Plateau concrete structure health digital twin system and life prediction method based on multi-element heterogeneous sensing and virtual-real mapping

Abstract

The application provides a highland concrete structure health digital twin system and life prediction method based on multi-element heterogeneous sensing and virtual-real mapping, and belongs to the technical field of concrete structure health monitoring.The application comprises a highland environment structure coupling sensing layer, a 5G edge intelligent transmission layer, a cloud collaborative data processing layer and a physical-data double-driven digital twin layer.The application first introduces an air pressure and resistivity gradient sensor to quantize the multi-field coupling effect of meteorology, freeze-thaw and mechanics;adopts a 5G architecture to realize edge multi-protocol analysis and weak network transmission resistance;finally, based on a physical guided dynamic Bayesian neural network, a deep neural network and a freeze-thaw damage mechanics model are nested, the confidence of the prediction result is quantized through a heterogeneous variance uncertainty output layer, and a virtual-real mapping closed loop of the structure metabolic state is formed, thereby providing a digital twin service covering intelligent monitoring and life prediction for the highland concrete, and breaking through the technical bottleneck of environment-mechanics multi-factor decoupling in traditional monitoring.

Description

Technical Field

[0001] This invention relates to a digital twin system for the health of plateau concrete structures and a method for predicting their lifespan based on multi-element heterogeneous sensing and virtual-real mapping, belonging to the field of concrete structure health monitoring technology. Background Technology

[0002] The plateau climate is characterized by strong ultraviolet radiation, dryness, frequent freeze-thaw cycles, and low air pressure. With the large-scale construction of water conservancy and hydropower projects in my country's plateau regions, some hydraulic concrete structures are exposed to the combined effects of extreme plateau climate and complex stresses. These factors have drastically accelerated the deterioration of concrete material properties, with freeze-thaw damage being a key factor threatening the safety of plateau hydraulic structures. Statistics show that the actual service life of plateau hydraulic concrete structures is shortened by an average of 30%-40% compared to their design life, seriously threatening the long-term, safe operation of water conservancy facilities.

[0003] At present, the performance monitoring of concrete building structures still mainly relies on traditional manual inspection and single-point sensor monitoring methods (such as rebound strength testing and handheld resistivity testing). The technical bottlenecks that need to be addressed in these technologies are summarized as follows: (1) Single-dimensional monitoring systems are difficult to characterize the multi-factor synergistic damage in plateau environments, and existing monitoring systems do not include low air pressure and resistivity gradient in the core monitoring indicators, which cannot truly reflect the synergistic damage mechanism of "meteorology-freeze-thaw-mechanics" in plateau environments; (2) Traditional data transmission units are only responsible for transparent transmission and cannot cope with the weak network and high latency environment in plateaus. Uploading massive amounts of raw data directly to the cloud can easily cause packet loss and congestion, and real-time cleaning cannot be performed at the edge; (3) The life prediction model of plateau concrete structures has poor universality. Existing deep learning models such as Convolutional Neural Network (CNN) and LSTM rely entirely on empirical formulas or simplified physical models obtained from massive sample training and accelerated experiments. These models generally have the disadvantages of fixed parameters and poor adaptability, and cannot dynamically evolve with the changes in the actual state of plateau concrete structures, making it difficult to ensure the integrity and timeliness of monitoring data, which directly affects the accuracy of subsequent damage assessment.

[0004] Existing monitoring technologies mostly focus on chloride ion corrosion or conventional temperature and humidity monitoring. There is a relative lack of technical literature on characterizing the multi-field coupling mechanisms of the plateau environment, and even more so a severe shortage of technologies that can effectively integrate monitoring systems with intelligent large-scale models. The survey found that...

[0005] The invention patent application with patent application number CN202511169464.X discloses a real-time monitoring system and method for the health of concrete structures based on multi-source sensor fusion. Although the proposed monitoring system based on multi-source sensor fusion involves environmental correction, it only considers the conventional effects of temperature and humidity on bearing capacity, ignoring the influence of the unique "low air pressure" of the plateau on the migration of moisture in micropores, as well as the unsteady evolution of resistivity gradient caused by "large temperature difference".

[0006] Patent application CN202511044192.0 discloses a method and device for dynamic probabilistic prediction of dam life. While the proposed dam life prediction method based on LSTM and dynamic Bayesian networks utilizes a time-series model, it is essentially a "purely data-driven" statistical method, heavily reliant on massive amounts of high-quality samples. In high-altitude regions, due to the difficulty in data acquisition and sparse samples, purely data-driven models are prone to overfitting and cannot incorporate the physical and mechanical constitutive relationships of concrete freeze-thaw damage, resulting in a lack of physical interpretability in the prediction results.

[0007] In summary, considering the unique deterioration characteristics of concrete structures in high-altitude environments, it is urgent to break through the bottleneck of quantitative characterization of multi-field coupled damage by integrating multi-dimensional parameter synchronous perception, highly reliable data transmission, and intelligent algorithms. This will provide technical support for the full life cycle management of water conservancy projects in high-altitude areas and lay the foundation for the deep integration of artificial intelligence and civil engineering. Summary of the Invention

[0008] The purpose of this invention is to address the shortcomings and deficiencies of existing high-altitude environment service performance monitoring technologies. It proposes a digital twin system for the health of high-altitude concrete structures and a life prediction method based on multi-element heterogeneous sensing and virtual-real mapping. For the first time, it introduces a barometric pressure sensor and resistivity gradient array into the monitoring system, quantifying the impact of low air pressure and large temperature differences on the evolution of concrete microstructures. This changes the traditional data transmission process, moving beyond mere transmission to address the technical challenges of data "cleaning, fusion, and caching" at the edge, thus solving the communication reliability problem in unattended high-altitude areas. Furthermore, compared to traditional algorithm models, this invention employs a physics-driven dynamic Bayesian model, embedding the damage mechanics constitutive equation as "physical knowledge" into the neural network. This allows the model to follow physical laws even with small sample sizes and accurately predict results through a heteroscedastic output layer. This enables the understanding of the health status of concrete structures under extreme environments, providing real-time, dynamic, and intelligent analysis of the remaining service life of the structure and offering a scientific basis for structural service performance assessment, protection, and repair.

[0009] The objective of this invention is achieved through the following technical solution:

[0010] A health digital twin system for plateau concrete structures based on multi-dimensional heterogeneous sensing and virtual-real mapping includes: a plateau environment structure coupling sensing layer, a 5G edge intelligent transmission layer, a cloud collaborative data processing layer, and a physical-data dual-drive digital twin layer; the signal output end of the plateau environment structure coupling sensing layer is connected to the signal input end of the 5G edge intelligent transmission layer, the signal output end of the 5G edge intelligent transmission layer is connected to the signal input end of the cloud collaborative data processing layer, and the signal output end of the cloud collaborative data processing layer is connected to the signal input end of the physical-data dual-drive digital twin layer;

[0011] The plateau environment structure coupling sensing layer is used to capture the external environmental parameters, internal physical fields and damage indicators of concrete structures in real time, and to build a synchronous monitoring system for environment-damage coupling.

[0012] The 5G edge intelligent transmission layer is used to receive data captured by the plateau environment structure coupling perception layer, and to store and preprocess it.

[0013] The cloud collaborative data processing layer is used to receive preprocessed data from edge nodes of the 5G edge intelligent transmission layer, perform cloud data processing, and generate a standardized time-series dataset of structural health status.

[0014] The physical-data dual-driven digital twin layer is used to capture the random noise and sparsity characteristics of the time-series dataset of structural health status generated by the cloud collaborative data processing layer, and simultaneously output the predicted mean and predicted variance of the remaining life of the concrete structure; and calculate the failure probability of the structure under a specific service life.

[0015] Preferably, the plateau environment structure coupling sensing layer includes: an external environment monitoring unit and an internal physical field and damage monitoring unit;

[0016] The external environment monitoring unit uses air temperature and humidity sensors, industrial-grade high-precision air pressure sensors, and intelligent monitoring cameras to capture the dynamic fluctuation characteristics of temperature, humidity, and air pressure in the plateau environment in real time, as well as on-site anti-theft early warning security video data.

[0017] The internal physical field and damage monitoring unit embeds temperature, humidity and resistivity sensors at different depths inside the concrete structure to invert freeze-thaw damage fronts; welds steel stress sensors and concrete strain sensors in key stress areas; and installs displacement sensors at expansion joints and surface cracks to collect data on temperature, humidity, resistivity, strain, stress and crack propagation.

[0018] Preferably, the 5G edge intelligent transmission layer includes: an adaptive signal gain amplifier, a multi-protocol acquisition chip, an energy management unit, and an edge computing unit; the signal input terminal of the multi-protocol acquisition chip is connected to the signal output terminals of the external environment monitoring unit and the internal physical field and damage monitoring unit of the plateau environment structure coupling perception layer, respectively; the signal output terminal of the multi-protocol acquisition chip is connected to the signal input terminal of the edge computing unit; the signal output terminal of the edge computing unit is connected to the signal input terminal of the adaptive signal gain amplifier; and the energy management unit is connected to the adaptive signal gain amplifier, the multi-protocol acquisition chip, and the edge computing unit, respectively.

[0019] An adaptive signal gain amplifier is used to dynamically adjust the transmit power and signal gain multiple, automatically compensate for link loss in areas with weak signal coverage, and ensure the stability of the data upload link.

[0020] A multi-protocol acquisition chip is used to parse the connections of various sensors in the high-altitude environment structure coupling sensing layer with different interface protocols, so as to realize multi-channel synchronous acquisition.

[0021] The energy management unit is used to monitor the external power supply status and the remaining power of the system, and to execute dynamic power scheduling strategies based on energy awareness.

[0022] Edge computing units are used to perform preprocessing and temporary storage of local data, and to remove random noise and anomalous jump values ​​collected by sensors.

[0023] Preferably, the multi-protocol acquisition chip parses different interface protocols, including bus type, vibrating wire type, and resistor type.

[0024] Preferably, the cloud collaborative data processing layer is built on a 5G edge computing architecture, and its algorithm implementation relies on an intelligent data analysis platform to receive preprocessed data from edge nodes and perform cloud data processing. The cloud data processing includes: performing multi-source heterogeneous data loading and spatiotemporal alignment, deep data cleaning and preprocessing, multi-scale feature engineering extraction, isolated forest anomaly detection, dimensionality reduction mapping and state clustering, and generating a standardized structural health state time series dataset.

[0025] Preferably, the physical-data dual-driven digital twin layer adopts a dynamic Bayesian neural network model, which includes: an input layer, a Bayesian LSTM layer, a heteroscedastic uncertainty output layer, and a post-processing module; the signal output terminal of the input layer is connected to the signal input terminal of the Bayesian LSTM layer, the signal output terminal of the Bayesian LSTM layer is connected to the signal input terminal of the heteroscedastic uncertainty output layer, and the signal output terminal of the heteroscedastic uncertainty output layer is connected to the signal input terminal of the post-processing module.

[0026] The input layer is used to map the standardized multi-source heterogeneous time series data into high-dimensional feature vectors and pass them to the Bayesian LSTM layer.

[0027] Bayesian LSTM layers are used for feature extraction and model uncertainty capture, capturing random noise and sparse features of the monitoring data;

[0028] Heteroscedastic uncertainty output layer, used to predict the probability distribution of the output concrete service status and quantify data noise;

[0029] The post-processing module, based on the output probability distribution information, is used to calculate the time-varying reliability index of the structure and introduces the spatiotemporal conversion coefficient.

[0030] Preferably, the time-varying reliability index of the computational structure adopts the first-order second-moment method.

[0031] A life prediction method for a health digital twin system of plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping includes the following steps:

[0032] Step 1: Deploy a plateau environment structure coupling sensing layer that includes external environmental monitoring units and internal physical field and damage monitoring units. Real-time synchronous acquisition of multi-dimensional and multi-physical field data such as temperature, humidity, air pressure, resistivity, strain, stress and crack propagation to form the original monitoring dataset of environmental and structural damage coupling.

[0033] Step 2: Based on the 5G edge computing architecture, the original monitoring dataset obtained in Step 1 is subjected to protocol parsing, edge-side data processing, and reliable transmission. Through data preprocessing, a standardized time-series dataset for deep learning model analysis is constructed. The standardized time-series dataset is input into the dynamic Bayesian neural network model, and the time-series features are captured through the Bayesian LSTM layer. Combined with heteroscedastic uncertainty, the remaining lifetime prediction results with confidence intervals are output.

[0034] Step 3: In the physical-data dual-driven digital twin layer, the damage characteristics of the concrete are intelligently identified and extracted based on the remaining service life prediction results in Step 2, and the dynamic evaluation level of the service status is output; combined with reliability theory, a service life prediction model that integrates material properties, service status and environmental coupling effects is constructed to calculate the remaining service life of the concrete structure.

[0035] Preferably, the specific steps for calculating the remaining service life of the concrete structure in step three are as follows:

[0036] First, the failure threshold is determined using the reliability index output by the Bayesian neural network model. Second, a spatiotemporal conversion coefficient is introduced to establish a mapping equation from the number of accelerated freeze-thaw cycles in the laboratory to the actual service life in engineering projects, thereby calculating the remaining service life, as shown in the following formula:

[0037]

[0038] in, To predict service life; To predict the cumulative number of cycles required for the model to reach the failure threshold, The average number of freeze-thaw cycles per year at the project site; This formula represents the equivalent damage coefficient between the laboratory and the field; it converts the number of rapid freeze-thaw cycles in the laboratory into actual engineering service life, providing a practical tool for engineering applications.

[0039] In determining In the process, this model also incorporates reliability theory, calculating reliability indices. The formula for determining whether a structure has entered a failure state is as follows:

[0040] The reliability index is calculated using the first-order second-moment method:

[0041]

[0042] in: To predict the mean, To predict the standard deviation, This is the failure threshold; when When the reliability is > 99.9%, the structure is safe; when At this time, the reliability is between 90% and 99.9%, requiring regular monitoring; when When the time comes, an early warning is issued, indicating the end of life.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] This invention constructs a multi-dimensional heterogeneous sensing and virtual-real mapping digital twin system for the health of plateau concrete structures. Through a multi-dimensional environment-structure digital twin (i.e., a plateau environment-structure coupled sensing layer) and a physical-data fusion driving mechanism (i.e., a 5G edge intelligent transmission layer and a physical-data dual-drive digital twin layer), it achieves intelligent management and control of the entire lifecycle of plateau concrete structures. The multi-dimensional environment-structure digital twin refers to the introduction of multi-dimensional mapping relationships between air pressure, drying cracking, and resistivity gradients to achieve integrated perception of the harsh external environment and the internal structural state. The physical-data fusion driving mechanism utilizes an upgraded 5G edge intelligent transmission layer to perform protocol cleaning and signal enhancement at the edge, and embeds a dynamic Bayesian network into the continuous damage mechanics equations to achieve probabilistic performance extrapolation from the hydration heat stage to long-term durability degradation. This not only solves the problem of "storage, management, and transmission" of massive multimodal data in the weak network environment of plateau areas, but also transforms monitoring data into engineering-universal reliability indicators, forming a closed-loop service covering intelligent diagnosis, life prediction, and maintenance decision-making.

[0045] This invention introduces pressure and resistivity gradient sensors for the first time to quantify the multi-field coupling effect of "meteorology-freeze-thaw-mechanics," simultaneously achieving real-time monitoring of the internal strain field of concrete, stress waves in steel reinforcement, and microcrack propagation. It also employs a 5G architecture to achieve multi-protocol parsing at the edge and transmission against weak networks. Finally, based on a physics-guided dynamic Bayesian neural network, a deep neural network is nested with a freeze-thaw damage mechanics model (implemented through the concrete damage mechanics constitutive equation in the algorithm). The confidence level of the prediction results is quantified through a heteroscedastic uncertainty output layer, forming a closed loop of virtual-real mapping of the structural metabolic state. This provides a digital twin service for high-altitude concrete, encompassing intelligent monitoring and life prediction, overcoming the technical bottleneck of decoupling environmental and mechanical factors in traditional monitoring. Furthermore, the virtual-real mapping mechanism constructed in this invention effectively overcomes the technical challenges of poor model generalization ability and lack of physical interpretability caused by sparse samples and high noise in extreme high-altitude environments. Through a deep mapping of mechanism and behavior driven by both physics and data, the confidence level of the prediction results is accurately quantified, achieving a leap from a simple "geometric twin" to a "physical behavioral twin." Attached Figure Description

[0046] Figure 1 This is a structural schematic diagram of the digital twin system for the health of plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping according to the present invention; wherein:

[0047] Figure 1 (a) is a schematic diagram of the structure of the plateau environment structure coupling sensing layer;

[0048] Figure 1 (b) is a schematic diagram of the 5G edge intelligent transmission layer structure;

[0049] Figure 1 (c) is a schematic diagram of the cloud collaborative data processing layer structure;

[0050] Figure 1 (d) is a schematic diagram of the physical-data dual-driven digital twin layer structure.

[0051] Figure 2 This is a structural block diagram of the digital twin system for the health of plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping, according to the present invention.

[0052] Figure 3 The graph shows the monitoring curve of the external environment of the concrete obtained in Example 1.

[0053] Figure 4 The graph shows the internal service status monitoring curve of the concrete obtained in Example 1. Detailed Implementation

[0054] The present invention will be further described in detail below with reference to the accompanying drawings: This embodiment is implemented under the premise of the technical solution of the present invention, and detailed implementation methods are given, but the protection scope of the present invention is not limited to the following embodiments.

[0055] like Figure 1 and Figure 2 As shown, the digital twin system for the health of plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping involved in this embodiment includes:

[0056] A digital twin system for the health of plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping includes: a plateau environment structure-coupled sensing layer, a 5G edge intelligent transmission layer, a cloud-collaborative data processing layer, and a physical-data dual-driven digital twin layer. The sensing layer is connected to the transmission layer, and data from the processing layer is imported into the digital twin layer for data prediction through an intelligent algorithm model (i.e., a dynamic Bayesian neural network model). Figure 1 As shown, through deep integration of edge computing and cloud collaboration, each level has achieved end-to-end digital twin monitoring, from environmental-structure coupled field capture to physically guided intelligent prediction.

[0057] Among them, the plateau environment structure coupling sensing layer, for the first time, integrates the atmospheric pressure-freeze-thaw damage environmental degradation characteristics of the plateau environment on the basis of traditional monitoring, realizing the transformation from "general-purpose monitoring" to "plateau-specific physical-guided monitoring". It is used to capture the external environmental parameters, internal physical fields and damage indicators of concrete structures in real time, and to build a synchronous monitoring system of environment-damage coupling. The external environment monitoring unit integrates a high-precision atmospheric pressure sensor (300-1100hPa) on the basis of traditional temperature and humidity environmental field monitoring. The internal physical field and damage monitoring unit integrates a resistivity sensor (0.001-1000kΩ.cm) by combining the special degradation effect of the plateau environment. At the same time, it is combined with steel reinforcement stress (-260-260Mpa), concrete micro-strain (-1500-1500μℇ) and crack displacement sensor (0-50mm) to build a meteorological-physical-mechanical multi-field coupled monitoring system. It also includes intelligent monitoring cameras to obtain on-site anti-theft early warning security video data, such as Figure 1 As shown in (a);

[0058] The 5G edge intelligent transmission layer (frequency 700-3600Hz, 16-32 channels) breaks through the limitations of traditional data transmission units' single transparent transmission and is specifically designed for high-altitude, cold, strong interference, and weak signal environments. This layer incorporates an adaptive signal gain amplifier and a multi-protocol acquisition chip, featuring an environmentally adaptive anti-interference mechanism. It executes data interruption resumption and local caching strategies at the edge and has a built-in power management unit to ensure long-term stable operation. Figure 1 As shown in (b);

[0059] like Figure 1 (c) shows the cloud collaborative data processing layer, which is built on the 5G edge computing architecture. Its algorithm implementation relies on an intelligent data analysis platform (such as a data processing model built on Python). It is used to receive preprocessed data from edge nodes, perform multi-source heterogeneous data loading and spatiotemporal alignment, deep data cleaning and preprocessing, multi-scale feature engineering extraction, isolated forest anomaly detection, dimensionality reduction mapping and state clustering, and generate a standardized structural health state time series dataset.

[0060] A physical-data dual-driven digital twin layer focuses on key damage characteristics in high-altitude environments. It embeds existing concrete damage mechanics constitutive equations as physical constraints into the neural network loss function. The aim is to utilize physical laws to limit the neural network's search space, thus providing predictions consistent with physical reality even when faced with "data sparsity" or "high random noise" (such as high-altitude environment monitoring data). Then, a Bayesian Long Short-Term Memory (LSTM) layer combined with variational inference techniques is used to capture the random noise and sparsity characteristics of the monitoring data. A designed heteroscedastic uncertainty output layer simultaneously outputs the predicted mean and variance of the remaining service life. The predicted mean and variance are statistical parameters describing the probability distribution of service status. Combined with reliability methods, the failure probability of the structure at a specific service life is calculated, such as... Figure 1 As shown in (d).

[0061] A life prediction method for a health digital twin system of plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping includes the following steps:

[0062] Step 1: Deploy a plateau environment structure coupling sensing layer containing external environment monitoring units and internal state sensing units to collect multi-dimensional and multi-physical field data such as temperature, humidity, air pressure, resistivity, strain, stress and crack propagation in real time and synchronously, forming a raw monitoring dataset of environmental and structural damage coupling.

[0063] Step 2: Based on the 5G edge computing architecture, perform protocol parsing, edge-side data processing, and reliable transmission on the raw data. Through data preprocessing, construct a standardized time-series dataset for deep learning model analysis. Input the standardized time-series dataset into the dynamic Bayesian neural network model, capture time-series features through the Bayesian LSTM layer, and output the remaining lifetime prediction results with confidence intervals by combining heteroscedastic uncertainty.

[0064] Step 3: Based on the training results (i.e., the inference results output by the dynamic Bayesian neural network model in Step 2 after processing the current standardized time-series dataset), intelligently identify and extract the damage characteristics of the concrete, and output the dynamic evaluation level of the service status. Combining reliability theory, construct a life prediction model that integrates material properties, service status, and environmental coupling effects to calculate the remaining service life of the concrete structure.

[0065] The specific steps for calculating the remaining service life of the concrete structure in step three are as follows:

[0066] First, the failure threshold is determined using the reliability index output by the Bayesian neural network model. Second, a spatiotemporal conversion coefficient is introduced to establish a mapping equation from "laboratory accelerated freeze-thaw cycle count" to "actual engineering service life," thereby calculating the remaining service life, as shown in the following formula:

[0067]

[0068] in: Predicted service life (years); The model predicts the cumulative number of cycles required to reach the failure threshold (RDME < 60%). The average number of freeze-thaw cycles per year at the project site (based on data from a certain observation station in Qinghai, taken as 100 times / year); The equivalent damage coefficients for laboratory and field use are based on the "Standard for Testing and Evaluation of Concrete Durability". This formula converts the number of rapid freeze-thaw cycles in the laboratory into the actual service life of engineering projects, providing a practical tool for engineering applications.

[0069] In determining In the process, this model also incorporates reliability theory, calculating reliability indices. The formula for determining whether a structure has entered a failure state is as follows:

[0070] The reliability index is calculated using the First Order Reliability Method (FORM):

[0071]

[0072] in: To predict the mean, To predict the standard deviation, This is the failure threshold (60% in this study). When... When the reliability is > 99.9%, the structure is safe; when At this time, the reliability is between 90% and 99.9%, requiring regular monitoring; when When the time comes, an early warning is issued, indicating the end of life.

[0073] The following five examples demonstrate the superiority of the multi-element heterogeneous sensing and virtual-real mapping digital twin system for the health of plateau concrete structures constructed in this invention:

[0074] Example 1

[0075] The plateau environment structure coupling sensing layer includes: an external environment monitoring unit and an internal physical field and damage monitoring unit;

[0076] The external environment monitoring unit employs air temperature and humidity sensors (temperature: -40-125℃, humidity: 0-100%RH, accuracy: ±0.3℃, ±2%RH) and industrial-grade high-precision barometric pressure sensors (measurement pressure range: 10-1100 hPa, accuracy: ±0.3hPa). One set is deployed every 50 meters along the dam to capture the dynamic fluctuations in temperature, humidity, and air pressure in the plateau environment in real time.

[0077] The internal physical field and damage monitoring unit embeds temperature, humidity, and resistivity sensors at different depths within the concrete to invert freeze-thaw damage fronts. Reinforcing steel stress sensors and concrete strain sensors are welded into key stress areas, while displacement sensors are installed at expansion joints and surface cracks. All sensor data is acquired synchronously at high frequency (10Hz) to ensure strict temporal and spatial alignment between environmental loads (air pressure / temperature difference) and structural responses (stress / cracks). The monitoring output results are as follows: Figure 3 , 4 As shown.

[0078] Example 2

[0079] The 5G edge intelligent transmission layer includes: an adaptive signal gain amplifier, a multi-protocol acquisition chip, an energy management unit, and an edge computing unit. The multi-protocol acquisition chip serves as the physical interface layer, directly connecting to the sensors in the perception layer. After acquisition, the data enters the edge computing unit for local preprocessing and caching. The processed data stream is modulated by the adaptive signal gain amplifier and then transmitted through the 5G radio frequency antenna. The energy management unit connects the above three modules in parallel and is responsible for the power supply scheduling and sleep / wake-up of the entire system.

[0080] The 5G edge intelligent transmission layer, with the gateway installed in the dam control room, integrates with sensors via a multi-protocol acquisition chip, achieving a wireless transmission range of 5 kilometers. The cloud platform's large screen uses data visualization technology to intuitively display complex data information in graphical and chart formats, and can trigger automatic warnings through customizable threshold settings. Furthermore, for the high-altitude, weak network environment, the gateway integrates an adaptive signal gain amplifier, an embedded edge computing unit, and a 1TB industrial-grade solid-state drive, implementing a breakpoint resume strategy. Simultaneously, the built-in power management unit dynamically adjusts the acquisition frequency based on the photovoltaic power supply status (sleep power consumption <0.5W), ensuring long-term stable operation of the system under conditions without mains power.

[0081] Example 3

[0082] The cloud-collaborative data processing layer's algorithm implementation is based on a data processing model built in Python, receiving preprocessed data from 5G edge nodes. It receives environmental and sensor data from the wireless module, first performing spatiotemporal alignment and multi-source data fusion to generate a structured time-series dataset. Then, it calculates statistical features, time-domain features, and frequency-domain features, completing feature engineering extraction. Finally, it normalizes the features to eliminate the influence of dimensions, standardizing the data, and combines isolated forest and principal component analysis for anomaly detection and dimensionality reduction analysis.

[0083] Example 4

[0084] The core algorithm model of the physical-data dual-driven digital twin engine layer employs a dynamic Bayesian neural network. The model structure includes: an input layer (receiving standardized time-series data), a Bayesian LSTM layer (3 LSTM layers, 128 neurons per layer), and a heteroscedastic uncertainty output layer (outputting the probability distribution of the concrete's service status). The dataset is input into the dynamic Bayesian neural network model, and the Bayesian LSTM layer learns the long-term dependence of freeze-thaw damage. The output layer provides probabilistic predictions (e.g., 95% confidence intervals).

[0085] The input layer serves as the data entry point, receiving standardized time-series monitoring data (such as strain and environmental parameters). It then maps the standardized multi-source heterogeneous time-series data into high-dimensional feature vectors and passes them to the Bayesian LSTM layer. The Bayesian LSTM layer takes the input features and uses variational inference mechanisms to capture the long-term and short-term time-series dependencies and the distribution characteristics of model parameters in the data, generating a hidden state vector containing randomness information. The heteroscedastic uncertainty output layer is tightly coupled to the LSTM layer, performing parallel analysis on the hidden state vector and synchronously outputting the mean and variance of the predicted values. This achieves a closed-loop implementation within a single network, from raw data input to remaining lifetime prediction output with confidence intervals.

[0086] The concrete damage mechanics constitutive equation serves as a physical constraint, and the physical constraint mechanism acts on the training and optimization stage of the overall network. Specifically, the concrete damage mechanics constitutive equation is constructed as a physical residual term and superimposed on the loss function.

[0087] Example 5

[0088] The post-processing module primarily uses the probability distribution information output by the Bayesian network and employs the first second-order moment method (FORM) to calculate the time-varying reliability index of the structure, seamlessly transforming the prediction results of the underlying algorithm into the reliability concept commonly used in structural safety assessment. Simultaneously, a spatiotemporal transformation coefficient is introduced to achieve an equivalent mapping from laboratory accelerated freeze-thaw cycles to the natural service life of actual engineering projects.

[0089] In dynamic Bayesian neural network models, prior probability distributions can be assigned to the weights of the input layer. After model training (variable inference), the variance of the posterior distribution of the weights of features that contribute little to the prediction result will approach 0; while the variance of the weights of features that contribute much to the prediction result will be larger. By extracting the statistical parameters of these posterior distributions, the probability weights of the features can be quantified.

[0090] Based on the probability weights of damage features obtained from training a dynamic Bayesian neural network model, key damage features affecting concrete performance degradation (in order of stress-strain and resistivity attenuation) are selected. The weights of each feature are determined using the analytic hierarchy process (AHP). In accordance with GB / T 50476-2019, service status is divided into five levels (Level 1: Excellent, damage <10%; Level 2: Good, damage 10%-25%; Level 3: Moderate, damage 25%-40%; Level 4: Poor, damage 40%-60%; Level 5: Critical, damage >60%). The quantitative values ​​of damage features output by the dynamic Bayesian neural network model are automatically matched to the corresponding evaluation level. Combining reliability theory, a life prediction model integrating material properties, service status, and environmental coupling effects is constructed to calculate the remaining service life.

[0091] The method for calculating the quantified value of the damage characteristics is as follows:

[0092] To standardize the evaluation criteria, the metrics output by the dynamic Bayesian neural network model at each prediction time step are transformed into a dimensionless structural damage level. :

[0093]

[0094] The model calculates in real time at each prediction time step. Indicators, when the model predicts Reduced to 60% (i.e. When the index reaches a certain threshold, the concrete structure is considered to have reached its durability limit state, i.e., the end of its service life. If the index reaches a certain threshold, the structure is considered to have entered a "failure state." Based on... The evolution trajectory of the value, according to the "Standard for Durability Design of Concrete Structures" (GB / T50476-2019), divides the service condition of concrete into 5 levels, as shown in the table below:

[0095] Table 1. Classification of Damage Levels in Service Condition

[0096]

[0097] The calculation of structural failure probability is implemented in the post-processing module of the physical-data dual-driven digital twin layer. The calculation method adopts the first-order second-moment method based on reliability theory. Although this calculation method itself belongs to the prior art, the significant feature of this invention is the construction of a three-level evaluation system of "qualitative classification - quantitative probability - spatiotemporal conversion": First, unlike traditional static statistics, this system directly uses the posterior mean and heteroscedastic uncertainty output by a dynamic Bayesian neural network (DBNN) as dynamic inputs to construct the function; Second, a spatiotemporal conversion coefficient is introduced to establish a mapping equation, converting the number of accelerated freeze-thaw cycles in the laboratory into the predicted service life of actual engineering projects, thereby calculating the failure probability under a specific service life, realizing the leap from single physical quantity prediction to comprehensive engineering risk assessment.

[0098] In summary, the innovation of this invention compared to existing monitoring systems lies in:

[0099] (1) A comprehensive monitoring system that “takes into account both internal and external factors and couples the environment and structure”: It breaks through the limitations of traditional monitoring that focuses on structural response (such as stress and strain), and realizes the capture of the entire chain of data from “environmental driving force” to “structural damage”, fundamentally changing the situation of traditional methods with single monitoring dimensions and disconnected causal analysis.

[0100] (2) End-to-end automation from “data acquisition” to “intelligent decision-making”: Breaking through the shortcomings of traditional assessment methods that isolate damage identification or life prediction, this invention constructs a progressive analysis framework that goes from wireless data acquisition to intelligent processing and analysis, realizing the interdisciplinary integration of artificial intelligence and structural health monitoring, and greatly improving the reliability of prediction results.

[0101] By comparing Example 1 with traditional monitoring methods, this invention not only achieves a synchronous acquisition accuracy of over 98% for internal and external service parameters of concrete in high-altitude areas, and a technological breakthrough in intelligent identification of structural damage with an accuracy rate of 97.8%, but also innovatively adopts DBNN to provide probabilistic predictions with a 95% confidence interval, quantifying uncertainty and supporting risk perception decision-making, providing a new data-driven paradigm for the study of the life-cycle performance evolution of concrete structures in special high-altitude environments.

[0102] As can be seen, the multi-dimensional heterogeneous sensing and virtual-real mapping digital twin system for the health of plateau concrete structures constructed in this invention, while fulfilling the traditional structural response monitoring function, further integrates the real-time sensing of key environmental parameters such as temperature, humidity, and air pressure, truly realizing multi-data fusion monitoring from "environmental driving forces" to "structural damage." Simultaneously, through deep integration of deep learning technology, efficiency is improved in data processing and performance prediction, fundamentally solving the technical bottlenecks of lagging data processing and asynchronous analysis in previous monitoring systems.

[0103] The above description is merely a preferred embodiment of the present invention. These specific embodiments are different implementations based on the overall concept of the present invention, and the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A health digital twin system for plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping, characterized in that, include: The system consists of a plateau environment structure coupling sensing layer, a 5G edge intelligent transmission layer, a cloud collaborative data processing layer, and a physical-data dual-drive digital twin layer. The signal output of the plateau environment structure coupling sensing layer is connected to the signal input of the 5G edge intelligent transmission layer, the signal output of the 5G edge intelligent transmission layer is connected to the signal input of the cloud collaborative data processing layer, and the signal output of the cloud collaborative data processing layer is connected to the signal input of the physical-data dual-drive digital twin layer. The plateau environment structure coupling sensing layer is used to capture the external environmental parameters, internal physical fields and damage indicators of concrete structures in real time, and to build a synchronous monitoring system for environment-damage coupling. The 5G edge intelligent transmission layer is used to receive data captured by the plateau environment structure coupling perception layer, and to store and preprocess it. The cloud collaborative data processing layer is used to receive preprocessed data from edge nodes of the 5G edge intelligent transmission layer, perform cloud data processing, and generate a standardized time-series dataset of structural health status. The physical-data dual-driven digital twin layer is used to capture the random noise and sparsity characteristics of the time-series dataset of structural health status generated by the cloud collaborative data processing layer, and simultaneously output the predicted mean and predicted variance of the remaining life of the concrete structure; and calculate the failure probability of the structure under a specific service life.

2. The digital twin system for the health of plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping as described in claim 1, characterized in that, The plateau environment structure coupling sensing layer includes: an external environment monitoring unit and an internal physical field and damage monitoring unit; The external environment monitoring unit uses air temperature and humidity sensors, industrial-grade high-precision air pressure sensors, and intelligent monitoring cameras to capture the dynamic fluctuation characteristics of temperature, humidity, and air pressure in the plateau environment in real time, as well as on-site anti-theft early warning security video data. The internal physical field and damage monitoring unit embeds temperature, humidity and resistivity sensors at different depths inside the concrete structure to invert freeze-thaw damage fronts; welds steel stress sensors and concrete strain sensors in key stress areas; and installs displacement sensors at expansion joints and surface cracks to collect data on temperature, humidity, resistivity, strain, stress and crack propagation.

3. The digital twin system for the health of plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping as described in claim 1, characterized in that, The 5G edge intelligent transmission layer includes: an adaptive signal gain amplifier, a multi-protocol acquisition chip, an energy management unit, and an edge computing unit; the signal input terminal of the multi-protocol acquisition chip is connected to the signal output terminals of the external environment monitoring unit and the internal physical field and damage monitoring unit of the plateau environment structure coupling perception layer, respectively; the signal output terminal of the multi-protocol acquisition chip is connected to the signal input terminal of the edge computing unit; the signal output terminal of the edge computing unit is connected to the signal input terminal of the adaptive signal gain amplifier; and the energy management unit is connected to the adaptive signal gain amplifier, the multi-protocol acquisition chip, and the edge computing unit, respectively. An adaptive signal gain amplifier is used to dynamically adjust the transmit power and signal gain multiple, automatically compensate for link loss in areas with weak signal coverage, and ensure the stability of the data upload link. A multi-protocol acquisition chip is used to parse the connections of various sensors in the high-altitude environment structure coupling sensing layer with different interface protocols, so as to realize multi-channel synchronous acquisition. The energy management unit is used to monitor the external power supply status and the remaining power of the system, and to execute dynamic power scheduling strategies based on energy awareness. Edge computing units are used to perform preprocessing and temporary storage of local data, and to remove random noise and anomalous jump values ​​collected by sensors.

4. The digital twin system for the health of plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping as described in claim 3, characterized in that, The multi-protocol acquisition chip parses different interface protocols, including bus type, vibrating wire type, and resistor type.

5. The digital twin system for the health of plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping as described in claim 1, characterized in that, The cloud collaborative data processing layer is built on a 5G edge computing architecture. Its algorithm implementation relies on an intelligent data analysis platform to receive preprocessed data from edge nodes and perform cloud data processing. The cloud data processing includes: performing multi-source heterogeneous data loading and spatiotemporal alignment, deep data cleaning and preprocessing, multi-scale feature engineering extraction, isolated forest anomaly detection, dimensionality reduction mapping and state clustering, and generating a standardized structural health state time series dataset.

6. The digital twin system for the health of plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping according to claim 1, characterized in that, The physical-data dual-driven digital twin layer adopts a dynamic Bayesian neural network model, which includes an input layer, a Bayesian LSTM layer, a heteroscedastic uncertainty output layer, and a post-processing module. The signal output terminal of the input layer is connected to the signal input terminal of the Bayesian LSTM layer, the signal output terminal of the Bayesian LSTM layer is connected to the signal input terminal of the heteroscedastic uncertainty output layer, and the signal output terminal of the heteroscedastic uncertainty output layer is connected to the signal input terminal of the post-processing module. The input layer is used to map the standardized multi-source heterogeneous time series data into high-dimensional feature vectors and pass them to the Bayesian LSTM layer. Bayesian LSTM layers are used for feature extraction and model uncertainty capture, capturing random noise and sparse features of the monitoring data; Heteroscedastic uncertainty output layer, used to predict the probability distribution of the output concrete service status and quantify data noise; The post-processing module, based on the output probability distribution information, is used to calculate the time-varying reliability index of the structure and introduces the spatiotemporal conversion coefficient.

7. The digital twin system for the health of plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping according to claim 6, characterized in that, The time-varying reliability index of the computational structure is calculated using the first-order second-moment method.

8. A life prediction method for a health digital twin system of plateau concrete structures based on multi-element heterogeneous sensing and virtual-real mapping as described in any one of claims 1-7, characterized in that, Includes the following steps: Step 1: Deploy a plateau environment structure coupling sensing layer that includes external environmental monitoring units and internal physical field and damage monitoring units. Real-time synchronous acquisition of multi-dimensional and multi-physical field data such as temperature, humidity, air pressure, resistivity, strain, stress and crack propagation to form the original monitoring dataset of environmental and structural damage coupling. Step 2: Based on the 5G edge computing architecture, the original monitoring dataset obtained in Step 1 is subjected to protocol parsing, edge-side data processing, and reliable transmission. Through data preprocessing, a standardized time-series dataset for deep learning model analysis is constructed. The standardized time-series dataset is input into the dynamic Bayesian neural network model, and the time-series features are captured through the Bayesian LSTM layer. Combined with heteroscedastic uncertainty, the remaining lifetime prediction results with confidence intervals are output. Step 3: In the physical-data dual-driven digital twin layer, the damage characteristics of the concrete are intelligently identified and extracted based on the remaining service life prediction results in Step 2, and the dynamic evaluation level of the service status is output; combined with reliability theory, a service life prediction model that integrates material properties, service status and environmental coupling effects is constructed to calculate the remaining service life of the concrete structure.

9. The lifetime prediction method according to claim 8, characterized in that, The specific steps for calculating the remaining service life of the concrete structure in step three are as follows: First, the failure threshold is determined using the reliability index output by the Bayesian neural network model. Second, a spatiotemporal conversion coefficient is introduced to establish a mapping equation from the number of accelerated freeze-thaw cycles in the laboratory to the actual service life in engineering projects, thereby calculating the remaining service life, as shown in the following formula: in, To predict service life; To predict the cumulative number of cycles required for the model to reach the failure threshold, The average number of freeze-thaw cycles per year at the project site; This formula represents the equivalent damage coefficient between the laboratory and the field; it converts the number of rapid freeze-thaw cycles in the laboratory into actual engineering service life, providing a practical tool for engineering applications. In determining In the process, this model also incorporates reliability theory, calculating reliability indices. The formula for determining whether a structure has entered a failure state is as follows: The reliability index is calculated using the first-order second-moment method: in: To predict the mean, To predict the standard deviation, This is the failure threshold; when When the reliability is > 99.9%, the structure is safe; when At this time, the reliability is between 90% and 99.9%, requiring regular monitoring; when When the time comes, an early warning is issued, indicating the end of life.

Images