal process is mapped to an event step sequence that can be directly processed by the spiking neural network model. The spiking neural network model only needs to process the order and interval pattern of the pulse events, which allows mature spiking neural network algorithms and hardware to be directly applied to bridge health monitoring, solving the core obstacle of mismatch between the algorithm model and the physical signal scale. Through time series analysis, the spiking neural network model can intelligently determine whether the strain pulse deviates from the normal correlation pattern with the temperature pulse trained by the health data, and thus output an indication signal. It can effectively distinguish between normal temperature strain and abnormal strain modulated by damage such as cracks in the background of strong, minute-hour-level thermal shock noise, and finally achieve accurate capture and separation of micro-fatigue crack propagation signals.

[0015] 2. This invention converts minute-level slow-varying temperature and strain signals into pulse sequences decoupled from physical time through asynchronous event-driven encoding, fundamentally solving the contradiction between sampling frequency and crack propagation timescale. At the same time, it enables the spiking neural network to process engineering slow-varying signals without modifying its millisecond-level time constant, overcoming the model mismatch problem. Furthermore, through the temporal analysis of the spiking neural network, it intelligently distinguishes between normal thermal strain and abnormal damage strain, and drives the digital twin model to dynamically switch compensation strategies, ultimately achieving accurate capture and separation of transient signals of micro-fatigue cracks under strong thermal shock noise. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments 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. Wherein: Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar symbols denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0018] Please refer to Figure 1 This invention provides a bridge temperature effect compensation method based on digital twins, comprising the following steps: S1. Obtain temperature and strain data for the bridge monitoring area; It should be noted that the acquired temperature and strain data specifically refer to the real-time signals collected by surface temperature sensors and structural strain sensors deployed in the conductive concrete area of ​​the bridge; these sensors are preferably deployed in the form of a wireless sensor network to achieve remote, synchronous data acquisition and transmission.

[0019] S2. Based on the temperature data and the strain data, generate corresponding temperature pulse sequences and strain pulse sequences respectively through asynchronous event-driven differential modulation; In S2, based on the temperature data and the strain data, corresponding temperature pulse sequences and strain pulse sequences are generated respectively through asynchronous event-driven differential modulation, as follows: Asynchronous event-driven differential modulation is performed on the temperature data and the strain data respectively to generate corresponding temperature pulse sequences and strain pulse sequences; The asynchronous event-driven differential modulation process is as follows: continuously compare the current value of the monitoring data with a dynamically updated reference value. When the difference between the current value and the dynamically updated reference value is greater than or equal to a preset change threshold, trigger a pulse event, generate a pulse that represents the change direction of this pulse event, and adjust the reference value to the current value. The monitoring data is temperature data or strain data. By continuously triggering the pulse events, a temperature pulse sequence and a strain pulse sequence are formed, which are composed of pulses in the triggering order.

[0020] The preset change threshold includes a temperature threshold set for changes in temperature data and a strain threshold set for changes in strain data. In the pulse sequence, each pulse event is characterized by the polarity of the pulse, and the sequential relationship between pulse events is characterized by the trigger interval.

[0021] It should be noted that the asynchronous event-driven differential modulation process specifically involves maintaining a dynamically updated reference value for both the temperature data stream and the strain data stream. And preset the threshold of change ( For temperature data, set a temperature threshold. (For example For strain data, set a strain threshold. (For example ). Continuously compare the current value of the sensor readings ( ) and the corresponding benchmark value Only when the difference Reaching or exceeding its preset threshold A valid event is triggered only once per event. If the difference is positive and exceeds the threshold, a positive (ON) pulse is generated; if the difference is negative and exceeds the threshold, a negative (OFF) pulse is generated. After each pulse trigger, the reference value is immediately reset. Update to current value This new benchmark is used to continue monitoring subsequent changes. By continuously triggering such pulse events, the original continuous temperature / strain data stream is converted into an asynchronous pulse sequence consisting of pulse polarity (characterizing the direction of change) and the trigger interval between pulses.

[0022] It's important to note that this encoding method achieves decoupling of the physical time scale. It doesn't record the absolute time of an event, but rather the event itself—that the change exceeds a threshold—and its relative order of occurrence. For example, regardless of whether the bridge experiences a linear cooling process lasting 10 minutes or 1 hour, as long as the total cooling magnitude is the same (e.g., a decrease in temperature), the encoding will be effective. ), and threshold The encoder will stably output 100 negative polarity pulses. The shape of the pulse sequence (the number of pulses and the polarity order) is determined only by the amplitude curve of the physical quantity change, and is independent of the absolute time taken for the change process.

[0023] It should be noted that in the generated pulse sequence, the core information of each pulse event consists of two parts: first, the polarity of the pulse (positive / negative), which directly corresponds to whether the physical quantity (temperature or strain) has increased or decreased beyond a threshold; and second, the trigger interval between the event and the previous event. The order and interval of all events together define the spatiotemporal pattern of the pulse sequence. It is this separation of absolute time that enables this invention to uniformly map thermo-mechanical processes at different time scales into an event step sequence that can be efficiently processed by a spiking neural network, thereby bridging the gap between macroscopic thermal shock monitoring and microscopic damage transient capture.

[0024] S3. Input the temperature pulse sequence and the strain pulse sequence into a pre-trained pulse neural network model for time series analysis, and output an indication signal based on the time series analysis results; It should be noted that step S3 is the core intelligent decision-making module of the present invention, which uses a spiking neural network model to perform biomimetic time-series analysis on the encoded pulse sequence, aiming to accurately separate abnormal signals representing structural damage from complex thermal shock noise.

[0025] In S3, the temperature pulse sequence and the strain pulse sequence are input into a pre-trained spiking neural network model for time series analysis, as follows: Using the aforementioned pulsation neural network model, a correlation time sequence pattern is established between the temperature pulse sequence and the strain pulse sequence under normal thermodynamic response; wherein, the correlation time sequence pattern is formed based on a pulse time sequence dependent plasticity mechanism. In real-time monitoring, the strain pulse sequence input in real time is compared with the established associated timing pattern, and the timing deviation of the strain pulse sequence is determined based on the comparison result.

[0026] A pre-trained spiking neural network model is constructed through the following steps: Acquire historical temperature and strain data of the bridge monitoring area under structural health conditions; The historical temperature data and historical strain data are converted into historical temperature pulse sequences and historical strain pulse sequences, respectively, using an asynchronous event-driven differential modulation method. The historical temperature pulse sequence and the historical strain pulse sequence are used as training sample pairs, input into an initial spiking neural network model, and trained using a pulse time-dependent plasticity mechanism. The training process enables the spiking neural network model to establish a temporal correlation pattern between the historical temperature pulse sequence and the historical strain pulse sequence under normal thermodynamic response.

[0027] It should be noted that the pre-trained spiking neural network model is preferably a dual-channel input network built based on a leak integral-fire neuron model. Its structural design has a clear physical meaning: the first input channel is dedicated to receiving temperature pulse sequences, and the second input channel is dedicated to receiving strain pulse sequences. The leak integral-fire neurons in each channel perform spatiotemporal integration of the input pulses. The core of the network's training and operation lies in utilizing the biomimetic learning rule of pulse temporal dependence plasticity. This rule, by adjusting the weights of connections between neurons, enables the network to learn and memorize the inherent causal relationship between two pulse sequences in time.

[0028] It should be noted that the associated timing pattern refers to the stable pulse timing relationship exhibited by the strain response caused by temperature changes under the healthy state of the bridge structure. For example, in the process of normal thermal expansion and contraction, after a negative pulse event indicating a decrease in temperature (from the first channel), a strain negative pulse event indicating contraction (from the second channel) will inevitably follow within a very short and stable time window (such as within an event step of a few milliseconds).

[0029] The training process of a spiking neural network model is the process of learning this health pattern: Data preparation: Collect historical temperature and strain data of the bridge under known health conditions and under various typical temperature changes (including simulated thermal shock).

[0030] Data encoding: Using the same asynchronous event-driven differential modulation method as real-time monitoring, historical data is converted into paired historical temperature pulse sequences and historical strain pulse sequences.

[0031] Model Training: These paired pulse sequences are used as training samples and input into the initialized spiking neural network model. The network performs unsupervised or self-supervised learning through the pulse temporal dependence plasticity mechanism: when the temperature pulse (preceding pulse) precedes the strain pulse (following pulse), the synaptic weights connecting these two events are strengthened. After training with a large amount of health data, this strong temporal correlation pattern of temperature pulse preceding strain pulse will be solidified within the network, forming a memory imprint representing the intrinsic relationship of thermo-mechanical response under structurally intact conditions.

[0032] Based on the timing analysis results, the following indicator signal is output: When the strain pulse sequence is detected to deviate from the associated timing pattern, an abnormality indication signal is output; otherwise, a normal indication signal is output.

[0033] The strain pulse sequence deviates from the associated timing pattern as follows: The strain pulse is missing within the preset timing window for the strain pulse to appear, or the strain pulse bursts out when there is no timing correlation with the temperature pulse.

[0034] It should be noted that during the real-time monitoring phase, the synchronously generated real-time temperature pulse sequence and strain pulse sequence are input into the pre-trained spiking neural network model. The spiking neural network model will perform real-time comparative analysis based on the fixed correlation time series patterns.

[0035] Normal state: If the timing relationship between the real-time strain pulse and the temperature pulse conforms to the learned health pattern (i.e., the strain pulse appears immediately after the corresponding temperature pulse within the expected time window), the firing activity of the neurons inside the network is stable, and eventually outputs a normal indication signal (or remains silent).

[0036] Abnormal state (timing deviation): When thermal shock induces the initiation or propagation of micro-fatigue cracks, the local stress field undergoes abrupt changes, leading to abnormal strain response. This abnormality manifests at the pulse sequence level as two typical timing deviations: Missing pulse: After a temperature pulse, the strain pulse that should have appeared within the expected time window does not occur. This may be because the crack releases local stress, causing the material to fail to produce the expected shrinkage strain.

[0037] Pulse burst: A sudden, dense series of pulses appears in the strain channel without any preceding temperature pulse triggering, or completely unrelated to the timing of the temperature pulses. This usually corresponds to stress waves and localized plastic deformation caused by instantaneous crack propagation or instability.

[0038] Any of the above-mentioned timing deviations will disrupt the balance established within the spiking neural network model through the pulse timing-dependent plasticity rule, causing downstream integrated neurons to fail to reach the firing threshold on time or to cause abnormally high-frequency firing, thereby triggering abnormal indicator signals to be output by the network.

[0039] It's important to note that using a spiking neural network model for time-series analysis offers a fundamental advantage over traditional algorithms based on thresholding or fixed-frequency spectrum analysis: it is sensitive to relationships, not absolute values. It doesn't concern itself with the specific numerical values ​​of temperature or strain, but only with whether the temporal coordination between the two pulse events is disrupted. This allows it to effectively distinguish between severe, normal thermal contraction (numerous temperature and strain pulses with a normal temporal relationship) and abnormal strain jumps caused by damage (abnormal temporal relationships). This enables the capture of damage events with an extremely high signal-to-noise ratio even in noisy environments, fundamentally avoiding the false alarm problem of traditional methods. This mechanism simulates the keen detection ability of biological nervous systems for abnormal patterns in perceptual information.

[0040] S4. Based on the indication signal output by the spiking neural network model, temperature effect compensation is performed on the strain data through a pre-constructed bridge digital twin model; wherein, the bridge digital twin model includes the thermo-mechanical coupling constitutive relationship of the bridge monitoring area.

[0041] It should be noted that by creating a bridge digital twin model that can intelligently understand the structural state and dynamically adjust the compensation logic, the true structural health status can be accurately extracted from strong thermal shock noise.

[0042] In S4, a digital twin model of the bridge is constructed: Obtain the geometric information and material parameters of the bridge monitoring area; Based on the geometric information and material parameters, a thermo-mechanical coupling constitutive relation is generated, and a digital twin model of the bridge is constructed using the finite element analysis method. The bridge digital twin model is used to simulate the theoretical strain response under temperature data changes and to compensate for temperature effects on the measured strain data.

[0043] It should be noted that the pre-built digital twin model of the bridge is a high-fidelity, parametric computer simulation model. Its construction is an offline process based on physical principles. Data input: First, it is necessary to obtain accurate geometric information (such as cross-sectional dimensions, reinforcement layout, thickness and range of conductive concrete layer) and material parameters (such as concrete elastic modulus, Poisson's ratio, resistivity of conductive concrete, coefficient of thermal expansion, and key thermo-mechanical coupling parameters) of the target monitoring area.

[0044] The core of the model is to establish a thermo-mechanical coupling constitutive relation based on the above information. This is particularly important for conductive concrete because its mechanical properties (stress-strain relationship) are closely coupled with the temperature field, especially when the internal temperature changes drastically due to power switching. This model needs to be able to describe the nonlinear changes in these material properties with temperature.

[0045] Construction Method: The finite element method is used to discretize the actual bridge structure into a mesh model composed of a large number of elements. The aforementioned material constitutive relations are then assigned to the corresponding elements, thereby constructing a digital twin model capable of simulating the internal stress and strain response of the structure under arbitrary temperature field changes. The core function of this model is: given an input temperature change history, it can calculate and output the strain response (i.e., theoretical thermal strain value) that the bridge should theoretically produce, which is purely caused by thermal effects.

[0046] In S4, based on the indicator signal output by the spiking neural network model, temperature effect compensation is performed on the strain data using a pre-constructed bridge digital twin model, as follows: When the indicator signal output by the spiking neural network model is a normal indicator signal, the bridge digital twin model uses a first compensation strategy based on the linear thermal expansion theory to compensate for the strain data. When the indicator signal output by the spiking neural network model is an abnormal indicator signal, the bridge digital twin model switches to a second compensation strategy based on nonlinear fracture mechanics to compensate for the strain data.

[0047] The first compensation strategy is as follows: based on the linear thermal expansion theory, the theoretical thermal strain value is calculated according to the change of the temperature data, and the theoretical thermal strain value is subtracted from the measured strain data; The second compensation strategy is as follows: taking the moment when the abnormal indication signal is triggered as the reference point, freezing the temperature compensation calculation based on the first compensation strategy, and identifying the residual between the measured strain data and the frozen theoretical thermal strain value after the reference point as a component related to structural damage for output or recording.

[0048] It should be noted that the digital twin of this invention is not a static reference model, but a dynamic compensation engine with dual states, whose state switching is entirely driven by indication signals provided by a spiking neural network model. First Compensation Strategy (Conventional Linear Mode): When the spiking neural network model outputs a normal indication signal, it indicates that the current temperature-strain response conforms to the correlation time-series pattern of a healthy structure, meaning all strain changes can be attributed to normal thermal expansion and contraction. In this case, the digital twin model employs the first compensation strategy based on the theory of linear thermal expansion. Specifically, it calculates the current theoretical thermal strain value based on real-time acquired temperature data and then directly subtracts this value from the measured total strain data. Theoretically, the residual after compensation should be close to zero or contain only measurement noise, thus obtaining a net structural response that eliminates temperature effects.

[0049] The second compensation strategy (nonlinear correction mode): When the spiking neural network model outputs an abnormal indication signal, it indicates that a temporal deviation in the strain pulse sequence has been detected, suggesting that damage such as fatigue cracks may be starting to occur under thermal shock. At this point, the system immediately switches to the second compensation strategy based on nonlinear fracture mechanics. This strategy recognizes that during the damage occurrence and development phase, classical thermal expansion theory is insufficient to describe the complex stress redistribution and energy release processes in the damage region.

[0050] It should be noted that the key operations and principles of the second compensation strategy are as follows: The core of the second compensation strategy is freezing the benchmark and identifying residuals, and its specific implementation is as follows: Determine the reference point: the moment when the spiking neural network model first issues an anomaly indication signal. As a key benchmark.

[0051] Freeze Compensation Benchmark: Immediate Freeze Digital Twin Model in The state at any given time, specifically the reference baseline frozen based on the theoretical thermal strain values ​​calculated using the first strategy. This means that for All data after that time will continue to be used. The theoretical thermal strain value at that moment or the thermal strain value calculated based on the state at that moment, rather than continuing to calculate using the real-time temperature according to the conventional formula.

[0052] Identify damage components: In After a certain time, the measured strain data is compared with the theoretical thermal strain value of the aforementioned freezing point, and the resulting residual is separated out. This residual is no longer considered as an error due to imperfect temperature effect compensation, but is positively identified as a component that is highly likely to be caused by structural damage (such as crack propagation), and is then specifically output, recorded, and analyzed.

[0053] It should be noted that, through the intelligent state discrimination of the spiking neural network model, when suspected damage occurs, the overfitting to thermal noise is actively stopped, and all subsequent changes deviating from classical thermodynamics are conservatively attributed to the potential damage. This is equivalent to establishing a data isolation zone for damage events in the digital world for the physical structure, ensuring that even under extreme thermal noise backgrounds, the weak signals of microscopic damage can be effectively captured and preserved, providing an unprecedentedly accurate data foundation for safety early warning and life assessment of bridge structures.

[0054] In summary, this patent application resolves the sampling scale contradiction through event-driven coding, solves the computational model mismatch through pulse sequence normalization, and achieves reliable monitoring of transient signals of microscopic damage under thermal shock conditions through the synergy of neuromorphic temporal analysis and digital twin dynamic compensation.

[0055] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0056] In the embodiments provided by this invention, it should be understood that the disclosed system or method can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0057] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0058] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0059] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the basic characteristics of the present invention.

[0060] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A bridge temperature effect compensation method based on digital twin, characterized in that, Includes the following steps: S1. Obtain temperature and strain data for the bridge monitoring area; S2. Based on the temperature data and the strain data, generate corresponding temperature pulse sequences and strain pulse sequences respectively through asynchronous event-driven differential modulation; S3. Input the temperature pulse sequence and the strain pulse sequence into a pre-trained pulse neural network model for time series analysis, and output an indication signal based on the time series analysis results; S4. Based on the indication signal output by the spiking neural network model, temperature effect compensation is performed on the strain data through a pre-constructed bridge digital twin model; wherein, the bridge digital twin model includes the thermo-mechanical coupling constitutive relationship of the bridge monitoring area.

2. The bridge temperature effect compensation method based on digital twin as described in claim 1, characterized in that: In S2, based on the temperature data and the strain data, corresponding temperature pulse sequences and strain pulse sequences are generated respectively through asynchronous event-driven differential modulation, as follows: Asynchronous event-driven differential modulation is performed on the temperature data and the strain data respectively to generate corresponding temperature pulse sequences and strain pulse sequences; The asynchronous event-driven differential modulation process is as follows: continuously compare the current value of the monitoring data with a dynamically updated reference value. When the difference between the current value and the dynamically updated reference value is greater than or equal to a preset change threshold, trigger a pulse event, generate a pulse that represents the change direction of this pulse event, and adjust the reference value to the current value. The monitoring data is temperature data or strain data. By continuously triggering the pulse events, a temperature pulse sequence and a strain pulse sequence are formed, which are composed of pulses in the triggering order.

3. The bridge temperature effect compensation method based on digital twin as described in claim 2, characterized in that: The preset change threshold includes a temperature threshold set for changes in temperature data and a strain threshold set for changes in strain data. In the pulse sequence, each pulse event is characterized by the polarity of the pulse, and the sequential relationship between pulse events is characterized by the trigger interval.

4. The bridge temperature effect compensation method based on digital twin as described in claim 1, characterized in that: In S3, the temperature pulse sequence and the strain pulse sequence are input into a pre-trained spiking neural network model for time series analysis, as follows: Using the aforementioned pulsation neural network model, a correlation time sequence pattern is established between the temperature pulse sequence and the strain pulse sequence under normal thermodynamic response; wherein, the correlation time sequence pattern is formed based on a pulse time sequence dependent plasticity mechanism. In real-time monitoring, the strain pulse sequence input in real time is compared with the established associated timing pattern, and the timing deviation of the strain pulse sequence is determined based on the comparison result.

5. The bridge temperature effect compensation method based on digital twin as described in claim 4, characterized in that: A pre-trained spiking neural network model is constructed through the following steps: Acquire historical temperature and strain data of the bridge monitoring area under structural health conditions; The historical temperature data and historical strain data are converted into historical temperature pulse sequences and historical strain pulse sequences, respectively, using an asynchronous event-driven differential modulation method. The historical temperature pulse sequence and the historical strain pulse sequence are used as training sample pairs, input into an initial spiking neural network model, and trained using a pulse time-dependent plasticity mechanism. The training process enables the spiking neural network model to establish a temporal correlation pattern between the historical temperature pulse sequence and the historical strain pulse sequence under normal thermodynamic response.

6. The bridge temperature effect compensation method based on digital twin as described in claim 5, characterized in that: Based on the timing analysis results, the following indicator signal is output: When the strain pulse sequence is detected to deviate from the associated timing pattern, an abnormality indication signal is output; otherwise, a normal indication signal is output.

7. The bridge temperature effect compensation method based on digital twin as described in claim 6, characterized in that: The strain pulse sequence deviates from the associated timing pattern as follows: The strain pulse is missing within the preset timing window for the strain pulse to appear, or the strain pulse bursts out when there is no timing correlation with the temperature pulse.

8. The bridge temperature effect compensation method based on digital twin as described in claim 1, characterized in that: In S4, a digital twin model of the bridge is constructed: Obtain the geometric information and material parameters of the bridge monitoring area; Based on the geometric information and material parameters, a thermo-mechanical coupling constitutive relation is generated, and a digital twin model of the bridge is constructed using the finite element analysis method. The bridge digital twin model is used to simulate the theoretical strain response under temperature data changes and to compensate for temperature effects on the measured strain data.

9. The bridge temperature effect compensation method based on digital twin as described in claim 8, characterized in that: In S4, based on the indicator signal output by the spiking neural network model, temperature effect compensation is performed on the strain data using a pre-constructed bridge digital twin model, as follows: When the indicator signal output by the spiking neural network model is a normal indicator signal, the bridge digital twin model uses a first compensation strategy based on the linear thermal expansion theory to compensate for the strain data. When the indicator signal output by the spiking neural network model is an abnormal indicator signal, the bridge digital twin model switches to a second compensation strategy based on nonlinear fracture mechanics to compensate for the strain data.

10. The bridge temperature effect compensation method based on digital twin as described in claim 9, characterized in that: The first compensation strategy is as follows: based on the linear thermal expansion theory, the theoretical thermal strain value is calculated according to the change of the temperature data, and the theoretical thermal strain value is subtracted from the measured strain data; The second compensation strategy is as follows: taking the moment when the abnormal indication signal is triggered as the reference point, freezing the temperature compensation calculation based on the first compensation strategy, and identifying the residual between the measured strain data and the frozen theoretical thermal strain value after the reference point as a component related to structural damage for output or recording.

Images