System for correcting stress monitoring data under influence of temperature of continuous beam bridge

Abstract

The invention relates to a stress monitoring data correction system for a continuous beam bridge under the temperature influence, in particular to the field of bridge health monitoring, and effectively solves the problem of bridge stress monitoring data distortion under the temperature influence through multi-module collaborative innovation. The trend modeling module dynamically predicts a drift evolution law and quantifies a confidence level, the compensation correction module fuses physical constraints to generate an anti-interference drift compensation amount, and the health prediction module realizes accurate life evaluation based on a time-varying mechanism, so that finally, the reliability of monitoring data is remarkably improved, and misjudgment of a structural damage state is avoided; maintenance decisions are optimized, full-life-cycle safety of the bridge is guaranteed, and meanwhile invalid operation and maintenance cost caused by false alarms is reduced.

Description

Technical Field

[0001] This invention relates to the field of bridge health monitoring, and more specifically, to a system for correcting stress monitoring data under the influence of temperature on continuous beam bridges. Background Technology

[0002] In the field of modern bridge engineering, continuous beam bridges are widely used due to their excellent spanning capacity and adaptability. To ensure the safe service of such large infrastructures during their design life of 20 to 30 years, structural health monitoring systems based on the concepts of fault prediction and health management have become an indispensable technical means. Especially under the significant influence of temperature effects, the accuracy of stress monitoring is directly related to the assessment and prediction of the safety status of bridge structures. Such systems usually require the long-term deployment of a large number of stress sensors in key areas of the bridge, such as beam joints and temperature-sensitive parts, to collect multi-variable data such as structural response and environmental temperature and humidity in real time. However, the service environment of bridges is often extreme and complex and variable, inevitably facing the combined effects of external factors such as periodic freeze-thaw cycles, continuous high humidity environments, and chemical media corrosion. At the same time, under the long-term continuous power-on operation, the sensor's sensitive elements and circuits will inevitably undergo internal degradation phenomena such as material aging and electrical characteristic drift. In addition, the possible slight loosening of the mounting base will cause the sensor output signal to accumulate a slow but continuous baseline shift or sensitivity change over time, i.e., sensor drift error. This phenomenon is common in the entire life cycle operation of structural health monitoring systems.

[0003] In existing technologies, the main approach to addressing long-term sensor drift issues relies on manual calibration at fixed time intervals. This includes periodic offline calibration using standard metrology instruments, or the application of traditional algorithms like Kalman filters for noise filtering and simple linear error compensation during data processing. However, these methods have significant limitations. First, manual calibration cycles are typically long, often monthly or even quarterly. During these intervals, drift errors inevitably accumulate, failing to meet the demands of real-time, accurate monitoring. Second, traditional data processing algorithms, such as standard Kalman filters, are usually based on linear or quasi-linear system assumptions, making it difficult to effectively characterize and compensate for the nonlinear and time-varying characteristics of sensor drift, especially the coupling relationships between different types of drift. Finally, and most importantly, there is an inherent deficiency in the design of existing technologies for fault prediction and health management frameworks. Current PHM models, when constructing structural health status predictions, mainly focus on the damage evolution and mechanical response of the structure itself, and generally lack explicit modeling and online prediction capabilities for the significant interfering factor of sensor performance degradation. This leads to a serious practical application challenge: during long-term operation, especially during periods of drastic external temperature fluctuations, spurious signals generated by sensor drift are easily misjudged as the actual stress response of the bridge structure due to temperature changes or signals of structural damage initiation. This seriously interferes with the accurate pre-diagnosis of the long-term health trend of the bridge by the PHM system. Such data distortion cannot be reliably identified and distinguished in bridge stress monitoring systems, forming a core technical bottleneck that urgently needs to be addressed. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing a stress monitoring data correction system for continuous beam bridges under the influence of temperature. Through multi-module collaborative innovation, it effectively solves the problem of distortion in bridge stress monitoring data under the influence of temperature, thereby resolving the issues mentioned in the background art.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: Specifically, it includes: a drift perception module, a trend modeling module, a compensation and correction module, and a health prediction module, wherein... Drift sensing module: When real-time stress data stream and ambient temperature and humidity data are input synchronously, it dynamically captures signal segments through an adaptive sliding window and applies a wavelet-empirical mode decomposition hybrid algorithm to separate the baseline drift component and the true stress component in the stress signal. Trend modeling module: configured to receive the baseline drift component and associated environmental variables, construct a dual-path deep reinforcement learning ensemble model, wherein the first path uses a long short-term memory network to encode environmental temporal features, the second path establishes a drift state transition equation, and finally outputs the drift prediction value and confidence level through a collaborative decision-maker; Compensation and correction module: configured to, based on the drift prediction value, introduce the theoretical quantity of structural thermal expansion as a physical constraint and drive the physical-guided adversarial generative network to generate drift compensation amount; Health prediction module: configured to fuse the drift compensation amount with the original stress data, perform the remaining life integral equation calculation with time-varying gain factor, and output the bridge structure health status and remaining life prediction results; In a preferred embodiment, the specific operation of the adaptive sliding window dynamically capturing signal segments in the drift sensing module is as follows: First, the phase space of the real-time stress data stream is reconstructed. The optimal time delay parameter is determined by calculating the minimum value of the mutual information function, and a multi-dimensional phase space trajectory matrix is ​​constructed with an embedding dimension of 3. Second, the maximum Lyapunov exponent of the phase space trajectory is calculated. Then, the window length is dynamically calculated based on the initial disturbance modulus threshold and the preset allowable error, combined with the reciprocal relationship of the maximum Lyapunov exponent. Finally, the environmental temperature and humidity data are synchronously locked into the window in a timestamp-aligned manner, and a fixed-length signal segment and associated environmental variables are output.

[0006] In a preferred embodiment, the specific process of separating the baseline drift component and the true stress component in the stress signal is as follows: A1. Calculate the local singularity index for the stress data within the window and generate the singularity weight distribution function; A2. The process of generating upper and lower envelopes based on singularity-weighted adjustment of empirical mode decomposition: the Hermite interpolation method is used to construct the weighted envelope, and the intrinsic mode function component set and residual components are obtained by iterative decomposition. A3. Perform discrete wavelet transform based on db8 wavelet basis functions on the residual components to decompose them into multi-level wavelet coefficients, and calculate the wavelet entropy of each level. A4. Establish an entropy-temperature coupling criterion based on the relationship between the rate of change of ambient temperature and the partial derivative of wavelet entropy. When the absolute value of the partial derivative exceeds a predetermined threshold, mark the corresponding segment of the current residual component as the baseline drift component. A5. Separate the baseline drift component from the original residual components and reorganize the intrinsic mode function components to output the true stress components.

[0007] In a preferred embodiment, the specific operation of encoding the temporal features of the long short-term memory network environment in the trend modeling module is as follows: First, the system receives a data stream of associated environmental variables from the drift sensing module, which includes synchronous time-series records of temperature and humidity parameters. Second, it constructs a bidirectional long short-term memory network architecture with a fixed number of hidden layer units of 64. Then, it sets the environmental analysis time window to 1200 seconds and performs a bidirectional time-dimensional scan of the temperature and humidity data within this window to capture the lagging and leading effects of historical and future environmental states on drift, respectively. Finally, it outputs an environmental feature vector that fuses the bidirectional features.

[0008] In a preferred embodiment, the process of constructing the drift state transition equation is as follows: First, the current value of the baseline drift component at the previous moment is obtained. Second, the environmental feature vector is extracted and input into the temperature change rate response gating function to generate environmental factor weighting coefficients. Then, the hardware aging effect is converted by a logarithmic scaling factor, based on the cumulative power-on time of the sensor. Next, the three sets of parameters—the baseline drift component, the environmental factor weighting coefficients, and the hardware aging effect—are substituted into the state transition equation, which uses a modified linear unit as the nonlinear activation function. Finally, the predicted drift state value is output.

[0009] In a preferred embodiment, the operation of the collaborative decision-maker in outputting the drift prediction value and confidence level is as follows: First, the environmental feature vector and the drift state prediction are merged to construct the decision state space. Second, the action space of the compensation adjustment magnitude is defined, with its upper and lower bounds symmetrically distributed. Then, the state action value function is calculated through a deep Q-network, with the hidden layer of this function adopting a two-layer fully connected structure. Subsequently, based on the absolute residual between the drift state prediction and the actual baseline drift component, and combined with the conditional mutual information entropy of the environmental feature vector and the drift state prediction, a normalized confidence index is generated. Finally, the compensation adjustment magnitude is output as the correction amount for the drift prediction, and the confidence index is used as the reliability indicator of health management decision.

[0010] In a preferred embodiment, the specific operation of introducing the theoretical quantity of structural thermal expansion as a physical constraint in the compensation and correction module includes: First, the characteristic length distribution of the temperature-sensitive region is analyzed using a bridge finite element model. Second, the real-time temperature difference value and the coefficient of thermal expansion of the material are obtained. Then, the median characteristic length, the temperature difference value, and the coefficient of thermal expansion are multiplied to generate the theoretical quantity of structural thermal expansion. Finally, this theoretical quantity is output to the input of the physics-guided adversarial generative network.

[0011] In a preferred embodiment, the operation process of the driving physical-guided adversarial generative network is as follows: First, a dual-channel temporal convolution generator is constructed, with the first channel inputting the drift prediction value and the second channel inputting the theoretical quantity of structural thermal expansion. Second, the discriminator incorporates the sensor aging physical equation constraint. Then, the generator generates an initial compensation amount based on the dual-channel input. Finally, a confidence-weighted fusion mechanism is used to correct the initial compensation amount: when the confidence level is higher than a threshold, the generated result is directly output; when it is lower than the threshold, it switches to a conservative correction mode based on baseline drift.

[0012] In a preferred embodiment, the specific operation of fusing the drift compensation amount with the original stress data in the health prediction module is as follows: First, the drift compensation amount and the original stress monitoring data stream are received; second, a drift elimination operation is performed, subtracting the drift compensation amount from the original stress monitoring data to generate the drift-corrected true stress value; then, the gradient field is calculated on the true stress value to obtain the multi-axis stress coupling gradient norm; at the same time, the environmental vibration energy spectral density parameter is extracted; finally, combined with the confidence parameter from the trend modeling module, a time-varying gain factor is generated through hyperbolic tangent function mapping.

[0013] In a preferred embodiment, the execution process of the remaining lifetime integral equation calculation with time-varying gain factor is as follows: First, the theoretical failure time point and the material failure stress threshold are set. Second, a time-varying decay factor with exponential decay characteristics is constructed, which gradually increases as time approaches the failure point. Then, the minimum value of the actual stress value during the service cycle is used as the fatigue strength inflection point criterion. Finally, a dual integral structure kernel function is established, which dynamically integrates the dynamic damage rate, decay factor and fatigue strength margin after gain modulation in the inner integral layer, and performs time-normalization calculation in the outer integral layer to output the remaining life prediction value and the three-level health status classification label.

[0014] The beneficial effects of this invention are as follows: through multi-module collaborative innovation, it effectively solves the problem of distortion of bridge stress monitoring data under the influence of temperature; the drift sensing module accurately separates environmental drift from the real stress signal; the trend modeling module dynamically predicts the drift evolution law and quantifies the confidence level; the compensation and correction module integrates physical constraints to generate anti-interference drift compensation amount; and the health prediction module achieves accurate life assessment based on time-varying mechanism. Ultimately, it significantly improves the reliability of monitoring data, avoids misjudging the structural damage state, optimizes maintenance decisions and ensures the safety of the bridge throughout its entire life cycle, while reducing the ineffective operation and maintenance costs caused by false alarms. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a block diagram of the system structure of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0018] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0019] Example 1 This embodiment provides, for example Figure 1-2 The system shown is a stress monitoring data correction system for continuous beam bridges under temperature influence, specifically including: a drift sensing module, a trend modeling module, a compensation correction module, and a health prediction module. Drift sensing module: Configured to dynamically capture signal segments through an adaptive sliding window and apply a wavelet-empirical mode decomposition hybrid algorithm when real-time stress data stream and ambient temperature and humidity data are input synchronously, it separates the baseline drift component and the true stress component in the stress signal. In continuous beam bridge stress monitoring systems, baseline drift caused by ambient temperature and humidity fluctuations and long-term sensor degradation can contaminate the true stress signal. This module uses dynamic feature-driven adaptive signal separation technology to extract high-fidelity stress components, laying a data foundation for subsequent drift modeling and compensation. Trend Modeling Module: Configured to receive baseline drift components and associated environmental variables, construct a dual-path deep reinforcement learning ensemble model. The first path uses a long short-term memory network to encode environmental temporal features, while the second path establishes the drift state transition equation. Finally, the drift prediction value and confidence level are output through a collaborative decision-maker. This module achieves high-precision modeling and confidence measurement of drift trends through a dual-path collaborative framework of physical constraints and deep reinforcement learning, providing a dynamic decision-making basis for subsequent compensation and correction. Compensation and Correction Module: Configured to be based on drift prediction values, introduce the theoretical quantity of structural thermal expansion as a physical constraint and drive a physical-guided adversarial generative network to generate drift compensation. This module uses a physical-guided adversarial generative network to inject the theoretical quantity of structural thermal expansion as a hard constraint into the network training, breaking through the limitations of traditional data-driven models and achieving accurate drift correction in temperature-sensitive scenarios. Health prediction module: Configured to fuse drift compensation with original stress data, perform residual life integral equation calculation with time-varying gain factor, and output bridge structure health status and residual life prediction results. This module achieves robust assessment of bridge health status through time-varying gain control and nonlinear damage integration, breaking through the limitations of traditional fixed threshold methods.

[0020] In this embodiment, it is specifically necessary to explain the specific operation of the adaptive sliding window dynamically capturing signal segments in the drift sensing module as follows: First, the real-time stress data stream Phase space reconstruction is performed. The optimal time delay parameter is determined by calculating the minimum value of the mutual information function, and a multidimensional phase space trajectory matrix is ​​constructed with an embedding dimension of 3. Its expression is as follows: ; in, Represents the reconstructed phase space trajectory matrix. express The original stress data at time 10:00. Indicates the embedding dimension (its value is set to ). ), This represents the time delay (interval parameter), calculated as follows: , This represents the minimum value of the mutual information function, used to calculate the nonlinear correlation of signals. In this formula, the mutual information entropy is used. The autocorrelation function is replaced to solve the time delay quantization problem of nonlinear systems; secondly, the maximum Lyapunov exponent of the phase space trajectory is calculated, which characterizes the divergence rate of the stress signal in the chaotic dynamic system; then, based on the initial perturbation mode length threshold and the preset allowable error, the window length is dynamically calculated in combination with the reciprocal relationship of the maximum Lyapunov exponent, and the formula is as follows: ; in, This represents the optimal window length, the signal segment length used for adaptive analysis, and its value ranges from [50, 500] sampling points. This represents the initial perturbation magnitude threshold, and its value is set to... , This indicates the tolerance, and its value is set to 0.01. The maximum Lyapunov exponent (used to describe the divergence rate of adjacent trajectories in phase space) is represented, with a value range of (0, +∞), and is calculated as follows: , Indicates the starting time of the data sequence. Indicates the end time of the data sequence. express Distance between adjacent trajectories at any given time (Euclidean distance). This represents the distance between adjacent trajectories at the initial moment; the window length adaptively expands or contracts according to the degree of signal chaos; finally, the ambient temperature and humidity data... Synchronize and lock the data within the window using timestamp alignment, and output a fixed-length signal segment and associated environment variables; The specific process of separating the baseline drift component and the true stress component in the stress signal is as follows: A1. For the stress data within the window... Calculate the local singularity index and generate the singularity weight distribution function. The formula for calculating the local singularity index is as follows: ; The formula for singularity weights is: ; in, This represents the stress data within the window (a fragment of the original stress signal extracted from the adaptive window). Indicates the current time point, Represents an infinitesimal quantity (the interval between adjacent sampling points is taken during calculation). Its value range is , Represents the local singularity index, where For smooth points, This represents a singular point; the smaller the value, the stronger the singularity. Its value ranges from [−1, 1]. The singularity weights are represented by... Exponential mapping generation is used to control the weighting of envelope generation and suppress singularity interference; its value range is... , Indicates no singularity. Indicates strong singularity. This represents the adjustment factor, used to control the sensitivity of the weights to singularities, and its value ranges from [value range missing]. (Default = 1) The larger the value, the more sensitive the weight is to singularity; A2. The process of generating upper and lower envelopes based on singularity-weighted adjusted empirical mode decomposition, using Hermite interpolation to construct the weighted envelope, the expression of which is: ; in, This represents the weighted envelope generation operator, used to output the corrected upper and lower envelopes. This represents the Hermite interpolation operator, used for smooth interpolation of the envelope point set to resolve endpoint effects. This represents the upper envelope (the sequence of local maxima of the original signal). This represents the lower envelope (the sequence of local minima of the original signal). Indicates the signal to be decomposed (i.e. The iterative decomposition yields the set of intrinsic mode function components and residual components, expressed as follows: ; in, Indicates the first The first-order eigenmode functions characterize vibration modes at different time scales. This represents the residual components, i.e., the low-frequency trend term (including the target drift component). The total number of IMF components is automatically determined by the iteration stopping condition; A3. Perform a discrete wavelet transform based on the db8 wavelet basis function on the residual components to decompose them into multiple wavelet coefficients, and calculate the wavelet entropy of each layer. The formula is as follows: ;in, Indicates the first Layer wavelet entropy (its range is) ), This represents the hierarchical index of the wavelet decomposition, and its value range is... , Indicates the first The sequence number of the wavelet coefficients within the layer, and its value range is: , Indicates the first Layer The normalized energy percentage of each coefficient (its value range is...) The calculation method is as follows: , Indicates the first Layer Wavelet coefficients (which are real numbers, generated by the wavelet basis function) are used in this equation to quantize the residual signal. Complexity at different frequency scales (layers); the higher the entropy value, the more chaotic the signal at that layer. A4. Based on the partial derivative relationship between the rate of change of ambient temperature and wavelet entropy, establish the entropy-temperature coupling criterion. The expression for the entropy-temperature coupling criterion is as follows: ; in, This represents the extracted baseline drift component. This represents the partial derivative of the wavelet entropy with respect to ambient temperature. The entropy threshold is set to 0.05. In this formula, the drift-dominant period is identified by the sensitivity of wavelet entropy to temperature (such as significant entropy changes when temperature changes abruptly), and the drift component is separated. When the absolute value of the partial derivative exceeds the predetermined threshold, the corresponding segment of the current residual component is marked as the baseline drift component. A5. Separate the baseline drift component from the original residual components and reconstruct the intrinsic mode function components to output the true stress components. The expression is as follows: ;in, This represents the corrected true stress value. This represents the total order of the IMF. Indicates the first eigenmode functions of order 1 Represents the residual components of empirical mode decomposition. The extracted baseline offset component is represented in this equation. First, the drift-free pure stress signal is removed through reconstruction. Second, the temperature sensitivity of wavelet entropy is used instead of a fixed threshold (as in traditional methods). This avoids misjudgments caused by subjective settings.

[0021] In this embodiment, it is specifically necessary to explain the specific operations of the long short-term memory network encoding environment temporal features in the trend modeling module as follows: First, the system receives a data stream of associated environmental variables from the drift sensing module. This data stream contains synchronized time-series records of temperature and humidity parameters. Secondly, a bidirectional long short-term memory network architecture is constructed, with a hidden layer unit number of... The time window is fixed at 64; then, the environmental analysis time window length is set to 1200 seconds. Within this window, a bidirectional time-dimension scan of the temperature and humidity data is performed to capture the lag and lead effects of historical and future environmental conditions on drift, respectively; finally, an environmental feature vector fused with bidirectional features is output. The dimension of this vector is consistent with the number of hidden layer units, and its expression is: ;in, This represents the environmental feature vector (encoding the temporal features of the environment for subsequent drift prediction). This represents the sequence of environmental variables within a time window (the input data format is temperature + humidity values ​​at n time points). This represents the current timestamp, used as a reference for aligning the data stream. This indicates the environmental analysis time window (set to 1200 seconds), used to control the length of historical data (approximately 20 minutes). This represents a bidirectional long short-term memory network, used to simultaneously extract forward and backward context dependencies. This represents the network weight matrix (parameters obtained during training, which determine the feature extraction capability). The process of constructing the drift state transition equation is as follows: First, obtain the current value of the baseline drift component from the previous moment; second, extract the environmental feature vector and input it into the temperature change rate response gating function to generate environmental factor weighting coefficients; then, combine the cumulative power-on time of the sensor and convert the hardware aging effect quantity through a logarithmic scaling factor; next, substitute the three sets of parameters—baseline drift component, environmental factor weighting coefficients, and hardware aging effect quantity—into the state transition equation, which uses a modified linear unit as the nonlinear activation function; finally, output the predicted drift state value, whose numerical range is consistent with the unit of the baseline drift component, and its expression is: ;in, This represents the predicted drift state value, which serves as the basis for compensation and correction in the next stage. This represents a corrected linear unit, used to ensure that the drift is non-negative and conforms to physical reality. Its calculation method is as follows: , This represents the drift state memory coefficient, used to control the weight of historical drift influence. This indicates the baseline drift at the previous moment. This represents the environmental response intensity coefficient, used to adjust for the impact of environmental shocks. This represents the Sigmoid gating function (with a value range of (0,1)), used to constrain the range of influence of sudden environmental changes (S1, when the environment changes drastically → gating value ≈ 1 → strong impact effect; S2, when the environment is stable → gating value ≈ 0 → suppressing noise interference), calculated as follows: , Represents the environmental feature weight vector. This represents an environmental feature vector, used to fuse historical / future environmental states. This represents the absolute value of the rate of temperature change, and it is calculated as follows: , This represents the aging effect coefficient (its value is set to 0.01 and calibrated through material testing). This indicates the duration of sensor power-on, used to characterize hardware fatigue strength; its value ranges from [0, +∞) seconds. This represents the time scaling factor (its value is set to 86400 seconds). This represents the logarithmic scaling function used to suppress the explosive growth of ultralong-term drift, in which... Quantify the short-term environmental impact. Characterizing long-term hardware degradation, the two are expressed through the gating factor λ and the aging factor. Achieve dynamic equilibrium; The collaborative decision-maker outputs drift predictions and confidence scores as follows: First, the environmental feature vector and the drift state prediction are merged to construct the decision state space. Secondly, define the adjustment range of the compensation amount, with its upper and lower bounds symmetrically distributed. Then, the state-action value function is calculated using a deep Q-network, and its formula is: ;in, Represents a state-space vector. Indicates the adjustment range of the compensation amount. This represents the weight matrix of a deep Q-network. This represents a modified linear unit activation function, used to enhance nonlinear modeling capabilities. The vector concatenation operation is used to merge state and action features. This function employs a two-layer fully connected hidden layer. Subsequently, based on the absolute residuals between the predicted drift state and the actual baseline drift components, and combined with the conditional mutual information entropy between the environmental feature vector and the predicted drift state, a normalized confidence index is generated, expressed as: ; in, The confidence level (with a value range of [0,1]) is a quantitative indicator of the reliability of the prediction results. This represents the attenuation factor (set to 0.1), used to adjust the sensitivity of the confidence level to the residuals. Indicates the baseline drift component. This represents the conditional mutual information between environmental characteristics and drift predictions, used to quantify the statistical dependence of the environment on drift (a higher value indicates higher confidence). Its calculation method is as follows: Finally, the adjustment range of the output compensation amount serves as the correction amount for the drift prediction value, and its expression is: ; in, This represents the final drift prediction value. This represents the predicted drift state value (input). This indicates that DQN outputs the optimal action, and its calculation method is as follows: The confidence index was used as an indicator of the reliability of health management decisions.

[0022] In this embodiment, it is specifically necessary to explain the specific operations of introducing the theoretical quantity of structural thermal expansion as a physical constraint in the compensation and correction module, including: First, the characteristic length distribution of the temperature-sensitive region is analyzed using a finite element model of the bridge. Second, the real-time temperature difference value and the coefficient of thermal expansion of the material are obtained. Then, the median characteristic length, the temperature difference value, and the coefficient of thermal expansion are multiplied together to generate the theoretical quantity of structural thermal expansion, the expression of which is: ; in, This represents the theoretical strain value (physical constraint reference) of the bridge due to temperature changes. The linear expansion coefficient of steel (reflecting an inherent property of the material) is set to a value of . , This represents the representative length of the temperature-sensitive region (such as beam joints) as defined by the finite element model. Non-homogeneous beams require partitioned calculations. This represents the difference between the current temperature and the ambient reference temperature (the reference temperature is the bridge's final closure temperature). The formula incorporates a spatially relevant characteristic length. (Non-homogeneous beams require partitioned calculations), replacing the traditional homogeneous material assumption; finally, the theoretical quantity is output to the input of the physics-guided adversarial generative network to ensure that it is input synchronously with the drift prediction value; The process of driving the physical-guided generative adversarial network is as follows: First, a dual-channel temporal convolution generator is constructed, with the first channel inputting the drift prediction value and the second channel inputting the theoretical quantity of structural thermal expansion. Its expression is as follows: ; in, Represents the generator input vector. This represents the drift prediction value. The first factor represents the theoretical quantity of structural thermal expansion; the second factor is the constraint imposed by the aging physical equation of the built-in sensor in the discriminator. This equation requires that the time-varying rate of the generator's output compensation quantity conforms to the exponential decay law of hardware degradation. The discriminator's physical constraint formula is as follows: ; in, The physical loss function (scalar) is used to quantify the deviation between the compensation quantity and the physical laws. This represents the derivative of the initial compensation amount with respect to time, characterizing the instantaneous rate of change of the compensation amount. This represents the drift attenuation coefficient (its value is set to...). ), This represents the time decay constant (set to 0.01), a parameter used to control the aging rate. Indicates the duration of sensor power-on. This represents the exponential decay term, characterizing the degradation effect of hardware under continuous power supply. This represents the initial compensation amount output by the generator; then, the generator generates the initial compensation amount based on the dual-channel input; finally, a confidence-weighted fusion mechanism is used to correct the initial compensation amount: when the confidence level is higher than a threshold... When (0.2), the generated result is output directly, and the confidence-driven correction formula is: ; in, This indicates the final drift compensation amount. This represents the confidence level, and its value ranges from [0,1]. This represents the initial compensation amount. This formula represents the baseline drift component and serves to dynamically balance the generated result with a conservative value. For example, when... (High confidence) The final compensation amount depends 90% on the generator result, when (Low confidence), 90% dependent on baseline drift value, below the threshold When switching to the baseline drift-based conservative correction mode, the formula for the conservative correction mode is: ; in, This represents the conservative weighting factor (its value range is [0,1]), and its calculation method is as follows: ( (represents the normalization coefficient) This expression represents baseline drift and serves to: resolve... The risk of divergence in the compensation amount at that time. In this embodiment, it is necessary to specifically explain the operation of fusing the drift compensation amount with the original stress data in the health prediction module as follows: First, the drift compensation amount and the original stress monitoring data stream are received; then, a drift elimination operation is performed, subtracting the drift compensation amount from the original stress monitoring data to generate the drift-corrected true stress value, the expression of which is: ;in, This represents the true stress after drift correction. This represents the original stress monitoring value. This represents the drift compensation amount. Its function is to: eliminate baseline drift caused by sensor aging or environmental interference, restoring the true structural response; subsequently, perform gradient field calculations on the true stress values ​​to obtain the multiaxial stress coupling gradient norm; simultaneously extract the environmental vibration energy spectral density parameters; finally, combine the confidence parameters from the trend modeling module, and generate a time-varying gain factor through hyperbolic tangent function mapping, used to modulate the contribution weight of the stress gradient to the damage rate. The gain control function formula is: ; in, This represents the time-varying gain factor, used to adjust the weight of the stress gradient's contribution to the damage rate; its value ranges from [0,1). This represents the confidence level, and its value ranges from [0,1]. This represents the hyperbolic tangent function, used for nonlinear mappings: (Trust damage rate) (Blocking abnormal data), in this formula, the function Compress the confidence level to the positive infinity interval, and then... Mapped to saturation gain, eliminating the risk of sudden damage rate changes at low confidence levels; The execution process of the remaining lifetime integral equation with time-varying gain factor is as follows: First, the theoretical failure time point and the material failure stress threshold are defined; second, a time-varying decay factor incorporating exponential decay characteristics is constructed, which gradually increases as time approaches the failure point. Its expression is: ; in, This represents the attenuation intensity coefficient (set to 0.05), used to control the attenuation rate: the larger the value, the higher the weight of long-term damage. This represents the margin before failure. The smaller the margin, the weaker the attenuation effect. (Approaching expiration) The attenuation effect disappears, enhancing the current damage. (Early service) The attenuation effect is enhanced, suppressing early noise interference; then, the minimum value of the actual stress during the service cycle is used as the fatigue strength inflection point criterion; finally, a double integral structure kernel function is established, the expression of which is: ; in, This represents the predicted remaining lifespan. Indicates the current assessment time. Indicates the theoretical failure time point (determined from material fatigue test data). Indicates dynamic damage rate. Represents the time-varying decay factor. This represents the material failure stress threshold (its value is determined by bridge design specifications). Indicates the historical minimum stress value ( The dynamic damage rate, attenuation factor, and fatigue strength margin after dynamic fusion gain modulation in the integral inner layer are calculated using the following formula: ; in, Represents the dynamic damage rate, used to quantify the accumulation of structural damage per unit time. It represents the true stress gradient norm, used to capture the spatiotemporal coupling effect of multiaxial stress (traditional methods use the Euclidean norm). This represents the energy of environmental vibration (data monitored by vibration sensors). This represents the stress damage weight (the proportion of the stress gradient contribution). This represents the vibration damage weight (the proportion of vibration energy contribution). 1.5 represents the nonlinear power term, used to amplify the impact of sudden vibration energy changes. A time-normalization operation is performed in the outer integral layer, outputting the predicted remaining lifespan and a three-level health status classification label. The content of the Level 3 health status classification label includes: when The health level is good, and the response action is routine monitoring. The health level is warning, and the response is increased monitoring plus manual inspection. The health level is dangerous, and the response action is to limit the passage and trigger a maintenance alarm.

[0023] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0024] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0025] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0026] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0027] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0028] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0029] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A system for correcting stress monitoring data under temperature influence in continuous beam bridges, characterized in that, Specifically, it includes: The module includes a drift perception module, a trend modeling module, a compensation and correction module, and a health prediction module. Drift sensing module: When real-time stress data stream and ambient temperature and humidity data are input synchronously, it dynamically captures signal segments through an adaptive sliding window and applies a wavelet-empirical mode decomposition hybrid algorithm to separate the baseline drift component and the true stress component in the stress signal. Trend modeling module: configured to receive the baseline drift component and associated environmental variables, construct a dual-path deep reinforcement learning ensemble model, wherein the first path uses a long short-term memory network to encode environmental temporal features, the second path establishes a drift state transition equation, and finally outputs the drift prediction value and confidence level through a collaborative decision-maker; Compensation and correction module: configured to, based on the drift prediction value, introduce the theoretical quantity of structural thermal expansion as a physical constraint and drive the physical-guided adversarial generative network to generate drift compensation amount; Health prediction module: configured to fuse the drift compensation amount with the original stress data, perform the remaining life integral equation calculation with time-varying gain factor, and output the bridge structure health status and remaining life prediction results.

2. The stress monitoring data correction system for continuous beam bridges under temperature influence according to claim 1, characterized in that: The specific operation of the adaptive sliding window to dynamically capture signal segments in the drift sensing module is as follows: First, the phase space of the real-time stress data stream is reconstructed. The optimal time delay parameter is determined by calculating the minimum value of the mutual information function, and a multi-dimensional phase space trajectory matrix is ​​constructed with an embedding dimension of 3. Second, the maximum Lyapunov exponent of the phase space trajectory is calculated. Then, the window length is dynamically calculated based on the initial disturbance modulus threshold and the preset allowable error, combined with the reciprocal relationship of the maximum Lyapunov exponent. Finally, the environmental temperature and humidity data are synchronously locked into the window in a timestamp-aligned manner, and a fixed-length signal segment and associated environmental variables are output.

3. The stress monitoring data correction system for continuous beam bridges under temperature influence according to claim 2, characterized in that: The specific process of separating the baseline drift component and the true stress component in the stress signal is as follows: A1. Calculate the local singularity index for the stress data within the window and generate the singularity weight distribution function; A2. The process of generating upper and lower envelopes based on singularity-weighted adjustment of empirical mode decomposition: the Hermite interpolation method is used to construct the weighted envelope, and the intrinsic mode function component set and residual components are obtained by iterative decomposition. A3. Perform discrete wavelet transform based on db8 wavelet basis functions on the residual components to decompose them into multi-level wavelet coefficients, and calculate the wavelet entropy of each level. A4. Establish an entropy-temperature coupling criterion based on the relationship between the rate of change of ambient temperature and the partial derivative of wavelet entropy. When the absolute value of the partial derivative exceeds a predetermined threshold, mark the corresponding segment of the current residual component as the baseline drift component. A5. Separate the baseline drift component from the original residual components and reorganize the intrinsic mode function components to output the true stress components.

4. The stress monitoring data correction system for continuous beam bridges under temperature influence according to claim 3, characterized in that: In the trend modeling module, the specific operations for encoding temporal features of the Long Short-Term Memory network are as follows: First, the system receives a data stream of associated environmental variables from the drift sensing module, which includes synchronous time-series records of temperature and humidity parameters. Second, it constructs a bidirectional long short-term memory network architecture with a fixed number of hidden layer units of 64. Then, it sets the environmental analysis time window to 1200 seconds and performs a bidirectional time-dimensional scan of the temperature and humidity data within this window to capture the lagging and leading effects of historical and future environmental states on drift, respectively. Finally, it outputs an environmental feature vector that fuses the bidirectional features.

5. The stress monitoring data correction system for continuous beam bridges under temperature influence according to claim 4, characterized in that: The process of constructing the drift state transition equation is as follows: First, the current value of the baseline drift component at the previous moment is obtained; second, the environmental feature vector is extracted and input into the temperature change rate response gating function to generate environmental factor weighting coefficients; then, the hardware aging effect is converted by a logarithmic scaling factor, based on the cumulative power-on time of the sensor; then, the three sets of parameters—the baseline drift component, the environmental factor weighting coefficients, and the hardware aging effect—are substituted into the state transition equation, which uses the modified linear unit as the nonlinear activation function. Finally, the predicted drift state is output.

6. The stress monitoring data correction system for continuous beam bridges under temperature influence according to claim 5, characterized in that: The operation of the collaborative decision-maker to output drift prediction values ​​and confidence levels is as follows: First, the environmental feature vector and the drift state prediction are merged to construct the decision state space. Second, the action space of the compensation adjustment magnitude is defined, with its upper and lower bounds symmetrically distributed. Then, the state action value function is calculated through a deep Q-network, with the hidden layer of this function adopting a two-layer fully connected structure. Subsequently, based on the absolute residual between the drift state prediction and the actual baseline drift component, and combined with the conditional mutual information entropy of the environmental feature vector and the drift state prediction, a normalized confidence index is generated. Finally, the compensation adjustment magnitude is output as the correction amount for the drift prediction, and the confidence index is used as the reliability indicator of health management decision.

7. The stress monitoring data correction system for continuous beam bridges under temperature influence according to claim 6, characterized in that: The specific operations of introducing the theoretical quantity of structural thermal expansion as a physical constraint in the compensation and correction module include: First, the characteristic length distribution of the temperature-sensitive region is analyzed using a bridge finite element model. Second, the real-time temperature difference value and the coefficient of thermal expansion of the material are obtained. Then, the median characteristic length, the temperature difference value, and the coefficient of thermal expansion are multiplied to generate the theoretical quantity of structural thermal expansion. Finally, this theoretical quantity is output to the input of the physics-guided adversarial generative network.

8. The stress monitoring data correction system for continuous beam bridges under temperature influence according to claim 7, characterized in that: The operation process of the driving physics-guided generative adversarial network is as follows: First, a dual-channel temporal convolution generator is constructed, with the first channel inputting the drift prediction value and the second channel inputting the theoretical quantity of structural thermal expansion. Second, the discriminator incorporates the sensor aging physical equation constraint. Then, the generator generates an initial compensation amount based on the dual-channel input. Finally, a confidence-weighted fusion mechanism is used to correct the initial compensation amount: when the confidence level is higher than a threshold, the generated result is directly output; when it is lower than the threshold, it switches to a conservative correction mode based on baseline drift.

9. A stress monitoring data correction system for continuous beam bridges under temperature influence according to claim 8, characterized in that: In the health prediction module, the specific operation of fusing the drift compensation amount with the original stress data is as follows: First, the drift compensation amount and the original stress monitoring data stream are received; second, a drift elimination operation is performed, subtracting the drift compensation amount from the original stress monitoring data to generate the drift-corrected true stress value; then, the gradient field is calculated on the true stress value to obtain the multi-axis stress coupling gradient norm; at the same time, the environmental vibration energy spectral density parameter is extracted; finally, combined with the confidence parameter from the trend modeling module, a time-varying gain factor is generated through hyperbolic tangent function mapping.

10. A stress monitoring data correction system for continuous beam bridges under temperature influence according to claim 9, characterized in that: The execution process of the remaining lifetime integral equation with time-varying gain factor is as follows: First, the theoretical failure time point and the material failure stress threshold are set; second, a time-varying decay factor with exponential decay characteristics is constructed, which gradually increases as time approaches the failure point. Then, the minimum value of the actual stress during the service life is used as the criterion for the fatigue strength inflection point; Finally, a dual-integral kernel function is established. The dynamic damage rate, attenuation factor and fatigue strength margin after gain modulation are dynamically fused in the inner integral layer, and the time-normalization operation is performed in the outer integral layer to output the remaining life prediction value and the three-level health status classification label.

Images