Dynamic visual monitoring and early warning method of bridge internet of things based on IFC

Abstract

The invention provides an IFC-based bridge Internet of Things dynamic visual monitoring and early warning method, and relates to the technical field of railway traffic infrastructure monitoring, and the method comprises the steps: obtaining the sensor monitoring data of a railway bridge; an original IFC dictionary file is obtained, entity expansion is carried out on the arranged sensors, attribute set expansion is carried out on railway bridge sensor types to obtain a basic railway bridge sensor data IFC file, and a standardized and visual data carrier is provided for bridge sensor monitoring information; and loading the file based on a BIM platform and constructing a railway bridge-oriented adaptive prediction model source monitoring data for analysis and trend prediction, thereby realizing whole-process linkage from data acquisition, digital fusion to intelligent evaluation and early warning. The method solves the problems of data redundancy, interaction difficulty and early warning lag in existing monitoring, is suitable for railway infrastructure operation and maintenance scenes, and has great significance in improving the safety and stability of a railway traffic system.

Description

Technical Field

[0001] The present invention relates to the technical field of railway transportation infrastructure monitoring, and in particular to an IFC-based bridge Internet of Things dynamic visualization monitoring and early warning method for railway bridge structure safety assurance. Background Art

[0002] Traditional railway bridge structural monitoring methods primarily rely on distributed, independent sensors to collect structural response data such as temperature, strain, and displacement. Data processing is primarily based on manual statistics and offline analysis, resulting in low data exchange efficiency, poor real-time performance, and information silos. Furthermore, these methods generally lack intelligent prediction and proactive early warning mechanisms, making it difficult to promptly identify potential operational risks. This severely impacts the safety management and operational response efficiency of railway bridges during their service life, posing potential safety hazards to railway operations.

[0003] In recent years, with the gradual adoption of BIM and IoT technologies in railway engineering, some railway bridge monitoring systems have initially achieved the integration and real-time visualization of multi-source sensor data. However, practical applications still face several technical bottlenecks: insufficient dynamic fusion capabilities of IoT data streams, inadequate real-time data-driven analysis and update mechanisms, limited intelligent early warning capabilities, and insufficiently intuitive visualization. As a result, the overall system struggles to meet the safety and security requirements of railway bridges in the high-intensity, high-stability operating environment.

[0004] One of the root causes of these issues is that the existing IFC (Industry Foundation Classes) standard has significant deficiencies in the modeling and semantic extension of railway bridge sensor data. This inability to uniformly represent the real-time data characteristics of different sensor types results in inefficient integration between bridge BIM models and IoT sensor data, making it difficult to support the precise management and timely analysis of dynamic monitoring data. This not only leads to the accumulation of redundant data and wastes computing resources, but also weakens the system's ability to respond to critical structural conditions, limiting the accuracy and real-time nature of early warning functions.

[0005] In addition, the prediction and early warning algorithms currently widely used in the field of railway bridges are mostly based on traditional neural network models. When faced with complex situations such as high coupling of structural response data with environmental factors, multi-source heterogeneity and nonlinear dynamic changes, they often have problems such as slow convergence, insufficient prediction accuracy, and poor model generalization ability. It is difficult to achieve highly reliable, low-latency dynamic perception and risk prediction of the operating status of key bridge components, and it is difficult to meet the operational safety assurance needs of railway bridges under all-weather, high-load and multiple working conditions. Summary of the Invention

[0006] The present invention provides an IFC-based bridge Internet of Things dynamic visualization monitoring and early warning method to overcome the above technical problems.

[0007] In order to achieve the above object, the technical solution of the present invention is: A bridge Internet of Things dynamic visualization monitoring and early warning method based on IFC, including: S1: Deploy different types of sensors at various structural parts of the railway bridge to obtain sensor monitoring data corresponding to each structural part of the railway bridge; S2: Obtain the original IFC dictionary file and perform entity expansion on the deployed sensors. The entity expansion is to add a railway bridge sensor type to the original IFC dictionary file to obtain a railway bridge sensor type IFC dictionary file, and then generate a railway bridge sensor type IFC file based on the railway bridge sensor IFC dictionary file. S3: Expanding the attribute set of the railway bridge sensor type, wherein the attribute set expansion is to add sensor data attributes corresponding to the railway bridge sensor type to the railway bridge sensor type IFC file to obtain a basic railway bridge sensor data IFC file; S4: Introduce the BIM platform, input the sensor monitoring data into the basic railway bridge sensor data IFC file based on the BIM platform, and update it in real time to obtain the railway bridge sensor data IFC file; S5: Design a custom activation function, introduce the tanh activation function, gating mechanism, and cell state. Based on the gating mechanism, cell state, custom activation function, and tanh activation function, build an adaptive prediction model for railway bridge monitoring data characteristics. Use sensor monitoring data to train and optimize the parameters of the adaptive prediction model, forming a multi-factor prediction model with time series perception and feature response capabilities. S6: Input the current sensor monitoring data into the multi-factor prediction model to predict the sensor monitoring data for a period of time in the future. Synchronously update the sensor monitoring data predicted by the multi-factor prediction model into the railway bridge sensor data IFC file to dynamically evaluate the structural health status and potential risks of the railway bridge. S7: Based on the updated IFC file of railway bridge sensor data, the system identifies railway bridge structural parts with potential risks, analyzes them in combination with real-time monitoring data, and conducts visual monitoring and intelligent early warning of railway bridges. The early warning results are pushed to the maintenance management platform and linked with the railway dispatching and command system and bridge maintenance system for control, thereby improving the structural safety assurance capability of railway bridges during operation and the operational stability of the railway transportation system.

[0008] Furthermore, the adaptive prediction model includes a feature introduction control unit, a state screening adjustment unit, a prediction derivation adjustment unit and a storage unit; The feature introduction control unit, state screening adjustment unit, prediction derivation adjustment unit and storage unit all contain weight matrices and bias items; the bias items are used to adjust the activation thresholds in the corresponding units and control the default flow tendency of information; The weight matrix of the storage unit is used to map the current sensor monitoring data and the predicted data output by the previous prediction and adjustment unit into a high-dimensional space to extract the nonlinear relationship between features; The weight matrix of the feature introduction control unit is used to determine the contribution of the output data of the previous moment and the current input data to the updated new data; the output data of the previous moment is the predicted data output by the prediction and derivation adjustment unit at the previous moment, and the current input data is the current sensor monitoring data; The output data of the previous moment and the current sensor monitoring data are input into a feature introduction control unit, and the processed sensor monitoring data are sorted in chronological order by the feature introduction control unit to form a multi-source time series input vector that can be recognized by the adaptive prediction model. The input data is processed based on the weight matrix and bias term of the feature introduction control unit and the storage unit using a custom activation function and a tanh activation function to obtain the output data of the feature introduction control unit and the state information of the candidate storage unit; the output data is a gating signal of the feature introduction control unit, and the state information of the storage unit is a nonlinear fusion representation of the output data of the previous moment and the current sensor monitoring data; The weight matrix of the state screening adjustment unit is used to determine the contribution of the output data of the previous moment and the current sensor monitoring data to forgetting old information; The output data of the previous moment and the current sensor monitoring data are input into the state screening and adjustment unit, and the current sensor monitoring data and the output data of the previous moment are processed based on the weight matrix and bias term of the state screening and adjustment unit using a custom activation function to obtain the output data of the state screening and adjustment unit, and the state information of the storage unit is recursively and updated based on the output data of the state screening and adjustment unit, the state information of the storage unit at the previous moment, the output data of the feature introduction control unit, and the state information of the candidate storage unit to obtain the updated state information of the storage unit at the current moment; The weight matrix of the prediction and export adjustment unit is used to determine the contribution of the output data of the previous moment and the current input data to the output information; Inputting the currently input sensor monitoring data and the output data at the previous moment into the prediction and derivation regulation unit, and using a custom activation function to process the currently input sensor monitoring data and the output data at the previous moment based on the weight matrix and bias term of the prediction and derivation regulation unit to obtain the output data of the prediction and derivation regulation unit; the output data of the prediction and derivation regulation unit is the gating signal of the prediction and derivation regulation unit; The tanh activation function is used to generate the corresponding predicted output based on the output data of the predicted deriving adjustment unit and the state information of the updated storage unit.

[0009] Furthermore, designing custom activation functions includes: Design a custom activation function. The custom activation function is shown in formulas (1) and (2). (1) (2) In formula (1), Represents the input sensor monitoring data; Represents the time parameter, which is used to simulate the changes of railway bridges at different times; Indicates environmental factors; These are all adjustable parameters of LSTM. represents the input weight, where Control the degree of influence of input sensor monitoring data on the custom activation function; Adjust time parameters Periodic effects on model outputs; Control the impact of environmental factors on sensor monitoring data; To adjust the amplitude of periodic changes; To adjust the weights of other nonlinear features; is a bias term used to adjust the output of the model; Indicates that the output is the maximum value between the value in the brackets and 0; In formula (2), T is the monitored temperature value; H is the monitored humidity value; W is the monitored wind speed value; To monitor the average values ​​of temperature, humidity and wind speed; To monitor the standard deviation of temperature, humidity and wind speed.

[0010] Furthermore, the sensor monitoring data is input into the adaptive prediction model for training, including: S51. For each time parameter , the sensor monitoring data are input into the feature introduction control unit, the state screening adjustment unit and the prediction export adjustment unit for processing. The specific steps are as follows: S511: Input the current sensor monitoring data and the output data of the previous moment into the feature introduction control unit for processing, as shown in formula (3), (3) in, Represents the output data of the feature introduction control unit, The weight matrix of the control unit is introduced for the feature, Represents the input data at the current moment, is the output data of the previous moment, Introducing bias terms of control units for features; S512: Input the current sensor monitoring data and the output data of the previous moment into the state screening and adjustment unit for processing, as shown in formula (4), (4) in, Represents the output data of the state screening and adjustment unit, is the weight matrix of the state screening adjustment unit, is the bias term of the state screening adjustment unit; S513: Input the current sensor monitoring data and the output data of the previous moment into the prediction and derivation adjustment unit for processing, as shown in formula (5), (5) in, Represents the output data of the prediction export adjustment unit, represents the weight matrix of the prediction-derived regulation unit, is the bias term of the state screening adjustment unit; S52, using the tanh activation function, combined with the output data of the feature introduction control unit and the output data of the state screening adjustment unit, update the state parameters of the storage unit at the current moment, as shown in formula (6), (6) in, Indicates the state parameters of the storage unit at the current moment, , indicating the part of storage unit information that needs to be retained at the current moment, represents the weight matrix of the storage unit, Represents the bias term of the storage unit; Using the tanh activation function, combined with the predicted output data of the regulation unit and the state parameters of the storage unit at the current moment, the output data at the current moment is calculated, as shown in formula (7). (7) in, Indicates the output data at the current moment.

[0011] Furthermore, the adaptive prediction model is optimized using sensor monitoring data to form a multi-factor prediction model with time series perception and feature response capabilities, including: S53. Construct an MSE loss function or a cross entropy loss function, use the loss function to calculate the error between the predicted output and the true value, calculate the gradient of the adjustable parameters of the custom activation function through the back propagation algorithm, and use the Adam optimization algorithm to update the adjustable parameters of the custom activation function. Continuously iterate during the training process until the error converges to obtain the final adjustable parameters of the custom activation function, and then obtain the final custom activation function. The predicted output of the adaptive prediction model is compared with the actual sensor monitoring data, and the prediction error is calculated. The error is backpropagated from the current moment to each time step through the backpropagation algorithm. The contribution of the weight matrix and bias term of the feature introduction control unit, state screening adjustment unit and prediction export adjustment unit to the error, that is, the error gradient, is calculated. The gradient descent algorithm is used to update the weight matrix and bias term according to the error gradient to reduce the prediction error until the prediction error of the adaptive prediction model converges to a relatively stable state, and the final weight matrix and bias term are obtained, and then the final multi-factor prediction model is obtained.

[0012] Furthermore, a BIM platform was introduced to input sensor monitoring data into the basic railway bridge sensor data IFC file based on the BIM platform and perform real-time updates, including: S41, performing data cleaning on the sensor monitoring data and converting the cleaned sensor monitoring data into an IFC standard format; S42, dynamically binding attributes for the IFC file, that is, using the IFC attribute mapping template, one-to-one mapping the converted sensor monitoring data with the railway bridge sensor entity type defined in IFC, and then writing the converted sensor monitoring data into the corresponding sensor data attributes to achieve dynamic binding; S43. Implement bidirectional synchronization of basic railway bridge sensor data IFC files and converted sensor monitoring data based on a publish-subscribe model.

[0013] Furthermore, the original IFC dictionary file is obtained, and entity expansion is performed on the deployed sensors. The entity expansion is to add the railway bridge sensor type to the original IFC dictionary file to obtain the railway bridge sensor type IFC dictionary file. The railway bridge sensor type IFC file is formed according to the railway bridge sensor IFC dictionary file, including: S21, obtaining the original IFC dictionary file; S22, based on the IFC standard, define the sensor space in the original IFC dictionary; S23, performing entity expansion on the sensor, adding structural elements of each type of sensor into the sensor space, and obtaining an IFC dictionary file of the railway bridge sensor type; the IFC dictionary file of the railway bridge sensor type includes key information and content of the sensor; S24. Generate a railway bridge sensor type IFC file according to the required sensor content in the railway bridge sensor type IFC dictionary file.

[0014] Furthermore, the attribute set of the railway bridge sensor type is expanded. The attribute set expansion is to add sensor data attributes corresponding to the railway bridge sensor type to the railway bridge sensor type IFC file, thereby obtaining a basic railway bridge sensor data IFC file, including: S31, read the railway bridge sensor type IFC file, and select the specified railway bridge sensor type; S32. Expand the attribute set of the railway bridge sensor type, that is, add the sensor data attributes to the corresponding railway bridge sensor type to obtain a basic railway bridge sensor data IFC file.

[0015] Beneficial effects: The present invention proposes a dynamic visualization monitoring and early warning method for bridge IoT based on IFC, which integrates the digital extension mechanism of the IFC standard and the adaptive prediction model for railway bridge structural health prediction, and has good technical practicality and engineering adaptability. By expanding the entity and attribute set of the IFC file structure, the unified expression and dynamic update of railway bridge sensor monitoring data are achieved within the standard semantic framework, providing a structured data carrier support for the integrated management and visualization analysis of monitoring information; the adaptive prediction model introduces an independently designed activation function, which can integrate multi-source heterogeneous feature information such as structural response, environmental load and time change, and improve the recognition ability and prediction accuracy of periodic fluctuations and sudden abnormal changes; the model output results are synchronously written into the IFC file, and the BIM platform is linked to realize the visualization display, forming a closed-loop logic mechanism of monitoring-prediction-update-early warning, which effectively improves the information perception and dynamic expression ability of the health status of the railway bridge structure. Based on the model judgment results, the system can also link the railway dispatching and operation and maintenance platform, trigger the risk warning response mechanism and push high-risk component information, provide data support for train speed limits, on-site inspections and maintenance interventions, significantly improve the safety assurance capabilities during railway bridge operations and the operational stability of the railway transportation system, and have good prospects for promotion and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0017] Figure 1A flow chart of a bridge IoT dynamic visualization monitoring and early warning method based on IFC provided by the present invention; Figure 2 This is the railway bridge Internet of Things monitoring and collection flow chart of the present invention; Figure 3 EXPRESS-G diagram of the spatial structure of sensors for IoT monitoring of railway bridges; Figure 4 EXPRESS-G diagram of physical elements of sensors for IoT monitoring of railway bridges; Figure 5 This is a diagram of the neural network structure based on a custom activation function; Figure 6 This is a visualization diagram of dynamic warning. DETAILED DESCRIPTION

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] This embodiment provides a bridge IoT dynamic visualization monitoring and early warning method based on IFC, such as Figure 1 Shown, including: S1: Deploy different types of sensors at various structural parts of the railway bridge to obtain sensor monitoring data corresponding to each structural part of the railway bridge; S2: Obtain the original IFC dictionary file and perform entity expansion on the deployed sensors. The entity expansion is to add a railway bridge sensor type to the original IFC dictionary file to obtain a railway bridge sensor type IFC dictionary file, and then generate a railway bridge sensor type IFC file based on the railway bridge sensor IFC dictionary file. S3: Expanding the attribute set of the railway bridge sensor type, wherein the attribute set expansion is to add sensor data attributes corresponding to the railway bridge sensor type to the railway bridge sensor type IFC file to obtain a basic railway bridge sensor data IFC file; S4: Introduce the BIM platform, input the sensor monitoring data into the basic railway bridge sensor data IFC file based on the BIM platform, and update it in real time to obtain the railway bridge sensor data IFC file; S5: Design a custom activation function, introduce the tanh activation function, gating mechanism, and cell state. Based on the gating mechanism, cell state, custom activation function, and tanh activation function, build an adaptive prediction model for railway bridge monitoring data characteristics. Use sensor monitoring data to train and optimize the parameters of the adaptive prediction model, forming a multi-factor prediction model with time series perception and feature response capabilities. S6: Input the current sensor monitoring data into the multi-factor prediction model to predict the sensor monitoring data for a period of time in the future. Synchronously update the sensor monitoring data predicted by the multi-factor prediction model into the railway bridge sensor data IFC file to dynamically evaluate the structural health status and potential risks of the railway bridge. S7: Based on the updated IFC file of railway bridge sensor data, the system identifies railway bridge structural parts with potential risks, analyzes them in combination with real-time monitoring data, and conducts visual monitoring and intelligent early warning of railway bridges. The early warning results are pushed to the maintenance management platform and linked with the railway dispatching and command system and bridge maintenance system for control, thereby improving the structural safety assurance capability of railway bridges during operation and the operational stability of the railway transportation system.

[0020] Specifically, different types of sensors are first deployed at various structural parts of railway bridges to obtain sensor monitoring data corresponding to each structural part of the railway bridge. Multiple sensors are deployed at various locations on the railway bridge. Through real-time, multi-dimensional data collection and analysis, comprehensive perception of the railway bridge structure status, early risk warning and scientific management are achieved, providing data support for subsequent visual warning of railway bridges. Secondly, the original IFC dictionary file is obtained, and the deployed sensors are entity expanded. The entity expansion is to add the railway bridge sensor type to the original IFC dictionary file to obtain the railway bridge sensor type IFC dictionary file. The railway bridge sensor type IFC file is formed according to the railway bridge sensor IFC dictionary file, and the attribute set of the railway bridge sensor type is expanded. The attribute set expansion is to add the sensor data attributes corresponding to the railway bridge sensor type to the railway bridge sensor type IFC file to obtain the basic railway bridge sensor data IFC file, introduce the BIM platform, and input the sensor monitoring data into the basic railway bridge sensor data IFC file based on the BIM platform, and update it in real time to obtain the railway bridge sensor data IFC file. By expanding the IFC standard and introducing dynamic data flow integration and real-time update mechanisms, using IFC as the medium, the Internet of Things data can be deeply integrated with the BIM model, and the monitoring data can be updated in real time and dynamically analyzed; Thirdly, a custom activation function is designed, and the tanh activation function, gating mechanism and cell state are introduced. Based on the gating mechanism, cell state, custom activation function and tanh activation function, an adaptive prediction model for the characteristics of railway bridge monitoring data is constructed. The adaptive prediction model is trained and parameter optimization is carried out using sensor monitoring data to form a multi-factor prediction model with time series perception and feature response capabilities. A custom activation function is designed to process the multi-dimensional data of railway bridges and the environment to form a multi-factor prediction model. This significantly improves the modeling and expression capabilities of nonlinear, multi-source heterogeneous monitoring information, and has the ability to automatically adjust the weights of multiple factors based on data distribution. This improves the stability, robustness and response timeliness of the model under high-frequency dynamic loads and operational disturbance conditions, providing an intelligent, explainable and engineering-practical prediction decision support method for railway bridge operation safety. Finally, the current sensor monitoring data is input into the multi-factor prediction model to predict the sensor monitoring data for a period of time in the future. The sensor monitoring data predicted by the multi-factor prediction model is synchronously updated to the railway bridge sensor data IFC file, and the structural health status and potential risks of the railway bridge are dynamically evaluated. Based on the updated railway bridge sensor data IFC file, the railway bridge structural parts with potential risks are identified, and combined with real-time monitoring data for analysis, visual monitoring and intelligent early warning of the railway bridge are carried out; the early warning results are pushed to the maintenance management platform, and linked control is achieved with the railway dispatching command system and the bridge maintenance system to improve the structural safety protection capability of the railway bridge during operation and the operational stability of the railway transportation system.

[0021] In a specific embodiment, different types of sensors are deployed at various structural parts of a railway bridge, and a solution for obtaining sensor monitoring data corresponding to each structural part of the railway bridge is as follows: In this embodiment, various types of sensors are deployed on key structural parts of railway bridges in a distributed manner. The Internet of Things monitoring system issues instructions to various collection modules to receive monitoring data collected by the sensors on various structural parts of the railway bridges. The specific steps are as follows: S11. Identify and analyze the key structural parts of the railway bridge on site and determine the main structural parts that need to be monitored, such as the bridge deck, piers, bridge supports, inclined cables, box girder webs, and abutments; S12. Select appropriate sensor types based on the monitoring requirements of each part. For example, if the bridge deck bears a large load, displacement sensors and strain sensors should be deployed. Tension sensors should be deployed on the cable-stayed part. The deployed sensors should be connected to the sensor acquisition module to collect data. S13. Perform intranet penetration on the client of the IoT monitoring system, so that the local client can be remotely accessed and connected. The IoT monitoring system client sends a collection instruction through the operator's wireless network. After the collection instruction is delivered to GPRS, GPRS transmits the corresponding collection instruction to the on-site collection module; S14. After receiving the instruction, the corresponding acquisition module collects data from various sensors on site. After the acquisition is completed, the sensor transmits the collected electrical signal to the acquisition module; the acquisition module performs signal conversion inside, converts the obtained electrical signal into hexadecimal coded data, and then sends it back to GPRS through the RS485 wired communication protocol. Finally, GPRS sends it back to the client of the monitoring system, that is, the client receives the data collected by various sensors. The railway bridge Internet of Things monitoring and acquisition process is as follows: Figure 2 As shown; In this embodiment, the collected data types are divided into the following categories: 1. Load and stress data: This includes monitoring data such as the stress of railway bridges under load, which is intended to assess the stress conditions and bearing capacity of railway bridge structures; 2. Structural and functional data: This data reflects the dynamic response of railway bridges during use, including displacement, strain, acceleration, vibration, and other indicators. By dynamically monitoring displacement, deformation, acceleration, and other responses, it is possible to determine whether the railway bridge has experienced abnormal deformation or vibration, thereby promptly identifying potential safety hazards. 3. Environmental impact data: This data reflects the impact of external climate and environmental factors on railway bridge structures, including temperature, humidity, wind speed, and other factors. It helps analyze the impact of environmental changes on the expansion and contraction of railway bridge materials (such as concrete and steel). 3. Monitoring time data: including the time and change trend of railway bridge monitoring data, used to record the monitoring status of railway bridge sensors; 4. Sensor identification data: This includes the sensor number, sensor model, acquisition frequency, and acquisition period. This type of data enables accurate tracking and positioning of each sensor, ensuring that the location and status of each sensor can be clearly identified during maintenance, acquisition, or management.

[0022] In this plan, multiple sensors are deployed at various locations on railway bridges. Through real-time, multi-dimensional data collection and analysis, comprehensive perception of the structural status of railway bridges, early risk warning and scientific management are achieved, providing data support for subsequent visual warnings of railway bridges.

[0023] In a specific embodiment, an original IFC dictionary file is obtained, and entity expansion is performed on the deployed sensors. The entity expansion is to add a railway bridge sensor type to the original IFC dictionary file to obtain a railway bridge sensor type IFC dictionary file. A scheme for forming a railway bridge sensor type IFC file based on the railway bridge sensor IFC dictionary file is: S21, obtaining the original IFC dictionary file; S22, based on the IFC standard, define the sensor space in the original IFC dictionary; Specifically, if Figure 3 As shown, Figure 3 This is a schematic diagram of the specific entities of IoT sensors and their inheritance relationships in the IFC model framework. The solid-line boxes in the figure represent some entities in the existing IFC, and the dashed-line boxes represent newly expanded entities and their inheritance relationships in the IFC model framework: First, under the IFC spatial structure class (IfcSpatialStructuralElement), a new IfcSensorSpace class is added as the spatial entity of the IoT sensor. This class defines the geometric position and spatial properties of the space where the IoT monitoring sensor of the railway bridge is located, that is, the spatial unit of the sensor installation area. Each IfcSensorSpace instance represents a specific spatial position, representing the sensor installation area of ​​a certain railway bridge structure. The spatial entity enumeration type (IfcSensorSpaceTypeEnum) of the IoT monitoring sensor of the railway bridge is added through the predefined type attribute definition method (PreDefinedType), from which the strain sensor space (IfcStrainSensorSpace), temperature sensor space (IfcTemperatureSensorSpace), displacement sensor space (IfcSensorSpaceEnum) are derived. Sensor space (IfcDisplacementSensorSpace), humidity sensor space (IfcHumiditySensorSpace), wind speed sensor space (IfcWindSpeedSensorSpace), vibration sensor space (IfcVibrationSensorSpace), stress sensor space (IfcStressSensorSpace), track acceleration sensor space (IfcTrackAccelerationSensor), rail temperature sensor space (IfcRailThermometer), wheel-rail force sensor space (IfcWheelRailForceSensor), axle counter sensor space (IfcAxleCounterSensor), and cable force sensor space (IfcAnchorStressSensor); S23. Perform entity expansion on the sensor, add structural elements of each type of sensor into the sensor space, and obtain an IFC dictionary file of the railway bridge sensor type; the IFC dictionary file of the railway bridge sensor type includes key information and content of the sensor; the key information includes but is not limited to the sensor model, sensor serial number, strain measurement range of the sensor, temperature measurement range of the sensor, strain measurement accuracy of the sensor, temperature measurement accuracy of the sensor, temperature correction coefficient of the sensor, installation time of the sensor, and service life of the sensor; Specifically, first determine the name of the new sensor entity class and the inheritance relationship of the new entity in the IFC model framework based on the type of the required railway bridge IoT monitoring sensor, and complete the IFC entity customization of the new railway bridge IoT monitoring sensor; The newly added railway bridge IoT monitoring sensor entity is added to the parent object of the defined entity according to its inheritance relationship in the IFC model framework. At the same time, based on the type of each sensor, its attribute set and attributes are expanded in the form of attribute set expansion, so that the corresponding sensor entity contains the various attributes shown in Table 1. Table 1 Attributes contained in each sensor entity ; S24. Generate a railway bridge sensor type IFC file according to the required sensor content in the railway bridge sensor IFC dictionary file.

[0024] In this solution, sensor entities are extended based on the IFC (Industry Foundation Classes) standard, and an IFC dictionary file for railway bridge sensor types is generated. This enables standardized mapping from physical sensors to digital models, facilitating subsequent data access.

[0025] In a specific embodiment, the attribute set of the railway bridge sensor type is expanded. The attribute set expansion is to add sensor data attributes corresponding to the railway bridge sensor type in the railway bridge sensor type IFC file. The solution for obtaining the basic railway bridge sensor data IFC file is: S31, read the railway bridge sensor type IFC file, and select the specified railway bridge sensor type; S32. Expand the attribute set of the railway bridge sensor type, that is, add the sensor data attributes to the corresponding railway bridge sensor type to obtain a basic railway bridge sensor data IFC file; Specifically, Figure 4 As shown, Figure 4It is a physical element diagram of the railway bridge Internet of Things monitoring sensor. Under the general category of physical structure elements (IfcElement) of the IFC model, add the entity civil engineering field element (IfcCivilElement) to make it at the same level as the building element entity (IfcBuildingElement). Then, based on IfcCivilElement, the railway bridge Internet of Things monitoring sensor physical element abstract entity (IfcSensorElement) is expanded downward as the parent class of all entities of the railway bridge Internet of Things monitoring sensor, that is, the railway bridge monitoring sensor class. Under IfcSensorElement, the strain sensor (IfcStrainSensor), temperature sensor (IfcTemperatureSensor), displacement sensor (IfcDisplacementSensor) are derived. The physical elements of the railway bridge IoT monitoring sensors are described as follows: humidity sensor (IfcHumiditySensor), wind speed sensor (IfcWindSpeedSensor), vibration sensor (IfcVibrationSensor), stress sensor (IfcStressSensor), track acceleration sensor (IfcTrackAccelerationSensor), rail temperature sensor (IfcRailThermometer), wheel-rail force sensor (IfcWheelRailForceSensor), axle counter sensor (IfcAxleCounterSensor), and cable force sensor (IfcAnchorStressSensor); Next, we add a specific attribute set (SensorData) to each sensor entity type. This is the enumeration type for the sensor in the diagram. This attribute set contains the sensor's real-time measurement data and some additional monitoring information. This attribute set includes fields such as CurrentReading (current reading), Timestamp (timestamp), and SensorStatus (sensor operating status). The IFC file also sets a unique identifier for the railway bridge equipped with the sensor to prepare for subsequent visualization.

[0026] In this solution, the attribute set of railway bridge sensor types is expanded to enable intercommunication with BIM and monitoring systems, facilitating subsequent data updates and presentations.

[0027] In a specific embodiment, a BIM platform is introduced, and sensor monitoring data is input into a basic railway bridge sensor data IFC file based on the BIM platform, and updated in real time. The solution for obtaining the railway bridge sensor data IFC file is: S41, performing data cleaning on the sensor monitoring data and converting the cleaned sensor monitoring data into an IFC standard format; Specifically, the data is first cleaned and outliers (such as sensor drift and signal noise) are filtered through the set API interface. Then, the data is normalized and filtered using the set API interface. That is, the collected raw data (such as strain values, temperature and humidity values, etc.) are converted into structured data that can be recognized by IFC files through a set of predefined formats; S42, dynamically binding attributes for the IFC file, that is, using the IFC attribute mapping template, one-to-one mapping the converted sensor monitoring data with the railway bridge sensor entity type defined in IFC, and then writing the converted sensor monitoring data into the corresponding sensor data attributes to achieve dynamic binding; Specifically, the IFC file is dynamically bound to attributes. That is, through the IFC attribute mapping template, the cleaned sensor monitoring data is dynamically bound to the sensor entity attributes (CurrentReading, Timestamp, SensorStatus) defined in the IFC. For example, the normalized strain value is written to the CurrentReading attribute of the corresponding sensor entity in real time, and the timestamp and status identifier are updated at the same time. S43, implementing bidirectional synchronization of basic railway bridge sensor data IFC files and converted sensor monitoring data based on a publish-subscribe model; Specifically, by automatically synchronizing and binding the SQL database of the IoT monitoring sensors on railway bridges, changes in sensor data can be monitored. Dynamic updates of the monitoring data and IFC files can be triggered based on a publish-subscribe model. That is, when sensor data changes, an IFC model update event is triggered. Data change events are broadcast to the IFC parsing engine via a message queue (such as RabbitMQ). The engine locates the corresponding railway bridge IoT monitoring sensor entities based on the sensor ID, calls the IfcOpenShell API interface to update the entity attributes, and generates an incremental version of the IFC file (Delta IFC). Based on bidirectional synchronization technology, data consistency between the sensor database and the IFC file is ensured. When the database record is updated, the IFC update service is automatically called through a trigger. Conversely, manual modifications to the IFC file (such as the adjustment of a certain attribute value) can also be synchronized back to the database.

[0028] In this solution, by introducing dynamic data stream integration and real-time update mechanisms, the expanded IFC model can not only represent static structural information, but also support data updates and real-time analysis of the railway bridge monitoring system through real-time data streams, making up for the shortcomings of the existing IFC standard in the dynamic update and streaming data processing capabilities of IoT device monitoring data.

[0029] In a specific embodiment, a custom activation function is designed, and a tanh activation function, a gating mechanism, and a cell state are introduced. An adaptive prediction model for railway bridge monitoring data characteristics is constructed based on the gating mechanism, the cell state, the custom activation function, and the tanh activation function. The adaptive prediction model is trained and parameterized using sensor monitoring data to form a multi-factor prediction model with time series perception and feature response capabilities. The solution is: like Figure 5 As shown, the adaptive prediction model includes a feature introduction control unit, a state screening adjustment unit, a prediction derivation adjustment unit and a storage unit; The feature introduction control unit, state screening adjustment unit, prediction derivation adjustment unit and storage unit all contain weight matrices and bias items; the bias items are used to adjust the activation thresholds in the corresponding units and control the default flow tendency of information; The weight matrix of the storage unit is used to map the current sensor monitoring data and the predicted data output by the previous prediction and adjustment unit into a high-dimensional space to extract the nonlinear relationship between features; The weight matrix of the feature introduction control unit is used to determine the contribution of the output data of the previous moment and the current input data to the updated new data; the output data of the previous moment is the predicted data output by the prediction and derivation adjustment unit at the previous moment, and the current input data is the current sensor monitoring data; The output data of the previous moment and the current sensor monitoring data are input into a feature introduction control unit, and the processed sensor monitoring data are sorted in chronological order by the feature introduction control unit to form a multi-source time series input vector that can be recognized by the adaptive prediction model. The input data is processed based on the weight matrix and bias term of the feature introduction control unit and the storage unit using a custom activation function and a tanh activation function to obtain the output data of the feature introduction control unit and the state information of the candidate storage unit; the output data is a gating signal of the feature introduction control unit, and the state information of the storage unit is a nonlinear fusion representation of the output data of the previous moment and the current sensor monitoring data; The weight matrix of the state screening adjustment unit is used to determine the contribution of the output data of the previous moment and the current sensor monitoring data to forgetting old information; The output data of the previous moment and the current sensor monitoring data are input into the state screening and adjustment unit, and the current sensor monitoring data and the output data of the previous moment are processed based on the weight matrix and bias term of the state screening and adjustment unit using a custom activation function to obtain the output data of the state screening and adjustment unit, and the state information of the storage unit is recursively and updated based on the output data of the state screening and adjustment unit, the state information of the storage unit at the previous moment, the output data of the feature introduction control unit, and the state information of the candidate storage unit to obtain the updated state information of the storage unit at the current moment; The weight matrix of the prediction and export adjustment unit is used to determine the contribution of the output data of the previous moment and the current input data to the output information; Inputting the currently input sensor monitoring data and the output data at the previous moment into the prediction and derivation regulation unit, and using a custom activation function to process the currently input sensor monitoring data and the output data at the previous moment based on the weight matrix and bias term of the prediction and derivation regulation unit to obtain the output data of the prediction and derivation regulation unit; the output data of the prediction and derivation regulation unit is the gating signal of the prediction and derivation regulation unit; Use the tanh activation function to generate the corresponding predicted output based on the output data of the predicted deriving adjustment unit and the updated state information of the storage unit; Design a custom activation function. The custom activation function is shown in formulas (8) and (9). (8) (9) In formula (8), Represents the input sensor monitoring data; Represents the time parameter, which is used to simulate the changes of railway bridges at different times; Indicates environmental factors; These are all adjustable parameters of LSTM. represents the input weight, where Control the degree of influence of input sensor monitoring data on the custom activation function; Adjust time parameters Periodic effects on model outputs; Control the impact of environmental factors on sensor monitoring data; To adjust the amplitude of periodic changes; To adjust the weights of other nonlinear features; is a bias term used to adjust the output of the model; Indicates that the output is the maximum value between the value in the brackets and 0; In formula (9), T is the monitored temperature value; H is the monitored humidity value; W is the monitored wind speed value; To monitor the average values ​​of temperature, humidity and wind speed; To monitor the standard deviation of temperature, humidity and wind speed; S51. For each time parameter , the sensor monitoring data are input into the feature introduction control unit, the state screening adjustment unit and the prediction export adjustment unit for processing. The specific steps are as follows: S511: Input the current sensor monitoring data and the output data of the previous moment into the feature introduction control unit for processing, as shown in formula (10), (10) in, Represents the output data of the feature introduction control unit, The weight matrix of the control unit is introduced for the feature, Represents the input data at the current moment, is the output data of the previous moment, Introducing bias terms of control units for features; S512: Input the current sensor monitoring data and the output data of the previous moment into the state screening and adjustment unit for processing, as shown in formula (11), (11) in, Represents the output data of the state screening and adjustment unit, is the weight matrix of the state screening adjustment unit, is the bias term of the state screening adjustment unit; S513: Input the current sensor monitoring data and the output data of the previous moment into the prediction and derivation adjustment unit for processing, as shown in formula (12), (12) in, Represents the output data of the prediction export adjustment unit, represents the weight matrix of the prediction-derived regulation unit, is the bias term of the state screening adjustment unit; S52, using the tanh activation function, combined with the output data of the feature introduction control unit and the output data of the state screening adjustment unit, update the state parameters of the storage unit at the current moment, as shown in formula (13), (13) in, Indicates the state parameters of the storage unit at the current moment, , indicating the part of storage unit information that needs to be retained at the current moment, represents the weight matrix of the storage unit, Represents the bias term of the storage unit; Using the tanh activation function, combined with the predicted output data of the regulation unit and the state parameters of the storage unit at the current moment, the output data at the current moment is calculated, as shown in formula (14). (14) in, Represents the output data at the current moment; S53. Construct an MSE loss function or a cross entropy loss function, use the loss function to calculate the error between the predicted output and the true value, calculate the gradient of the adjustable parameters of the custom activation function through the back propagation algorithm, and use the Adam optimization algorithm to update the adjustable parameters of the custom activation function. Continuously iterate during the training process until the error converges to obtain the final adjustable parameters of the custom activation function, and then obtain the final custom activation function. The predicted output of the adaptive prediction model is compared with the actual sensor monitoring data, and the prediction error is calculated. The error is backpropagated from the current moment to each time step through the backpropagation algorithm. The contribution of the weight matrix and bias term of the feature introduction control unit, state screening adjustment unit and prediction export adjustment unit to the error, that is, the error gradient, is calculated. The gradient descent algorithm is used to update the weight matrix and bias term according to the error gradient to reduce the prediction error until the prediction error of the adaptive prediction model converges to a relatively stable state, and the final weight matrix and bias term are obtained, and then the final multi-factor prediction model is obtained.

[0030] The adaptive prediction model for railway bridges proposed in this proposal fully incorporates the actual characteristics of railway bridge structural health monitoring data during the modeling process. This is based on two main premises: First, the monitoring data comes from diverse sources, covering structural response information such as strain, displacement, and cable tension, as well as environmental factors such as temperature, humidity, and wind speed, and there is a significant nonlinear coupling relationship between the data; second, the multi-source monitoring data exhibits dynamic characteristics of periodic fluctuations and sudden anomalies over time, making it difficult for conventional prediction methods to achieve efficient and accurate responses in such situations; To address these issues, the adaptive prediction model has been specifically designed with respect to its network structure and feature fusion mechanism. It is capable of integrating time series features, structural response information, and environmental interference factors. Through a learnable weighted control mechanism, it enables adaptive modeling and dynamic adjustment of multi-factor inputs, thereby constructing a high-precision prediction model suitable for railway bridge scenarios. Specifically, this model has the following beneficial effects: First, it significantly improves the modeling and expression capabilities of nonlinear, multi-source heterogeneous monitoring information, and enhances the sensitivity of identifying periodic operational fluctuations (such as temperature-stress response) and sudden structural anomalies (such as pier settlement and sudden displacement changes at beam ends), providing reliable support for early fault diagnosis and hidden danger warning of railway bridges. Second, it has the ability to automatically adjust the weights of multiple factors based on data distribution, effectively overcoming the problem of insufficient adaptability of traditional models to complex environments, and significantly enhancing the prediction accuracy and generalization performance of bridge structural behavior under multiple working conditions and multiple load conditions; Third, by introducing a dynamic weight adjustment mechanism, a coupled response relationship between structural state, time factors and environmental interference is constructed, which improves the stability, robustness and response timeliness of the model under high-frequency dynamic loads and operational disturbance conditions, and provides an intelligent, explainable and engineering-practical prediction and decision-making support means for the safe operation of railway bridges.

[0031] In a specific embodiment, the current sensor monitoring data is input into a multi-factor prediction model to predict the sensor monitoring data for a period of time in the future. The sensor monitoring data predicted by the multi-factor prediction model is synchronously updated to the railway bridge sensor data IFC file. The solution for dynamically evaluating the structural health status and potential risks of the railway bridge is: S61. Define safety thresholds (such as acceleration, displacement, and stress) for each railway bridge structural component (e.g., main beams, piers, and supports). These thresholds can be set based on railway bridge design specifications, historical monitoring data, and expert experience. Threshold data is embedded in the extended attributes of the railway bridge sensor data IFC file and dynamically linked to the predicted data. S62. Compare the preset threshold data with the predicted sensor monitoring data updated in the railway bridge sensor data IFC file. The comparison adopts a multi-level threshold classification strategy and is divided into three levels: normal, warning, and danger: When the predicted data is below the safety threshold but below the danger threshold, it is considered normal; when it exceeds the safety threshold but is below the danger threshold, it is considered a warning; if it exceeds the danger threshold, it is considered dangerous. The comparison results will be stored in the railway bridge sensor data IFC file as an attribute set, and corresponding warning level attributes will be added to each structural part for subsequent visualization and management. S63. Use the globally unique identifiers in the IFC file of the railway bridge sensor data to locate and compare each railway bridge component. Different risk levels are color-coded using different rules: normal status corresponds to green, warning status corresponds to yellow, and dangerous status corresponds to red. The risk status of each part of the railway bridge is intuitively reflected on the visualization platform. The specific operations are as follows: Modify the RGB value under the IfcSurfaceStyleRendering property under the IfcMaterial entity in the IFC file. The color coding strategy is: Normal (green): RGB (0, 255, 0); Warning (yellow): RGB (255, 255, 0); Danger (red): RGB (255, 0, 0); S64. After adjusting the RGB values, add additional attributes of the risk category in the form of an attribute set in the IFC file of the railway bridge sensor data to indicate the specific risk source of the component, such as stress exceeding the limit, abnormal displacement, or vibration exceeding the limit, to facilitate analysis and management on the visualization platform; S65. Update the IFC file of railway bridge sensor data and perform version management to ensure traceability of changes while ensuring data integrity and consistency. S66. Use the BIM platform to read the expanded IFC file of railway bridge sensor data, load and display the railway bridge IoT monitoring data in real time, and dynamically present the prediction results and real-time monitoring data through the 3D visualization interface of the BIM platform. At the same time, the health status and risk warning level of the structure are marked and displayed with set colors, dynamically evaluating the structural health status and potential risks of the railway bridge.

[0032] In a specific embodiment, based on the updated IFC file of railway bridge sensor data, railway bridge structural parts with potential risks are identified, analyzed in combination with real-time monitoring data, and visual monitoring and intelligent early warning are performed on the railway bridge. The early warning results are pushed to the maintenance management platform, and coordinated control is achieved with the railway dispatching and command system and the bridge maintenance system. The solution to improve the structural safety assurance capability of the railway bridge during operation and the operational stability of the railway transportation system is as follows: S71. Automatically update the IFC file of railway bridge sensor data based on real-time monitoring and prediction results to ensure that changes in the bridge structure status are visible in real time; the BIM platform automatically refreshes and updates the results each time it receives new data, realizing continuous monitoring and early warning closed-loop management of the health status of railway bridge structures, such as Figure 6 As shown, it effectively ensures the structural safety and stability of railway bridges during operation and improves the operational reliability and safety of railway transportation systems; S72. Based on the risk levels and structural status information contained in the updated IFC files for railway bridge sensor data in S71, establish linkage interfaces with the railway dispatching and command system and bridge maintenance system, push identified high-risk components and their monitoring data to relevant management platforms, and trigger corresponding safety control strategies based on the warning level, such as speed restrictions, temporary bridge closures, and on-site inspections. By directly applying structural health assessment results to railway traffic dispatching and operation and maintenance decisions, a closed-loop monitoring-warning-response management and control system is implemented, effectively improving the safety assurance capabilities of railway bridges during operation and the overall stability of railway traffic operations.

[0033] To verify the effectiveness of the present invention, a steel box girder cable-stayed bridge with a main span of 620 meters is used as an implementation object. Combined with the operational safety requirements of railway lines, the application process of the technical solution of the present invention in actual engineering and its supporting role in railway traffic safety are elaborated in detail: 1. Sensor deployment and data collection process First, based on the structural system and operational risk characteristics of railway bridges, key monitoring locations were identified. These included structural nodes sensitive to train loads, such as the girder midspan, the cable anchorage area, and the tower tops and support locations. Sensors were deployed based on the mechanical properties of different structural locations, as shown in Table 2.

[0034] Table 2 Sensor layout

[0035] On-site IoT sensors are connected to the field acquisition module via an RS485 bus. The acquisition module, equipped with built-in signal conditioning circuitry, converts analog signals into digital signals, which are then transmitted to the cloud-based monitoring platform via the GPRS wireless network. The data transmission protocol uses Modbus-TCP, ensuring a data update frequency of at least once per second. The collected raw data includes strain values, acceleration, temperature, humidity, and other information from each sensor. Each frame of data is timestamped and stored in a time series database.

[0036] 2. IFC Standard Extension and Dynamic Data Integration Within the IFC4×2 standard framework, two new entities have been added to support the structured description of IoT monitoring data for railway bridges. One is the sensor spatial entity (IfcSensorSpace), which inherits from IfcSpatialStructuralElement and defines the spatial properties of the sensor installation location, including geometric coordinates (X / Y / Z), installation orientation (pitch angle, yaw angle), and structural location (such as main beam, pylon). The other is the sensor physical element entity (IfcSensorElement), which inherits from IfcCivilElement and is further subdivided into subclasses such as strain sensor (IfcStrainSensor) and acceleration sensor (IfcAccelerationSensor). Each entity extends the attribute set SensorData, which includes fields such as current reading (CurrentReading), timestamp (Timestamp), and device status (SensorStatus).

[0037] Establish real-time mapping. That is, when sensor data is updated in the database, the IFC parsing engine (IfcOpenShell) is called through the REST API to locate the corresponding IfcSensorElement entity and update its CurrentReading property, thus achieving automatic synchronization of the sensor's SQL database and binding with the IFC file data.

[0038] 3. Case Verification of Multi-factor Prediction Model Taking the mid-span monitoring of a railway cable-stayed bridge as an example, we selected six consecutive hours of high-frequency data (with a 5-minute interval) to verify the model's predictive ability under the influence of multiple factors. The monitoring data is shown in Table 3 (partial example): Table 3 Original monitoring data (partial examples)

[0039] t = 14 seconds: This is the historical moment. The strain recorded is 300 microstrain. The environmental data are T = 29°C, H = 63%, and W = 11 m / s. t = 15 seconds: At this moment, the strain at the mid-span of the main beam collected by the sensor is 305 microstrains, and the environmental data is updated to T = 30°C, H = 65%, and W = 12 m / s; t = 16 seconds: At the prediction moment, the environmental data obtained through on-site monitoring or forecasting are T = 30.5°C, H = 66%, and W = 12.5 m / s. We use the sensor data at t = 15 seconds and the environmental data at t = 16 seconds to predict the strain value at the next moment.

[0040] Model parameter setting: For ease of explanation, the parameters used in each module are obtained after training and learning the data samples. The specific parameters are as follows: (1) Custom activation function parameters:

[0041] (2) Statistical data (used for normalization of environmental data):

[0042] (3) Feature introduction control unit I: Weight , bias

[0043] (4) State screening and adjustment unit F: Weight , bias

[0044] (5) Prediction and export of regulatory unit O: Weight , bias ; (5) Storage unit: Weight , bias ; Calculate the environmental factors using the environmental data at t=16 seconds:

[0045] Use current input , the time parameter is the predicted time t=16s, and env=5.65 calculated in the previous step;

[0046] Introduce data input features into the control unit for processing:

[0047] The data is input into the state screening and adjustment unit for processing:

[0048] Input data into the forecast export adjustment unit for processing:

[0049] Use the tanh activation function to update the storage unit at the current moment by combining the output data of the feature introduction control unit and the output data of the state screening adjustment unit:

[0050] In the example, to predict the output at the 16th second, the storage unit at the 16th second should essentially be used. To calculate However, due to the actual sensor data The strain at the mid-span of the main beam is usually unknown, so an alternative assumption is often made, that is, the last known value is used. replace , or directly assume that the input has not changed significantly, and use the existing update results to predict the output at the next moment.

[0051] Therefore, the sensor data at t=15 seconds and the historical state (from t=14 seconds) are used to update, and the calculated storage unit is marked as ; Under this simplification, we can Approximately .

[0052] Finally, the formula is used (tanh(357.2) is approximately regarded as 1):

[0053] Through the above complete calculation process, we obtain that the strain in the mid-span of the main beam at the predicted time t=16 seconds is about 290 microstrain.

[0054] Based on this, the mid-span strain of the main beam at t=70 seconds is predicted to be approximately .

[0055] 4. Safety threshold comparison and IFC file attribute adjustment The safety threshold of the mid-span strain of the railway bridge main beam is set as The predicted strain data and safety threshold Perform a comparison. Since the predicted value exceeds the safety threshold, the system determines that the component is at risk. According to the multi-level threshold classification strategy set in the invention (for example: normal, warning, dangerous), here it exceeds the safety threshold but does not reach a higher danger level, it is judged to be in the "warning" state; while updating the attribute set, the IfcSurfaceStyleRendering attribute statement under the IfcMaterial entity corresponding to the component in the IFC file is also adjusted to achieve an intuitive display of the risk status. The component in the warning state is displayed in yellow (RGB value is 255,255,0). At the same time, a specific risk source description is attached to the attribute set, such as "Strain exceeds the limit-predicted value , threshold ”.

[0056] 5. Dynamic visualization and early warning display of BIM platform The updated IFC file is read in real time by the BIM platform to realize the visualization of the dynamic monitoring and early warning of the railway bridge IoT. In this case, the system detects that the strain prediction value of the main beam mid-span at t=70s is , exceeding the safety threshold of railway bridge structures The component's risk level has been marked as "warning" in the IFC file, and its RGB color code has been adjusted to yellow (255, 255, 0). This status indicates a minor overload risk at the girder midspan, potentially impacting train safety. After reading the latest IFC file, the BIM platform dynamically refreshes the 3D model of the railway bridge, highlighting the girder midspan in yellow. This intuitively communicates to operators and maintenance personnel the presence of a minor overload risk, providing visual support for monitoring the bridge's operational safety. Based on the dynamic integration of monitoring data and IFC files, whenever new monitoring data is uploaded and triggers an IFC file update, the model's color and warning information are synchronously adjusted. This allows changes in the bridge's structural health to be promptly reflected on the platform, improving operators' efficiency in identifying and responding to bridge operational anomalies. This supports intelligent and proactive decision-making in railway traffic safety management, effectively ensuring the structural stability and train safety of railway bridges under high-intensity operating conditions.

[0057] To enhance the system's ability to quickly identify sudden structural anomalies (such as pier settlement and sudden displacement changes at beam ends), the prediction model incorporates historical mutation data samples during its training phase, and a custom activation function is used to increase the weight of the response to abnormal change patterns. Furthermore, thresholds for abnormal change rates, such as the displacement change rate per unit time and the slope of vibration acceleration, are embedded in the IFC file's extended properties. These indicators serve as important references for identifying significant deviations from the monitoring data. When the platform identifies any of these indicators as significantly exceeding the preset thresholds, the relevant components are automatically annotated, highlighted in flashing red in the 3D model, and a "high risk" status is displayed in the information panel.

[0058] In addition, while displaying high-risk status, the platform can establish an interface mechanism with the railway dispatching system: once such sudden anomalies are detected and component marking is completed, automatic data push can be triggered, and the railway operation control system can be linked to trigger corresponding response strategies, including speed limit operation, dispatch avoidance, temporary traffic control and other measures, thereby realizing the safe closed-loop management of "anomaly identification - visual presentation - control linkage", and further improving the emergency response efficiency and traffic safety assurance capabilities of railway bridges under complex operating conditions.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A bridge IoT dynamic visualization monitoring and early warning method based on IFC, characterized in that: include: S1: Deploy different types of sensors at various structural parts of the railway bridge to obtain sensor monitoring data corresponding to each structural part of the railway bridge; S2: Obtain the original IFC dictionary file and perform entity expansion on the deployed sensors. The entity expansion is to add a railway bridge sensor type to the original IFC dictionary file to obtain a railway bridge sensor type IFC dictionary file, and then generate a railway bridge sensor type IFC file based on the railway bridge sensor IFC dictionary file. S3: Expanding the attribute set of the railway bridge sensor type, wherein the attribute set expansion is to add sensor data attributes corresponding to the railway bridge sensor type to the railway bridge sensor type IFC file to obtain a basic railway bridge sensor data IFC file; S4: Introduce the BIM platform, input the sensor monitoring data into the basic railway bridge sensor data IFC file based on the BIM platform, and update it in real time to obtain the railway bridge sensor data IFC file; S5: Design a custom activation function, introduce the tanh activation function, gating mechanism, and cell state. Based on the gating mechanism, cell state, custom activation function, and tanh activation function, build an adaptive prediction model for railway bridge monitoring data characteristics. Use sensor monitoring data to train and optimize the parameters of the adaptive prediction model, forming a multi-factor prediction model with time series perception and feature response capabilities. S6: Input the current sensor monitoring data into the multi-factor prediction model to predict the sensor monitoring data for a period of time in the future. Synchronously update the sensor monitoring data predicted by the multi-factor prediction model into the railway bridge sensor data IFC file to dynamically evaluate the structural health status and potential risks of the railway bridge. S7: Based on the updated IFC file of railway bridge sensor data, identify railway bridge structural parts with potential risks, analyze them in combination with real-time monitoring data, and conduct visual monitoring and intelligent early warning of railway bridges; The early warning results are pushed to the maintenance management platform, and linked control is achieved with the railway dispatching command system and bridge maintenance system to improve the structural safety assurance capability of railway bridges during operation and the operational stability of the railway transportation system.

2. The IFC-based bridge IoT dynamic visualization monitoring and early warning method according to claim 1 is characterized in that: The adaptive prediction model includes a feature introduction control unit, a state screening adjustment unit, a prediction derivation adjustment unit and a storage unit; The feature introduction control unit, state screening adjustment unit, prediction derivation adjustment unit and storage unit all contain weight matrices and bias items; the bias items are used to adjust the activation thresholds in the corresponding units and control the default flow tendency of information; The weight matrix of the storage unit is used to map the current sensor monitoring data and the predicted data output by the previous prediction and adjustment unit into a high-dimensional space to extract the nonlinear relationship between features; The weight matrix of the feature introduction control unit is used to determine the contribution of the output data of the previous moment and the current input data to the updated new data; The output data at the previous moment is the predicted data output by the prediction and export adjustment unit at the previous moment, and the current input data is the current sensor monitoring data; The output data of the previous moment and the current sensor monitoring data are input into a feature introduction control unit, and the processed sensor monitoring data are sorted in chronological order by the feature introduction control unit to form a multi-source time series input vector that can be recognized by the adaptive prediction model. The input data is processed based on the weight matrix and bias term of the feature introduction control unit and the storage unit using a custom activation function and a tanh activation function to obtain the output data of the feature introduction control unit and the state information of the candidate storage unit; the output data is a gating signal of the feature introduction control unit, and the state information of the storage unit is a nonlinear fusion representation of the output data of the previous moment and the current sensor monitoring data; The weight matrix of the state screening adjustment unit is used to determine the contribution of the output data of the previous moment and the current sensor monitoring data to forgetting old information; The output data of the previous moment and the current sensor monitoring data are input into the state screening and adjustment unit, and the current sensor monitoring data and the output data of the previous moment are processed based on the weight matrix and bias term of the state screening and adjustment unit using a custom activation function to obtain the output data of the state screening and adjustment unit, and the state information of the storage unit is recursively and updated based on the output data of the state screening and adjustment unit, the state information of the storage unit at the previous moment, the output data of the feature introduction control unit, and the state information of the candidate storage unit to obtain the updated state information of the storage unit at the current moment; The weight matrix of the prediction and export adjustment unit is used to determine the contribution of the output data of the previous moment and the current input data to the output information; Input the current input sensor monitoring data and the output data of the previous moment into the prediction and derivation adjustment unit, use the custom activation function to process the current input sensor monitoring data and the output data of the previous moment based on the weight matrix and bias term of the prediction and derivation adjustment unit, and obtain the output data of the prediction and derivation adjustment unit; The output data of the prediction-derived regulation unit is a gating signal of the prediction-derived regulation unit; The tanh activation function is used to generate the corresponding predicted output based on the output data of the predicted deriving adjustment unit and the state information of the updated storage unit.

3. The IFC-based bridge IoT dynamic visualization monitoring and early warning method according to claim 2 is characterized in that: Designing a custom activation function involves: Design a custom activation function. The custom activation function is shown in formulas (1) and (2). （1） （2） In formula (1), Represents the input sensor monitoring data; Represents the time parameter, which is used to simulate the changes of railway bridges at different times; Indicates environmental factors; are all adjustable parameters of the adaptive prediction model. represents the input weight, where Control the degree of influence of input sensor monitoring data on the custom activation function; Adjust time parameters Periodic effects on model outputs; Control the impact of environmental factors on sensor monitoring data; To adjust the amplitude of periodic changes; To adjust the weights of other nonlinear features; is a bias term used to adjust the output of the model; Indicates that the output is the maximum value between the value in the brackets and 0; In formula (2), T is the monitored temperature value; H is the monitored humidity value; W is the monitored wind speed value; To monitor the average values ​​of temperature, humidity and wind speed; To monitor the standard deviation of temperature, humidity and wind speed.

4. The IFC-based bridge IoT dynamic visualization monitoring and early warning method according to claim 3 is characterized in that: Input sensor monitoring data into the adaptive prediction model for training, including: S51. For each time parameter , the sensor monitoring data are input into the feature introduction control unit, the state screening adjustment unit and the prediction export adjustment unit for processing. The specific steps are as follows: S511: Input the current sensor monitoring data and the output data of the previous moment into the feature introduction control unit for processing, as shown in formula (3), （3） in, Represents the output data of the feature introduction control unit, The weight matrix of the control unit is introduced for the feature, Represents the input data at the current moment, is the output data of the previous moment, Introducing bias terms of control units for features; S512: Input the current sensor monitoring data and the output data of the previous moment into the state screening and adjustment unit for processing, as shown in formula (4), （4） in, Represents the output data of the state screening and adjustment unit, is the weight matrix of the state screening adjustment unit, is the bias term of the state screening adjustment unit; S513: Input the current sensor monitoring data and the output data of the previous moment into the prediction and derivation adjustment unit for processing, as shown in formula (5), （5） in, Represents the output data of the prediction export adjustment unit, represents the weight matrix of the prediction-derived regulation unit, is the bias term of the state screening adjustment unit; S52, using the tanh activation function, combined with the output data of the feature introduction control unit and the output data of the state screening adjustment unit, update the state parameters of the storage unit at the current moment, as shown in formula (6), （6） in, Indicates the state parameters of the storage unit at the current moment, , indicating the part of storage unit information that needs to be retained at the current moment, represents the weight matrix of the storage unit, Represents the bias term of the storage unit; Using the tanh activation function, combined with the predicted output data of the regulation unit and the state parameters of the storage unit at the current moment, the output data at the current moment is calculated, as shown in formula (7). （7） in, Indicates the output data at the current moment.

5. The IFC-based bridge IoT dynamic visualization monitoring and early warning method according to claim 4 is characterized in that: Sensor monitoring data is used to optimize the parameters of the adaptive prediction model to form a multi-factor prediction model with time series perception and feature response capabilities, including: S53. Construct an MSE loss function or a cross entropy loss function, use the loss function to calculate the error between the predicted output and the true value, calculate the gradient of the adjustable parameters of the custom activation function through the back propagation algorithm, and use the Adam optimization algorithm to update the adjustable parameters of the custom activation function. Continuously iterate during the training process until the error converges to obtain the final adjustable parameters of the custom activation function, and then obtain the final custom activation function. The predicted output of the adaptive prediction model is compared with the actual sensor monitoring data, and the prediction error is calculated. The error is backpropagated from the current moment to each time step through the backpropagation algorithm. The contribution of the weight matrix and bias term of the feature introduction control unit, state screening adjustment unit and prediction export adjustment unit to the error, that is, the error gradient, is calculated. The gradient descent algorithm is used to update the weight matrix and bias term according to the error gradient to reduce the prediction error until the prediction error of the adaptive prediction model converges to a relatively stable state, and the final weight matrix and bias term are obtained, and then the final multi-factor prediction model is obtained.

6. The IFC-based bridge IoT dynamic visualization monitoring and early warning method according to claim 1 is characterized in that: Introducing a BIM platform, based on which sensor monitoring data is input into the basic railway bridge sensor data IFC file, and performing real-time updates, including: S41, performing data cleaning on the sensor monitoring data and converting the cleaned sensor monitoring data into an IFC standard format; S42, dynamically binding attributes for the IFC file, that is, using the IFC attribute mapping template, one-to-one mapping the converted sensor monitoring data with the railway bridge sensor entity type defined in IFC, and then writing the converted sensor monitoring data into the corresponding sensor data attributes to achieve dynamic binding; S43. Implement bidirectional synchronization of basic railway bridge sensor data IFC files and converted sensor monitoring data based on a publish-subscribe model.

7. The IFC-based bridge IoT dynamic visualization monitoring and early warning method according to claim 1 is characterized in that: Obtain the original IFC dictionary file and perform entity expansion on the deployed sensors. The entity expansion is to add the railway bridge sensor type to the original IFC dictionary file to obtain the railway bridge sensor type IFC dictionary file. The railway bridge sensor type IFC file is formed according to the railway bridge sensor IFC dictionary file, including: S21, obtaining the original IFC dictionary file; S22, based on the IFC standard, define the sensor space in the original IFC dictionary; S23, performing entity expansion on the sensor, adding structural elements of each type of sensor into the sensor space, and obtaining an IFC dictionary file of the railway bridge sensor type; the IFC dictionary file of the railway bridge sensor type includes key information and content of the sensor; S24. Generate a railway bridge sensor type IFC file according to the required sensor content in the railway bridge sensor type IFC dictionary file.

8. The IFC-based bridge IoT dynamic visualization monitoring and early warning method according to claim 1 is characterized in that: The attribute set of the railway bridge sensor type is expanded. The attribute set expansion is to add sensor data attributes corresponding to the railway bridge sensor type in the railway bridge sensor type IFC file to obtain the basic railway bridge sensor data IFC file, including: S31, read the railway bridge sensor type IFC file, and select the specified railway bridge sensor type; S32. Expand the attribute set of the railway bridge sensor type, that is, add the sensor data attributes to the corresponding railway bridge sensor type to obtain a basic railway bridge sensor data IFC file.

Images