Building structure safety monitoring system based on flexible sensor

Abstract

The invention discloses a building structure safety monitoring system based on a flexible sensor, and relates to the technical field of building structure safety monitoring, and the system comprises a flexible sensing network which is composed of a plurality of distributed flexible sensing units and communicates with adjacent flexible sensing units and edge computing nodes through a wireless ad hoc network mode; the edge computing node is used for carrying out preprocessing and preliminary feature extraction on the original data from the flexible sensor network; the data aggregation gateway is responsible for aggregating the data of the plurality of edge computing nodes, performing protocol conversion and compression, and uploading the data to the cloud monitoring analysis platform; and the cloud monitoring analysis platform is used for carrying out storage, deep analysis and safety state evaluation on the monitoring data and generating early warning information. According to the invention, conformal fitting with the surface of the structure, self-powered wireless monitoring and crossing from data acquisition to intelligent diagnosis are realized, and the accuracy, real-time performance and reliability of monitoring are improved.

Description

Technical Field

[0001] This invention relates to the field of building structure safety monitoring technology, and in particular to a building structure safety monitoring system based on flexible sensors. Background Technology

[0002] In the field of civil engineering and structural health monitoring, real-time and accurate monitoring of the safety status of building structures is crucial for ensuring public safety and extending structural lifespan. This field involves the interdisciplinary integration of multiple disciplines, including mechanics, materials science, sensing technology, and data analysis, aiming to assess performance degradation, damage accumulation, and safety risks through continuous collection of structural response data. The core of structural health monitoring lies in deploying sensor networks to acquire changes in key physical quantities of structures under environmental and operational loads.

[0003] Among them, sensor-based building structure safety monitoring is an important technological direction for ensuring the safe operation of critical infrastructure. This technology involves installing various sensors at key structural components to collect parameters such as strain, displacement, acceleration, and crack width in real time, and using data analysis methods to assess the structural integrity, stiffness changes, and potential damage, thereby enabling early warning and assessment of the structural safety status.

[0004] Existing technologies primarily rely on traditional rigid sensors, such as resistance strain gauges, fiber Bragg grating sensors, or accelerometers. These sensors have significant limitations in long-term outdoor applications in complex environments: their rigid substrates exhibit poor deformation compatibility with building structure surfaces, easily leading to stress concentration at the interface, resulting in measurement distortion or sensor damage; installation is typically complex and requires high-quality surface treatment, hindering rapid deployment over large areas and at high density; furthermore, traditional sensor systems often employ wired transmission, resulting in cumbersome and costly wiring, and the wiring itself is susceptible to environmental corrosion or human damage, affecting the reliability of the monitoring network. For large and complex structures, existing monitoring solutions struggle to achieve truly distributed, full-surface strain field sensing, failing to accurately capture the initiation and evolution of local micro-damage, leading to delayed early warnings. Summary of the Invention

[0005] The purpose of this invention is to provide a building structure safety monitoring system based on flexible sensors to solve the problems of poor coordination between rigid sensors and structural surface deformation, difficulty in deployment, and difficulty in achieving distributed full-surface accurate perception in the prior art, which leads to monitoring distortion and delayed early warning.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A building structure safety monitoring system based on flexible sensors, comprising: The flexible sensor network consists of multiple distributed flexible sensor units. Each flexible sensor unit integrates a flexible strain sensor array, a flexible temperature compensation sensor, and a self-powered module, and communicates with neighboring flexible sensor units and edge computing nodes through a wireless self-organizing network. Edge computing nodes are deployed near the monitoring area to preprocess and perform preliminary feature extraction on raw data from the flexible sensor network; The data aggregation gateway is responsible for aggregating data from multiple edge computing nodes, performing protocol conversion and compression, and then uploading it to the cloud-based monitoring and analysis platform. The cloud-based monitoring and analysis platform is used to store, deeply analyze, and assess the security status of massive amounts of monitoring data, and generate early warning information.

[0007] The flexible sensing unit employs a multilayer heterogeneous integrated structure, with its substrate being a flexible polymer film of polyimide or polydimethylsiloxane, with a thickness ranging from 50 micrometers to 200 micrometers. A serpentine, meandering metal nanowire strain sensing grid is fabricated on the flexible substrate using micro-nano processing techniques. This grid constitutes a flexible strain sensor array, with each sensing unit measuring 2 millimeters by 2 millimeters. The overall coverage area of ​​the array can be cut and spliced ​​according to monitoring requirements. Flexible temperature compensation sensors based on negative temperature coefficient thermistors are integrated between adjacent channels of the strain sensing grid for real-time measurement of the temperature field at the location of the sensing unit. The self-powered module consists of a flexible piezoelectric energy harvesting sheet and a micro supercapacitor. The flexible piezoelectric energy harvesting sheet is attached to the underside of the strain sensing grid, converting the mechanical energy of the structural micro-vibrations into electrical energy and charging the micro supercapacitor, which then powers the sensing and wireless communication circuitry of the entire flexible sensing unit.

[0008] The flexible strain sensor array operates in two modes: static and dynamic. In static mode, the building structure safety monitoring system based on flexible sensors acquires the quasi-static strain distribution of the structure at a frequency of once per minute. In dynamic mode, when the edge computing node detects that the structural response exceeds a preset threshold, it triggers the flexible sensing unit to acquire dynamic strain time history data at a sampling frequency of up to 1000 Hz. The microprocessor built into each flexible sensing unit performs analog-to-digital conversion on the multi-channel raw signals acquired by the array and executes a temperature drift compensation algorithm based on data from the flexible temperature compensation sensor to eliminate the influence of ambient temperature changes on the strain measurement values.

[0009] The wireless ad hoc network employs a hybrid protocol combining time division multiple access (TDMA) and frequency hopping spread spectrum. During network initialization, a synchronization beacon is broadcast by the data aggregation gateway. Each flexible sensing unit and edge computing node dynamically selects the optimal parent node and joins the network based on the received signal strength, forming a multi-hop tree topology. Each communication time slot is divided into multiple sub-time slots, used for beacon broadcasting, data uploading, command issuance, and network maintenance, respectively. The flexible sensor-based building structure safety monitoring system operates in the 2.4 GHz industrial, scientific, and medical frequency band and employs a frequency hopping pattern to avoid co-channel interference, ensuring communication reliability in complex electromagnetic environments.

[0010] The preprocessing and feature extraction process for the edge computing nodes is as follows: Step S1: The edge computing node first receives the compensated strain data packets uploaded by all flexible sensing units within its jurisdiction. The data packets contain the sensing unit's identity, timestamp, and strain values ​​for each channel. Step S2: The edge computing node performs a data alignment operation to synchronize the data collected by different sensing units at the same time in the time dimension. Step S3: The edge computing node runs a local strain field reconstruction algorithm. This algorithm generates a continuous two-dimensional strain cloud map covering the monitoring area based on the position coordinates of the sensing unit and the strain measurement value through bilinear interpolation. Step S4: The edge computing node extracts a feature vector from the reconstructed strain field. The feature vector includes the maximum principal strain value, the minimum principal strain value, the magnitude of the strain gradient vector, and the strain energy density distribution statistics of the monitoring area. The feature vector is then compared with the corresponding strain contour map. Figure 1 The package is sent to the data aggregation gateway.

[0011] The cloud-based monitoring and analysis platform comprises a data warehouse, a digital twin model engine, a damage identification algorithm module, and an early warning decision module. The data warehouse employs a hybrid architecture of time-series and relational databases to store historical and real-time monitoring data. The digital twin model engine loads a refined finite element model of the monitored building structure. This model can dynamically simulate real-time environmental load data, including wind speed, temperature, and traffic flow, to calculate the theoretical strain response of the structure under healthy conditions. The core of the damage identification algorithm module is a deep residual convolutional neural network. The input to this network is the actual monitored strain feature vector uploaded from edge computing nodes and the digital twin. The difference between the theoretical strain feature vectors calculated by the model engine, i.e., the strain residual field features, is used by the network after training to map the strain residual field features to a preset damage pattern library. The output includes damage probability, possible damage location, and damage severity level. The early warning decision module receives the output results of the damage identification algorithm module and generates early warning instructions according to a preset multi-level early warning threshold strategy. The early warning level is divided according to the combination of damage probability and severity. When a high-level early warning is determined, the building structure safety monitoring system based on flexible sensors automatically sends alarm information containing a damage location diagram and suggested treatment measures to a preset management personnel terminal.

[0012] The building structure safety monitoring system based on flexible sensors also includes a self-calibration and health diagnosis subsystem, which operates periodically, and its process is as follows; Step L1: The cloud monitoring and analysis platform sends a self-test command to the designated edge computing node and flexible sensing unit via the data aggregation gateway. Step L2: After receiving the command, the flexible sensing unit uses its built-in microprocessor to control a micro exciter to generate a micro strain of known amplitude, while simultaneously recording the response output of the flexible strain sensor array. Step L3: The edge computing node collects the self-test response data of each sensing unit and compares it with the standard response curve to calculate the sensitivity coefficient drift and zero-point drift of each sensing channel. In step L4, if the drift exceeds the allowable tolerance, the sensing unit is marked as requiring calibration or in a faulty state, and its data is weighted or removed in subsequent data analysis. At the same time, the self-calibration and health diagnosis subsystem will record the event and prompt maintenance requirements.

[0013] The flexible sensing unit is deployed using a reversible smart adhesive interface, consisting of two layers: a lower pressure-sensitive adhesive layer providing initial adhesion, and an upper heat-reversible phase-change polymer layer. During deployment, a specialized heating tool is used to briefly heat the back of the flexible sensing unit, softening the phase-change polymer layer. The sensing unit is then adhered to the structural surface under pressure. Upon cooling, the phase-change polymer layer solidifies, forming a strong bond with the structural surface and the pressure-sensitive adhesive layer. When the sensing unit needs to be replaced or removed, it is locally heated again above the phase-change temperature, significantly reducing the adhesion and allowing for non-destructive peeling. This method ensures long-term conformal adhesion between the sensing unit and the structural surface without damaging the structural coating.

[0014] Compared with the prior art, the beneficial technical effects of the present invention are as follows: This invention achieves mechanical compatibility and conformal fit between the sensor and the building structure surface by employing a flexible sensing unit with a serpentine, meandering metal nanowire mesh as its core. The flexible substrate can adapt to the curvature changes and minute deformations of the structural surface, fundamentally avoiding the measurement distortion and self-damage problems caused by interface stress concentration in traditional rigid sensors. The distributed flexible sensing network can achieve large-area, high-density strain field coverage, accurately capturing local stress concentration and micro-damage initiation, significantly improving the spatial resolution of monitoring and the ability to identify early damage. This invention constructs a complete monitoring system integrating self-powered, wireless self-organizing network, edge intelligent preprocessing, and cloud-based deep analysis. The flexible piezoelectric self-powered module eliminates the constraints of wired power supply, making sensor deployment extremely flexible, especially suitable for areas with existing structures or where wiring is difficult. The wireless self-organizing network protocol based on time division multiple access and frequency hopping ensures reliable, low-power communication of large-scale sensor networks in complex environments. Edge computing nodes perform data preprocessing and feature extraction, significantly reducing the amount of data that needs to be uploaded, lowering cloud load and communication bandwidth requirements, while improving system real-time performance.

[0015] This invention achieves a leap from "data acquisition" to "intelligent diagnosis" in monitoring by deeply integrating real-time monitoring data with theoretical simulations based on digital twins and applying deep residual convolutional neural networks for damage identification. The digital twin model provides a strain benchmark under healthy conditions, and the strain residual field effectively amplifies the abnormal signals caused by damage. The deep neural network can automatically learn the complex nonlinear mapping relationship between damage features and strain residual fields, achieving high-precision and automated identification and location of multiple damage modes, greatly reducing the workload and subjective errors of manual interpretation, and making structural safety early warning more timely, accurate, and reliable.

[0016] The self-calibration and health diagnosis subsystem and reversible intelligent bonding interface integrated in this invention enhance the long-term service stability and maintainability of the building structure safety monitoring system based on flexible sensors. Periodic self-calibration can promptly detect and compensate for sensor performance drift, ensuring the long-term accuracy of monitoring data. The reversible bonding interface makes the deployment, replacement, and maintenance of flexible sensing units convenient and non-destructive, reducing the operation and maintenance costs throughout the entire life cycle. This lays a solid technical foundation for the large-scale and sustainable application of the building structure safety monitoring system based on flexible sensors in various building structures. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall technical architecture of the building structure safety monitoring system based on flexible sensors proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of intelligent damage recognition that deeply integrates real-time monitoring data and digital twin simulation in this invention. Detailed Implementation

[0018] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the present invention and not to limit the present invention. For those skilled in the art, the present invention can be practiced without some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present invention by illustrating examples of the invention.

[0019] Example 1 The present invention provides a building structure safety monitoring system based on flexible sensors, the overall technical architecture of which is shown in the attached figure. Figure 1 As shown, the system consists of four main parts: a flexible sensor network, edge computing nodes, a data aggregation gateway, and a cloud-based monitoring and analysis platform. These parts communicate efficiently via wired or wireless communication links, forming a closed-loop, intelligent structural health perception and decision-making system. The overall system design aims to overcome the inherent limitations of traditional rigid sensors deployed on building structures, such as mechanical incompatibility, sparse spatial coverage, difficulties in power supply wiring, and data processing lag. This allows for high-precision, low-power, and adaptive safety status monitoring of large and complex building structures throughout their entire lifecycle.

[0020] The flexible sensing network is the front-end data sensing layer of this system, with flexible sensing units as its core components. Each flexible sensing unit adopts a multi-layer heterogeneous integrated structure, with an overall thickness controlled between 50 and 200 micrometers to ensure excellent flexibility and adhesion. The substrate material of the flexible sensing unit is a flexible polymer film made of polyimide or polydimethylsiloxane, both of which have good mechanical ductility, thermal stability, and chemical inertness, enabling them to withstand long-term exposure to ultraviolet radiation, humidity changes, and temperature cycling in outdoor environments. On this flexible substrate, a serpentine, meandering metal nanowire strain sensing grid is fabricated using micro-nano fabrication processes. This grid constitutes a flexible strain sensor array. The serpentine structure design of individual sensing channels effectively improves their electrical conductivity stability under tensile or bending deformation, avoiding signal failure caused by fracture of traditional linear conductors under strain. In the flexible strain sensor array, each independent sensing unit has a physical size of 2 mm by 2 mm. The array as a whole can be cut, spliced ​​or rolled up according to the geometry of the monitored structure to achieve seamless coverage of planar, curved or even complex topological surfaces.

[0021] Between adjacent sensing channels of the flexible strain sensor array, a flexible temperature compensation sensor based on a negative temperature coefficient thermistor material is integrated. This temperature compensation sensor shares the same flexible substrate with the strain sensing grid and is manufactured synchronously using the same micro-nano processes, ensuring high spatial alignment and consistent thermal response characteristics. The temperature compensation sensor acquires the temperature value of the local microenvironment of the flexible sensing unit in real time, with a sampling frequency synchronized with the strain sensing, typically once per minute in static mode or up to 1000 Hz in dynamic mode. This temperature data is used in subsequent temperature drift compensation algorithms to eliminate strain measurement errors caused by ambient temperature fluctuations. Since the resistivity of metal nanowires has a significant temperature dependence, without compensation, a 1-degree Celsius temperature change can lead to equivalent strain drift of tens of microstrains, severely affecting monitoring accuracy. Therefore, the integration of the temperature compensation sensor is a crucial step in ensuring the long-term measurement accuracy of the system.

[0022] The flexible sensing unit also integrates a self-powered module, which consists of a flexible piezoelectric energy harvesting sheet and a miniature supercapacitor. The flexible piezoelectric energy harvesting sheet is made of lead zirconate titanate or polyvinylidene fluoride composite material, its physical dimensions matched to the strain sensing grid, and is attached to the back of the flexible substrate, i.e., the side furthest from the structural surface, to maximize the capture of the micro-amplitude mechanical vibration energy generated by the structure under wind loads, traffic vibrations, or seismic excitation. After the piezoelectric energy harvesting sheet converts the mechanical energy into alternating current (AC), it is then converted into direct current (DC) by a built-in rectifier and voltage regulator circuit to charge the miniature supercapacitor. The miniature supercapacitor uses a flexible solid-state electrolyte encapsulation, is small in size, has a long cycle life, and high charge / discharge efficiency, enabling it to continuously power the analog-to-digital converter, microprocessor, and wireless communication chip inside the flexible sensing unit without an external power source. This self-powered module completely eliminates the dependence on wired power supply in traditional monitoring systems, allowing the flexible sensing unit to operate stably for extended periods in areas where wiring is difficult, such as bridge cables, high-rise curtain walls, and the ridges of ancient buildings.

[0023] The flexible sensing unit integrates a low-power microprocessor, responsible for tasks such as local data acquisition, analog-to-digital conversion, temperature compensation, and communication scheduling. The system supports two operating modes: static and dynamic. In static mode, the system performs routine monitoring, polling all sensing channels once per minute to obtain the quasi-static strain distribution of the structure under constant load. When an edge computing node detects that the strain response in a certain area exceeds a preset threshold, it triggers the relevant flexible sensing unit to enter dynamic mode via a downlink command. The preset threshold may include, for example, a principal strain exceeding 50 microstrains or a sudden increase in strain gradient. In dynamic mode, the flexible sensing unit increases its sampling frequency to a maximum of 1000 Hz, continuously acquiring dynamic strain time-history data for at least 10 seconds to capture the transient response characteristics of the structure under sudden loads. After dynamic data acquisition is complete, the flexible sensing unit automatically returns to static mode to save energy. Constant loads include self-weight and temperature fields, while sudden loads include strong winds, vehicle impacts, and earthquakes.

[0024] Communication between flexible sensing units and connections with edge computing nodes are achieved through a wireless ad hoc network. This wireless network is built on a hybrid protocol combining time division multiple access (TDMA) and frequency hopping spread spectrum. During network initialization, the data aggregation gateway first broadcasts a synchronization beacon frame, which includes a global time base, network identifier, and current frequency hopping pattern. Each flexible sensing unit and edge computing node listens to this beacon after power-on and dynamically evaluates the communication quality of surrounding nodes based on the received signal strength. It then selects the node with the strongest signal and lowest load as its parent node and joins the network. Thus, the entire network spontaneously forms a multi-hop tree topology with the data aggregation gateway as the root node, exhibiting good scalability and fault tolerance. Each communication cycle is divided into fixed-length time slots, each of which is further subdivided into four sub-time slots: the first sub-time slot is used for the root node to broadcast a synchronization beacon; the second sub-time slot is allocated to each flexible sensing unit to upload compensated strain data packets; the third sub-time slot is used for edge computing nodes or data aggregation gateways to issue control commands, such as switching operating modes or initiating self-tests; and the fourth sub-time slot is reserved for network maintenance, including node joining / leaving, route updates, and link quality detection. The system operates in the 2.4 GHz industrial, scientific, and medical frequency band. The frequency hopping pattern is generated by a pseudo-random sequence, with no less than 50 hops per second, effectively avoiding co-channel interference and multipath fading, ensuring communication reliability in densely reinforced concrete and electromagnetically noisy building environments. Each data packet employs a dual protection mechanism of forward error correction coding and cyclic redundancy check, controlling the bit error rate to within 10%. -6 the following.

[0025] Edge computing nodes are deployed in physical proximity to the monitored area, typically installed in equipment rooms within structures, bridge tower platforms, or protective enclosures on building facades. Each edge computing node manages a local sensor cluster, with a typical coverage radius of 30 to 50 meters, and can connect to dozens to hundreds of flexible sensing units. The core function of the edge computing node is to preprocess and perform preliminary feature extraction on the raw sensor data to reduce the burden on subsequent transmission and cloud computing. The processing flow is as follows: First, the edge computing node receives data packets uploaded from its managed flexible sensing units. Each data packet contains the identification of the flexible sensing unit, a precise timestamp with an accuracy better than 1 millisecond, the original strain values ​​of each channel, and the corresponding temperature compensation values. Next, the edge computing node performs data alignment. Due to slight delays in wireless transmission, the arrival times of data from different sensing units are slightly offset. The edge computing node uses its own high-precision real-time clock as a reference and employs linear interpolation methods to uniformly align the data from all sensing units to the same time grid, ensuring the spatiotemporal consistency of subsequent strain field reconstruction.

[0026] After time alignment, the edge computing nodes run a local strain field reconstruction algorithm. This algorithm, based on the known physical coordinates of each flexible sensing unit on the structural surface and its corresponding strain measurements, uses bilinear interpolation to generate a continuous two-dimensional strain distribution cloud map between discrete measurement points. The physical coordinates are imported during deployment via a laser rangefinder or BIM model. Specifically, for any point P(x,y) within the monitoring area, its strain value ε(P) is obtained by a weighted average of the strain values ​​at the four vertices of its corresponding grid unit. The spatial resolution of this strain cloud map can reach the order of 1 millimeter, far exceeding the sparse sampling of traditional point sensors. Subsequently, the edge computing nodes extract a set of high-dimensional feature vectors from this strain cloud map. These feature vectors contain four core elements: the maximum principal strain value of the monitoring area, reflecting the tensile state of the most critical section of the structure; the minimum principal strain value, reflecting the compressive state; the magnitude of the strain gradient vector, characterizing the drastic changes in strain space and sensitive to crack initiation; and statistics of the strain energy density distribution, including the mean, variance, and peak value. In addition, the edge computing nodes generate a thumbnail of the strain cloud map, package it together with the feature vector, and send it to the data aggregation gateway via wired Ethernet or 4G / 5G wireless link. The thumbnail's resolution is compressed to 128×128 pixels. This process compresses the original data volume to less than 5% of its original size, significantly improving the overall system efficiency.

[0027] The data aggregation gateway, located in the central control room or communication room at the monitoring site, is primarily responsible for aggregating data streams from multiple edge computing nodes, performing protocol conversion, data compression, and security encryption before uploading them to the cloud-based monitoring and analysis platform. The data aggregation gateway employs an industrial-grade embedded computer architecture, equipped with dual power supply redundancy and lightning protection to ensure 24 / 7 uninterrupted operation. Its uplink supports multiple WAN interfaces, including fiber optic, 4G / 5G, and satellite communication, and can automatically switch according to on-site network conditions. Before data upload, the data aggregation gateway performs secondary compression on data packets using the LZ77 algorithm and adds a digital signature to prevent data tampering or theft during transmission.

[0028] The cloud-based monitoring and analysis platform is the intelligent decision-making center of this system, and its logical architecture is shown in the attached figure. Figure 2 As shown, the platform is deployed on a public or private cloud server cluster and comprises four core components: a data warehouse, a digital twin model engine, a damage identification algorithm module, and an early warning decision module. The data warehouse employs a hybrid architecture of time-series and relational databases. The time-series database efficiently stores massive, high-frequency real-time monitoring data streams, supporting millisecond-level timestamp queries and downsampling aggregation; the relational database stores structural metadata, historical event records, and user permission information. Structural metadata includes sensor locations, material parameters, and design drawings. The two databases are periodically synchronized using ETL tools to ensure data consistency.

[0029] The digital twin model engine is a key support for the intelligent diagnosis of this system. This engine loads a refined finite element model of the monitored building structure. The model mesh is refined to the component level, and material properties, boundary conditions, and connection methods are strictly set according to as-built drawings and on-site survey data. During operation, the digital twin model engine receives real-time external environmental load data, including wind speed and direction, temperature and humidity from weather stations, traffic flow and vehicle type distribution from traffic management departments, and real-time seismic motion parameters from earthquake early warning systems. Based on these inputs, the model engine dynamically simulates and calculates the theoretical strain response field of the structure under the current load combination and outputs a theoretical strain feature vector with dimensions completely consistent with the feature vectors of the edge computing nodes. This theoretical feature vector represents the expected behavior of the structure in a "healthy and undamaged" state, providing a benchmark reference for subsequent damage identification.

[0030] The core of the damage identification algorithm module is a deep residual convolutional neural network. The input to this network is not the original strain data, but rather the difference between the actual monitored feature vector and the theoretical strain feature vector, i.e., the strain residual field feature. The strain residual field effectively amplifies the abnormal signals caused by structural damage while suppressing the common-mode response caused by normal load changes. Structural damage includes cracks, corrosion, and loose connections. The deep residual convolutional neural network uses a 50-layer ResNet architecture, containing multiple residual blocks. Each residual block consists of two 3×3 convolutional layers, a batch normalization layer, and a ReLU activation function, with skip connections introduced to alleviate the gradient vanishing problem. The network's output layer uses the Softmax activation function to map the input to a pre-defined damage pattern library. This damage pattern library covers common building structural damage types, such as concrete cracking, rebar yielding, support detachment, and node loosening. Each type of damage is labeled with its typical location, severity level (1 to 5), and probability confidence. During the training phase, the network uses a large amount of simulated and experimental data for supervised learning. The loss function is a weighted combination of cross-entropy and mean squared error. The optimizer is Adam, and the initial learning rate is set to 0.001. The model converges after 200 iterations. After training, the model is deployed on a cloud GPU server, with inference latency of less than 200 milliseconds.

[0031] The early warning decision module receives the output from the damage identification algorithm module and generates the final early warning instruction based on a preset multi-level early warning threshold strategy. Early warning levels are divided into three levels: Level 1 (yellow) indicates a damage probability between 30% and 60% or a severity level of 1 to 2; the system only logs this information in the management backend and provides a notification. Level 2 (orange) indicates a damage probability between 60% and 90% or a severity level of 3; the system sends SMS and app notifications to designated technical personnel, along with a damage location diagram and preliminary handling suggestions. Level 3 (red) indicates a damage probability exceeding 90% or a severity level of 4 to 5; the system immediately and automatically dials a preset management personnel number and sends a high-priority alert via email, SMS, and app; the alert content includes a 3D damage location map, historical evolution trends, suggested emergency measures, and a contact list. All early warning events are stored in the data warehouse and trigger the work order system to generate maintenance tasks.

[0032] To ensure the long-term accuracy and reliability of the system, this invention also integrates a self-calibration and health diagnosis subsystem. This subsystem operates automatically according to a preset cycle or on demand, for example, once a month or upon receiving an alarm for abnormal data. During operation, the cloud-based monitoring and analysis platform first sends a self-test command to the data aggregation gateway, specifying the edge computing nodes and flexible sensing units to be tested. The data aggregation gateway forwards the command to the target edge computing node, which then broadcasts it to all flexible sensing units under its jurisdiction. Upon receiving the command, the flexible sensing unit activates a micro-exciter—typically a micro-piezoelectric ceramic sheet or shape memory alloy wire integrated within the flexible substrate—through its built-in microprocessor. Driven by a control signal, the exciter generates a micro-strain excitation with a known amplitude (e.g., 100 micro-strain) and waveform (e.g., step or sine). Simultaneously, the flexible strain sensor array synchronously records the response output of each channel. This response data contains a complete excitation-response time-series curve.

[0033] After collecting the self-test response data of all flexible sensing units, the edge computing node compares it with pre-stored standard response curves. These standard response curves, obtained during laboratory calibration, represent the input-output relationship of the sensor under ideal conditions. The comparison process uses least squares fitting to calculate the sensitivity coefficient of each sensing channel, i.e., the rate of change of resistance per unit strain and the zero-point offset. If the sensitivity drift of a channel exceeds ±5% or the zero-point drift exceeds ±20 microstrains, the channel is considered to have abnormal performance. The edge computing node marks the abnormal channel as "needs calibration" or "faulty," and applies a low weight (e.g., a weighting coefficient of 0.3) or directly removes the channel's data in subsequent data analysis. Simultaneously, the system generates a health diagnostic report, uploads it to the cloud, and prompts maintenance personnel in the management interface to arrange on-site calibration or replacement. This mechanism effectively solves the performance degradation problem caused by material aging and environmental corrosion during long-term sensor service, ensuring the reliability of monitoring data.

[0034] The physical deployment of the flexible sensing unit employs a reversible smart adhesive interface composed of two functional materials. The lower layer is an acrylic pressure-sensitive adhesive layer, approximately 20 micrometers thick, providing initial adhesion and allowing the flexible sensing unit to be temporarily fixed to the structural surface during initial deployment. The upper layer is a heat-reversible phase-change polymer layer, made of an epoxy-amine system or crystalline polyurethane containing dynamic covalent bonds, with its glass transition temperature or melting point set between 60°C and 80°C. During deployment, maintenance personnel use specialized heating tools such as a hot air gun or infrared heating plate to uniformly heat the back of the flexible sensing unit to 70°C for 10 to 15 seconds, softening the phase-change polymer layer into a viscous flow state. At this point, the sensing unit is precisely adhered to the cleaned structural surface, and a uniform pressure of 0.1 MPa to 0.3 MPa is applied and maintained for 30 seconds. After cooling to room temperature, the phase change polymer layer re-cures, forming a three-dimensional interpenetrating network structure with the pressure-sensitive adhesive layer and the structural surface. The bonding strength can reach over 1 MPa, sufficient to resist wind and rain erosion and structural vibration. When it is necessary to replace or remove the sensing unit, it can be locally heated to over 70 degrees Celsius again. The phase change polymer layer softens, and the bonding strength drops sharply to below 0.01 MPa, allowing for easy and non-destructive peeling by hand or tools without leaving adhesive residue or damaging the original coating or finish of the structure. This deployment method balances the robustness of long-term service with the convenience of later maintenance, greatly reducing the operation and maintenance costs throughout the system's life cycle.

[0035] In summary, the building structure safety monitoring system based on flexible sensors described in this embodiment constructs a high-precision, highly robust, and low-maintenance intelligent monitoring system through conformal bonding of flexible sensing units, self-powered wireless networking, edge intelligent preprocessing, cloud-based digital twin fusion diagnosis, and self-calibration maintenance mechanisms. It is suitable for the safety assurance of various major infrastructures such as bridges, dams, super high-rise buildings, and stadiums.

[0036] Example 2 Based on Example 1, this example adapts and optimizes the system to meet the specific monitoring needs of cross-sea bridges in a marine environment. Cross-sea bridges are exposed to harsh environments with high salt spray, high humidity, strong ultraviolet radiation, and typhoon loads for extended periods, placing higher demands on the sensors' corrosion resistance, communication interference resistance, and dynamic response accuracy.

[0037] The substrate material for the flexible sensing unit is preferably polydimethylsiloxane, as its dimensional stability in humid and hot environments is superior to that of polyimide. A 50-nanometer-thick atomic-layer deposited alumina film is coated on the surface of the metal nanowire strain sensing mesh. This dense, non-porous film effectively blocks chloride ion penetration and prevents electrochemical corrosion of the silver or copper nanowires. The flexible temperature-compensated sensor uses a platinum resistance film instead of a negative temperature coefficient thermistor, due to its superior linearity and long-term stability over a wide temperature range of -40°C to 120°C.

[0038] The wireless ad hoc network protocol adds an adaptive power control mechanism to the original protocol. Each flexible sensing unit dynamically adjusts its transmit power, ranging from 1 milliwatt to 100 milliwatts, based on its distance from the parent node and channel quality. This ensures communication reliability while maximizing the battery life of the self-powered module. Simultaneously, the frequency hopping pattern update cycle is shortened to 10 seconds to address rapid channel fading caused by multipath effects over the sea.

[0039] Edge computing nodes are deployed in temperature- and humidity-controlled cabinets inside the bridge tower, with added surge protection and electromagnetic shielding measures. Its local strain field reconstruction algorithm incorporates a wave load spectrum correction term, considering high-frequency vibration components caused by wave impact, thus avoiding misinterpreting normal dynamic responses as damage signals.

[0040] The digital twin model engine of the cloud-based monitoring and analysis platform additionally integrates a marine environmental load module, which receives real-time data on tide level, wave height, ocean current velocity, and salt spray concentration to correct theoretical strain responses. The training dataset for the damage identification algorithm module has been expanded to include a large number of simulated marine corrosion damage samples, such as typical defects like concrete cover cracking caused by steel reinforcement rust and broken prestressed steel strands.

[0041] The self-calibration subsystem operates once a week and is automatically triggered after each typhoon. The phase change polymer layer of the reversible smart adhesive interface is now formulated with a salt spray resistant formula, and its peel strength retains more than 90% of its initial value after 1000 hours of salt spray testing.

[0042] Through the above optimizations, the system in this embodiment demonstrates excellent environmental adaptability and monitoring reliability in cross-sea bridge applications, successfully achieving all-weather safety monitoring of the main beam, cable anchorage zone, and pier scour zone.

[0043] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A building structure safety monitoring system based on flexible sensors, characterized in that, include: The flexible sensor network consists of multiple distributed flexible sensor units. Each flexible sensor unit integrates a flexible strain sensor array, a flexible temperature compensation sensor, and a self-powered module, and communicates with neighboring flexible sensor units and edge computing nodes through a wireless self-organizing network. Edge computing nodes, deployed near the monitoring area, are used to preprocess and perform preliminary feature extraction on raw data from the flexible sensor network. The data aggregation gateway is used to aggregate data from multiple edge computing nodes, perform protocol conversion and compression, and then upload the data. The cloud-based monitoring and analysis platform is used to store, deeply analyze, and assess the security status of massive amounts of monitoring data, and generate early warning information.

2. The building structure safety monitoring system based on flexible sensors according to claim 1, characterized in that: The flexible sensing unit adopts a multilayer heterogeneous integrated structure, with its substrate being a flexible polymer film of polyimide or polydimethylsiloxane with a thickness between 50 micrometers and 200 micrometers. A serpentine metal nanowire strain sensing grid is fabricated on the flexible substrate using micro-nano processing technology, and the strain sensing grid constitutes a flexible strain sensor array. A flexible temperature compensation sensor based on a negative temperature coefficient thermistor material is integrated between adjacent channels of the strain sensing grid. The self-powered module consists of a flexible piezoelectric energy harvesting plate and a micro supercapacitor. The flexible piezoelectric energy harvesting plate is attached to the underside of the strain sensing grid, converting the mechanical energy of the structural micro-vibration into electrical energy and charging the micro supercapacitor.

3. The building structure safety monitoring system based on flexible sensors according to claim 1, characterized in that: The preprocessing and feature extraction process for the edge computing nodes is as follows: Step S1: Receive compensated strain data packets uploaded from all flexible sensing units within its jurisdiction. Step S2: Perform data alignment operation to synchronize the data collected by different sensing units at the same time in the time dimension; Step S3: Run the local strain field reconstruction algorithm. The local strain field reconstruction algorithm generates a continuous two-dimensional strain cloud map covering the monitoring area based on the position coordinates of the sensing unit and the strain measurement value through bilinear interpolation. Step S4: Extract feature vectors from the reconstructed strain field. The feature vectors include the maximum principal strain value, the minimum principal strain value, the magnitude of the strain gradient vector, and the strain energy density distribution statistics of the monitoring area. The feature vectors are then packaged together with the corresponding strain cloud map thumbnails and sent to the data aggregation gateway.

4. The building structure safety monitoring system based on flexible sensors according to claim 1, characterized in that: The cloud-based monitoring and analysis platform includes a data warehouse, a digital twin model engine, a damage recognition algorithm module, and an early warning decision module. The digital twin model engine is loaded with a refined finite element model of the monitored building structure. The finite element model can perform dynamic simulation based on real-time environmental load data and calculate the theoretical strain response of the structure in a healthy state. The core of the damage identification algorithm module is a deep residual convolutional neural network. The input of the deep residual convolutional neural network is the difference between the actual monitored strain feature vector uploaded from the edge computing node and the theoretical strain feature vector calculated by the digital twin model engine, i.e., the strain residual field feature. The early warning decision module receives the output of the damage identification algorithm module and generates early warning instructions based on the preset multi-level early warning threshold strategy.

5. The building structure safety monitoring system based on flexible sensors according to claim 4, characterized in that: The data warehouse adopts a hybrid architecture of time-series database and relational database to store historical and real-time monitoring data; the damage identification algorithm module, after training, can map strain residual field features to a preset damage pattern library, and the output includes damage probability, possible damage location, and damage severity level; the early warning decision module generates early warning instructions, and the early warning level is divided according to the combination of damage probability and severity; when a high-level early warning is determined, the system automatically sends alarm information containing a damage location diagram and suggested treatment measures to the preset management personnel terminal.

6. The building structure safety monitoring system based on flexible sensors according to claim 1, characterized in that, The flexible strain sensor array operates in two modes: static and dynamic. In static mode, the system acquires the quasi-static strain distribution of the structure once per minute. In dynamic mode, when the edge computing node detects that the structural response exceeds a preset threshold, it triggers the flexible sensing unit to acquire dynamic strain time history data at a sampling frequency of up to 1000 Hz. The microprocessor built into each flexible sensing unit performs analog-to-digital conversion on the multi-channel raw signals acquired by the array and executes a temperature drift compensation algorithm based on the data from the flexible temperature compensation sensor. The size of a single sensing unit in the flexible strain sensor array is 2 mm by 2 mm, and the overall coverage area of ​​the array is cut and spliced ​​according to monitoring requirements.

7. The building structure safety monitoring system based on flexible sensors according to claim 1, characterized in that, The wireless ad hoc network uses a hybrid protocol combining time division multiple access and frequency hopping spread spectrum for communication. During network initialization, the data aggregation gateway broadcasts a synchronization beacon, and each flexible sensing unit and edge computing node dynamically selects the optimal parent node and joins the network based on the received signal strength indication, forming a multi-hop tree topology. Each communication time slot is divided into multiple sub-time slots, which are used for beacon broadcasting, data uploading, command issuance, and network maintenance, respectively. The building structure safety monitoring system based on flexible sensors operates in the 2.4 GHz industrial, scientific, and medical frequency band and uses a frequency hopping pattern to avoid co-channel interference.

8. The building structure safety monitoring system based on flexible sensors according to claim 1, characterized in that, The system also includes a self-calibration and health diagnosis subsystem, which operates periodically, and its process is as follows: Step L1: The cloud monitoring and analysis platform sends a self-test command to the designated edge computing node and flexible sensing unit via the data aggregation gateway. Step L2: After receiving the command, the flexible sensing unit uses its built-in microprocessor to control a micro exciter to generate a micro strain of known amplitude, while simultaneously recording the response output of the flexible strain sensor array. Step L3: The edge computing node collects the self-test response data of each sensing unit and compares it with the standard response curve to calculate the sensitivity coefficient drift and zero-point drift of each sensing channel. In step L4, if the drift exceeds the allowable tolerance, the sensing unit is marked as requiring calibration or in a fault state. In subsequent data analysis, the data of the sensing units marked as requiring calibration or in a fault state are weighted or removed.

9. The building structure safety monitoring system based on flexible sensors according to claim 1, characterized in that, The flexible sensing unit is deployed using a reversible smart adhesive interface. The smart adhesive interface consists of two layers: a pressure-sensitive adhesive layer on the bottom and a heat-reversible phase change polymer layer on the top. During deployment, a special heating tool is used to briefly heat the back of the flexible sensing unit to soften the phase change polymer layer. At this time, the sensing unit is attached to the surface of the structure and pressure is applied. After cooling, the phase change polymer layer solidifies to form a strong and tough bond. When it is necessary to replace or remove the sensing unit, it is locally heated again to above the phase change temperature and then peeled off without damage.

Images