A traffic infrastructure construction management and maintenance integrated data management system

Abstract

The application discloses a kind of traffic infrastructure's construction management maintenance integration data management system, and the application relates to traffic infrastructure whole life cycle management technical field, including dynamic space-time grid encoding engine, multimodal feature vector fusion engine, double verification module and self-evolution feedback module connected in turn.The traffic infrastructure's construction management maintenance integration data management system, through dynamic space-time grid encoding engine, solves the space-time benchmark misplacement problem of multi-source heterogeneous data, realizes the accurate spatial correlation of design model, construction record, monitoring data;Based on the cross-modal fusion mechanism of lightweight feature extraction and twin network hard constraint, the intelligent alignment problem of image, text, time series data is solved, and the real-time and accuracy of disease state perception are improved.The architecture of physical rule engine priority interception and double verification cooperation eliminates the decision risk caused by data error, and solves the delay hidden danger of structure safety early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of whole life cycle management of traffic infrastructure, in particular to a traffic infrastructure construction management and maintenance integrated data management system. BACKGROUND

[0002] In the whole life cycle management of traffic infrastructure, the integrated data management of construction, management and maintenance is the key target to improve efficiency, ensure safety and optimize decision-making. Although the existing related data management systems have integrated information at different stages to a certain extent, they still face a significant and fundamental bottleneck in dealing with actual business needs. The construction, management and maintenance activities of traffic infrastructure involve extremely complex and diverse data sources, including design models, construction records, manual inspection reports, vehicle-mounted or airborne sensor information, fixed-point monitored time series data and various image and video materials. These data have great differences in format specification, collection accuracy, update frequency and most importantly, spatio-temporal reference.

[0003] The existing systems generally lack efficient and automated core capabilities, and cannot intelligently identify, compare and correlate data from different sources, describing the same infrastructure entity or the same type of state characteristics, in a unified spatio-temporal framework. For example, it is difficult to automatically match and correlate the abnormal position point found by automated detection with the defect description in the historical manual record, and the real-time changes captured by the adjacent sensor. The lack of cross-source heterogeneous data fusion capability results in a serious dependence on manual data searching, checking and splicing work, which is inefficient and prone to errors. As a direct result, the system cannot timely and accurately capture the subtle evolution trend of infrastructure state over time, making early warning of potential risks lag, and ultimately making maintenance decisions based on incomplete and timely data, which is difficult to support accurate judgment and action based on whole life cycle data. Therefore, the technical problem to be solved at present is how to efficiently and automatically realize the accurate comparison, correlation and deep fusion of multi-source heterogeneous construction, management and maintenance data in a unified spatio-temporal reference. SUMMARY

[0004] To achieve the above purpose, the present application is implemented by the following technical scheme: a traffic infrastructure construction, management and maintenance integrated data management system, comprising a dynamic spatio-temporal grid coding engine, a multi-modal feature vector fusion engine, a dual verification module and a self-evolution feedback module connected in sequence.

[0005] The dynamic spatio-temporal grid coding engine divides the bridge and road into grid units with unique spatio-temporal coding, and integrates Beidou positioning data, BIM component coordinates and sensor physical coordinates through a nonlinear coordinate mapping algorithm.

[0006] The multi-modal feature vector fusion engine includes a structured data processing pipeline and an unstructured data processing pipeline, respectively generating a statistical feature vector, a visual feature vector and a semantic feature vector, and outputting a unified state description vector after alignment by a twin network;

[0007] The double verification module is composed of a physical rule engine and a dynamic adversarial verification network in series, the physical rule engine preloads a crack propagation model and a material fatigue equation, and the dynamic adversarial verification network includes a generator and a discriminator.

[0008] The self-evolution feedback module adjusts the grid size and fusion weight according to the verification result.

[0009] Preferably, the dynamic space-time grid encoding engine performs grid dynamic adjustment operation, the crack detection area uses a grid unit with a preset edge length of 5-10 cm, and the settlement monitoring area uses a grid unit with a preset edge length of 1-2 m, and the numerical interval of the grid unit with a preset edge length of 5-10 cm and the grid unit with a preset edge length of 1-2 m is set according to the monitoring accuracy requirement.

[0010] Preferably, the nonlinear coordinate mapping algorithm includes: establishing a rigid transformation matrix of Beidou coordinates and BIM component center points; when the deviation between sensor coordinates and grid center points exceeds a preset spatial deviation threshold, triggering grid redivision based on sensor coordinates, wherein the preset spatial deviation threshold is set according to sensor positioning accuracy and grid resolution.

[0011] Preferably, the structured data processing pipeline uses a time series feature extractor to calculate mean, variance and linear regression slope.

[0012] In the unstructured data processing pipeline:

[0013] Image data is extracted by a three-convolution layer lightweight network to extract edge features;

[0014] Text data extracts crack and spalling disease entities by a named entity recognition model.

[0015] Preferably, the twin network alignment process includes:

[0016] The visual feature vector and the semantic feature vector are input into a double-branch network with shared weights.

[0017] The space-time grid encoding is used as a similarity calculation constraint condition.

[0018] The state description vector with 0-1 confidence score is output.

[0019] Preferably, the physical rule engine execution process comprises: determining that the concrete crack propagation equation is violated when the crack daily propagation amount in the state description vector exceeds a preset propagation rate threshold value; triggering an alarm instruction and interrupting data output, wherein the preset propagation rate threshold value is set based on the concrete material characteristics and engineering specifications.

[0020] Preferably, the dynamic adversarial verification network executes:

[0021] The generator simulates the associated results based on historical maintenance records;

[0022] The discriminator compares the simulation results with the actual maintenance effect data;

[0023] Only when the output confidence is not lower than a preset confidence threshold value and passes the physical rule check, the data output interface is activated, wherein the preset confidence threshold value is configured according to the maintenance simulation accuracy requirement.

[0024] Preferably, the connection relationship of the double verification module is:

[0025] The output end of the physical rule engine is connected to the input end of the dynamic adversarial verification network;

[0026] The data that fails to pass the verification is returned to the fusion engine through a feedback loop.

[0027] Preferably, the self-evolution feedback module executes:

[0028] According to the alarm frequency of the physical rule engine, the corresponding grid is subdivided into a preset scaling ratio of the original size, and the implementer can independently set the preset scaling ratio according to the monitoring accuracy requirement;

[0029] Based on the confidence decrease amplitude of the discriminator, the twin network fusion weight is updated according to the gradient.

[0030] Preferably, the traffic infrastructure construction management and maintenance integrated data management system further comprises:

[0031] A visual decision terminal connected to the output end of the double verification module;

[0032] A distributed database for storing associated data according to the space-time grid code.

[0033] The present application provides a traffic infrastructure construction management and maintenance integrated data management system. The following advantages are provided:

[0034] The traffic infrastructure construction management and maintenance integrated data management system solves the time and space benchmark misalignment problem of multi-source heterogeneous data through a dynamic time and space grid coding engine, realizes the accurate spatial correlation of design models, construction records and monitoring data, solves the intelligent alignment problem of images, texts and time series data based on the cross-modal fusion mechanism of lightweight feature extraction and twin network hard constraint, and improves the real-time performance and accuracy of disease state perception. The architecture of the physical rule engine preferentially intercepts and cooperates with double verification, eliminates the decision risk caused by data errors, and solves the delay hidden danger of structure safety warning.

[0035] The traffic infrastructure construction management and maintenance integrated data management system gives the system dynamic optimization capability through the self-evolution feedback module, the grid subdivision and weight adjustment mechanism can adapt to the long-period evolution characteristics of the infrastructure, and reduces the dependence on artificial parameter adjustment; the whole link blood relationship tracing under the unified index of time and space coding makes the construction management and maintenance whole cycle data traceable, verifiable and credible. And realize the change from passive response to active prediction of maintenance decision, prolong the service life of the facility. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 It is the overall framework diagram of the present application;

[0037] Figure 2 It is the flowchart of the present application. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0039] Please refer to Figure 1 and Figure 2 , the present application provides a technical solution: a traffic infrastructure construction management and maintenance integrated data management system, comprising a dynamic time and space grid coding engine, a multi-modal feature vector fusion engine, a double verification module and a self-evolution feedback module connected in turn;

[0040] The dynamic time and space grid coding engine divides the bridge and road into grid units with unique time and space coding, integrates Beidou positioning data, BIM component coordinates and sensor physical coordinates through a nonlinear coordinate mapping algorithm;

[0041] The multi-modal feature vector fusion engine includes a structured data processing pipeline and an unstructured data processing pipeline, which respectively generate statistical feature vectors, visual feature vectors and semantic feature vectors, and output a unified state description vector after alignment by a twin network;

[0042] The double verification module is composed of a physical rule engine and a dynamic adversarial verification network in series, the physical rule engine is preset with a crack propagation model and a material fatigue equation, and the dynamic adversarial verification network includes a generator and a discriminator;

[0043] The self-evolution feedback module adjusts the grid size and fusion weight according to the verification result.

[0044] It needs to be further explained that in the specific implementation process, the system first processes multi-source data through a dynamic space-time grid coding engine, including: after receiving Beidou positioning data, BIM component coordinates and sensor physical coordinates, the bridge pier column, road section and other entities are divided into dynamic grid units, and a unique space-time code is generated for each unit. For the crack detection scene, a fine grid with a side length of 5-10 cm is used, and for the settlement monitoring scene, a coarse grid with a size of 1-2 m is used. When the deviation between the sensor coordinates and the grid center point exceeds 10 cm, the grid is automatically redivided based on the high-precision sensor to ensure the unity of the spatial reference.

[0045] The multi-modal feature vector fusion engine performs shunt translation on the input data, including: structured data extraction mean, variance and trend slope to form a statistical vector, wherein the structured data includes vibration sensor time series; in unstructured data, the crack image is extracted by a three-layer convolution lightweight network to form an edge feature vector, and the artificial inspection text is extracted by a named entity recognition model to form a semantic vector such as "crack width" and "peeling position". All feature vectors are input into a twin network for cross-modal alignment, i.e., the similarity weight between visual features and semantic features is calculated under the condition of space-time grid coding as a hard constraint, and a unified state description vector with a 0-1 confidence score is output.

[0046] The double verification module strictly screens the fusion results, including: the physical rule engine presets a daily crack expansion threshold of 0.5 mm for concrete, and if the state vector shows that the daily widening of a crack exceeds the limit, the data output is immediately frozen and an alarm is given; data that passes the physical check is input into the dynamic adversarial verification network, the generator simulates the associated results based on historical maintenance records, and the discriminator compares the simulation results with the actual maintenance effect data, and only when the output confidence is ≥0.85 does the data output interface activate.

[0047] The self-evolution feedback module dynamically optimizes the system according to the verification result, including: when the physical rule engine alarms the same grid area for three consecutive times, the area is automatically subdivided into one-fourth of the original size; if the discriminator of the dynamic adversarial verification network detects that the correlation confidence of a certain type of data is continuously decreasing, the fusion weight coefficient of the corresponding modal in the twin network is updated according to the gradient. The data that passes the verification is pushed to the visual decision terminal and stored in the distributed database according to the space-time grid coding index.

[0048] The dynamic space-time grid coding engine performs grid dynamic adjustment operation. The crack detection area adopts a grid unit with a preset side length of 5-10 cm, and the settlement monitoring area adopts a grid unit with a preset side length of 1-2 m. It needs to be further explained that in the specific implementation process, the system initializes the grid unit size according to the type of the detection target. When processing the crack detection task, the fine grid unit with a side length of 5-10 cm is automatically enabled, which can capture millimeter-level crack width changes and avoid data overload; and in the settlement monitoring scene, the coarse grid unit with a side length of 1-2 m is switched to, which takes into account the efficiency and accuracy requirements of large ground deformation monitoring.

[0049] The grid division operation is performed in real time in the space-time coding engine. If the same area needs to process crack and settlement monitoring data at the same time, the fine grid for crack detection is used preferentially, and the interpolation data of the settlement sensor is fused.

[0050] The grid dynamic adjustment process follows preset rules, including: the minimum grid size is enabled by default for vulnerable parts such as bridge expansion joints and tunnel linings, and the maximum grid size is used for uniform structures such as embankment filling areas.

[0051] When the physical coordinates of the inclination sensor deployed on the pier column deviate from the current grid center point by more than 10 cm, the 3x3 grid cluster around the sensor is immediately redivided based on the high-precision coordinates of the sensor, until the deviation between the new grid center point and the sensor coordinates is less than 2 cm.

[0052] The space topology relationship of the BIM component is updated synchronously during the redivision process to ensure that different detection data of the same component always belong to the same grid code. The historical version data generated by the grid size switching and redivision operation is stored according to the timestamp index to support state evolution backtracking analysis.

[0053] The nonlinear coordinate mapping algorithm includes: establishing a rigid transformation matrix between the Beidou coordinates and the center point of the BIM component; when the deviation between the sensor coordinates and the grid center point exceeds the preset spatial deviation threshold of 10 cm, the grid is redivided based on the sensor coordinates. It needs to be further explained that in the specific implementation process, when the system receives the Beidou positioning data, the WGS-84 geodetic coordinates are first solved and converted into the local engineering coordinate system, and the center point design coordinates of the component in the BIM model are extracted; a rigid transformation matrix is established between them, and the rotation and translation parameters are fitted by the least squares method to realize the initial alignment of the design coordinates and the measured coordinates.

[0054] If the physical coordinates of the devices such as the tilt sensor, strain gauge, etc. deployed on the surface of the component deviate from the current grid center point by more than 10 cm, it is determined that the coordinate reference is misaligned, and the calibration procedure is immediately started, including: taking the high-precision sensor coordinates as the reference point, freezing the data processing flow of the 3x3 grid area around the sensor, recalculating the grid boundary and dividing new cells according to the reference point coordinate value, until the new grid center point deviates from the sensor coordinates by less than 2 cm.

[0055] During the redrawing process, the original grid space-time coding is automatically inherited and a version identifier is added, and the BIM component space topology relationship library is updated to ensure that the same component's multi-period monitoring data can be traced back to the same coding system. For areas where multiple high-precision sensors exist simultaneously, the device with the highest measurement accuracy is selected as the reference; if the accuracy is the same, the sensor closest to the geometric center of the grid is selected to avoid local calibration causing global topology conflicts. The grid parameters and transformation matrix increment after calibration are updated to the spatial index library for subsequent automatic matching of multi-source data.

[0056] The structured data processing pipeline uses a time series feature extractor to calculate the mean, variance, and linear regression slope;

[0057] In the unstructured data processing pipeline:

[0058] Image data is extracted edge features through a three-convolution layer lightweight network;

[0059] Text data extracts crack and spalling disease entities through a named entity recognition model.

[0060] It should be further noted that in the specific implementation process, after structured data enters the processing pipeline, for time series data generated by vibration sensors, temperature and humidity monitors, etc., the mean is calculated with a 15-minute window to reflect the stability level, the variance represents the fluctuation intensity, and the trend slope is obtained through least squares fitting to identify the long-term evolution direction; when the absolute value of the slope increases continuously for three windows, the window is automatically shortened to 5 minutes to capture sudden abnormal fluctuations.

[0061] In unstructured data processing, crack images are input into a three-convolution layer lightweight network, including: the first layer of 64-channel 3x3 convolution extracts coarse edges, the second layer of 128-channel 1x1 convolution enhances crack texture, and the third layer of 64-channel 3x3 convolution suppresses noise, outputting a 512-dimensional edge feature vector, i.e. this structure reduces the parameter amount to 18% of the standard ResNet while ensuring a 95% recognition rate through the bridge crack library verification.

[0062] The artificial inspection text is processed by a named entity recognition model, and the disease entities such as “crack”, “peeling” and “water seepage” and their modifiers such as “horizontal crack width 3mm” are preferentially extracted. If a sentence contains a location description such as “K25+300 pier top”, the disease entity is associated to generate a spatial semantic vector. When the text description conflicts with the image detection result, such as the text report “no crack” but the image detects a crack, the image feature weight is automatically increased and the review process is triggered.

[0063] The twin network alignment process includes: inputting the visual feature vector and the semantic feature vector into a double-branch network with shared weights; using the spatiotemporal grid code as a similarity calculation constraint; and outputting a state description vector with a 0-1 confidence score. It should be further explained that in the specific implementation process, the visual feature vector and the semantic feature vector are input into a double-branch network structure with shared weight parameters in parallel, the left branch processes the 512-dimensional edge feature vector extracted from the image, and the right branch processes the disease semantic vector generated from the text; when the input data carries the same spatiotemporal grid code, the similarity calculation module is forcibly activated, the cosine similarity of the two feature vectors is calculated and multiplied by the grid position weight coefficient, and if the grid is located in a key area such as the bridge expansion joint, the weight is increased to 1.5 times the reference value.

[0064] The similarity calculation result is converted into a confidence score in the 0-1 interval, including: when the score is higher than 0.7, the fused state description vector is directly generated; when the score is in the interval of 0.4-0.7, the cross-modal enhancement module is triggered, and the attention mechanism is used to focus on key attributes such as crack width and position coordinates for secondary matching; when the score is lower than 0.4, it is determined as invalid association, and the data is automatically discarded and marked as grid code abnormal.

[0065] The output vector is accompanied by a confidence score and a key attribute matching list. When the position coordinate deviation in the list exceeds 5 cm or the crack width difference is greater than 0.2 mm, the confidence is forced to be degraded to below 0.3 and returned to the feature extraction engine for reprocessing.

[0066] The physical rule engine execution process includes: when the crack daily expansion in the state description vector exceeds the preset expansion rate threshold > 0.5 mm, it is determined that the concrete crack propagation equation is violated; trigger the alarm instruction and interrupt the data output, wherein the preset expansion rate threshold is set based on the material properties of concrete and engineering specifications.

[0067] It needs to be further explained that in the specific implementation process, the physical rule engine execution process includes: the system presets the concrete crack daily expansion threshold value 0.5mm as the core judgment basis, which is derived from the fracture mechanics model corresponding to the tensile strength limit of C40 concrete, and when the generated state description vector shows that the single-day width increment of a crack exceeds this limit, the data output stream of the grid element is immediately frozen; automatically switch the verification rules for different structural materials, that is: the fatigue stress amplitude threshold value of steel structure is 150MPa, and the rut depth daily change threshold value of asphalt pavement is 3mm.

[0068] Triggering an alarm synchronously starts root cause analysis, including: if the over-limit crack is located in the bearing area of the pier column, it is marked as a structural risk, and a red warning is generated and pushed to the maintenance terminal; if it is in a non-bearing part such as a crash barrier, it is marked as an appearance damage, and a UAV review process is started.

[0069] For the application of material fatigue equation, when the trend slope of vibration sensor data after feature extraction exceeds 0.15 for three consecutive hours, and the heavy vehicle proportion in the vehicle load spectrum during this period exceeds 40%, the steel structure S-N curve is automatically called to predict the remaining life. All frozen data generate abnormal snapshots with time and space encoding, and associate historical similar event handling records for decision reference until manual confirmation to release the frozen state.

[0070] The dynamic adversarial verification network executes: the generator simulates the associated results based on historical maintenance records; the discriminator compares the simulation results with the actual maintenance effect data; only when the output confidence is not less than the preset confidence threshold ≥0.85 and passes the physical rule check, the data output interface is activated, wherein the preset confidence threshold is configured according to the maintenance simulation accuracy requirement. It needs to be further explained that in the specific implementation process, the dynamic adversarial verification network execution process includes: the generator simulates the associated results based on the typical disease mode library in the near three years of maintenance records, and preferentially extracts historical cases with crack width 0.2-0.5mm and located at the top of the pier column to generate virtual state vectors; the discriminator receives real-time fusion data stream, compares the Mahalanobis distance between the current vibration frequency spectrum of the tested member and the historical normal state after maintenance, and automatically reduces the confidence score when the distance value exceeds three times the standard deviation range.

[0071] The verification process implements a double interception mechanism, that is: if the physical rule engine has marked a structural risk, output a basic confidence of 0.3; only when the physical check passes and the discriminator confidence is ≥0.85, activate the data output interface and generate a state report with a green trusted identifier.

[0072] For data with confidence in the interval of 0.75-0.85, the grid cell is automatically associated with the last three unmanned aerial vehicle inspection high-definition images for manual review to assist decision-making; the data stream with a confidence decrease for three consecutive times triggers the weight coefficient resetting procedure of the feature fusion engine, and the grid data output is frozen until the resetting is completed. All intercepted data generate verification logs with time and space coding, marking the root cause types such as "physical rule conflict", "insufficient discrimination confidence" or "continuous decrease anomaly", and are summarized weekly to drive the self-evolution module optimization.

[0073] The connection relationship of the double verification module is that the output end of the physical rule engine is connected to the input end of the dynamic adversarial verification network; the data that does not pass the verification is returned to the fusion engine through a feedback loop. It needs to be further explained that in the specific implementation process, the double verification module execution process includes: the physical rule engine preferentially receives the state description vector output by the feature fusion engine, and when the crack expansion amount is detected to be more than 0.5 millimeters or the steel structure stress amplitude is more than 150 megapascals, the grid data stream is immediately frozen and a red alarm code is generated; only the vector that passes the physical check is transmitted to the dynamic adversarial verification network.

[0074] The adversarial network discriminator calculates the Mahalanobis distance between the real-time data and the historical normal state. If the confidence score is ≥0.85, the green pass instruction is activated, and the data is pushed to the decision terminal; if the confidence is in the interval of 0.75-0.85, the last three unmanned aerial vehicle images are automatically associated to generate a yellow review report; when it is lower than 0.75, the data backflow mechanism is triggered, that is, the vector that does not pass any verification carries an error code back to the feature fusion engine.

[0075] The backflow data triggers priority resetting, including: the vector that fails the physical rule is forced to re-extract the original sensor data, and the vector that fails the adversarial verification is adjusted to the fusion weight of the twin network and then aligned again. When the same grid data backflows for three consecutive times, the comprehensive verification procedure of the grid time and space coding is started and the associated business processing is suspended.

[0076] The self-evolution feedback module executes:

[0077] According to the alarm frequency of the physical rule engine, the corresponding grid is subdivided into a preset scaling ratio of 1 / 4 of the original size. The implementer can independently set the preset scaling ratio according to the monitoring accuracy requirement;

[0078] Based on the confidence decrease amplitude of the discriminator, the fusion weight of the twin network is updated by gradient. It needs to be further explained that in the specific implementation process, the self-evolution feedback execution process includes: when the physical rule engine triggers an alarm for a specific grid cell, the system automatically records the alarm frequency of the grid; if the same grid alarms for three consecutive times and the crack expansion exceeds the limit, the area grid is immediately subdivided into one-fourth of the original size, and an edge computing node is deployed in the new grid to improve the sampling frequency.

[0079] For the confidence score of dynamic adversarial verification network output, the average confidence value of each grid cell is monitored every week; when the confidence of a specific grid decreases by more than 0.15 for two consecutive weeks, the weight update program is automatically started, that is, if the main reason for the decrease is the conflict between image and text features, the weight coefficient of the text branch of the twin network is reduced by 30%, and the image branch weight is increased to 1.3 times the original value.

[0080] After weight adjustment, the recent 72-hour data stream is reprocessed, and if the average confidence returns to above the baseline, it is confirmed that the optimization is effective; if it continues to decrease, expert intervention mode is triggered, and automatic update is frozen and a structure health assessment request is pushed. All optimization records are archived according to grid code, and when new infrastructure of similar structure accesses the system, the corresponding type of optimization parameters is automatically loaded as the initial configuration.

[0081] The integrated data management system for the construction, management and maintenance of transportation infrastructure also includes a visual decision terminal connected to the output end of the dual verification module, and a distributed database for storing and associating data according to space-time grid code. It needs to be further explained that the data output and storage execution process includes:

[0082] The visual decision terminal receives the green pass data stream activated by the dual verification module in real time, and for structural risk areas marked by physical rules, such as crack exceeding grid, it is marked with a flashing red dot in the three-dimensional BIM model, and the latest three comparison graphs of high-definition images are displayed in association; disease data with adversarial verification confidence ≥0.85 generates a state card with a trusted identifier, and automatically associates similar historical maintenance schemes and material consumption list.

[0083] The distributed database uses space-time grid code as the main index key, and stores data bloodline information synchronously, that is, structured data retains sensor serial number and collection timestamp, and unstructured data adds lightweight network feature extraction parameter version; when a user queries the crack evolution of a pier column from 2023 to 2024, the system preferentially retrieves hot data from the edge node, and if complete analysis requires access to the central database, a bandwidth estimation prompt is automatically generated. For grid cells that continuously trigger the fuse mechanism, mark the database with a yellow warning identifier, and force load the verification log and coordinate redivision record of the physical rule engine for subsequent queries to ensure the completeness and traceability of decision information.

[0084] It needs to be further explained that in the specific implementation process, the system loads the infrastructure BIM model when starting, and decomposes entities such as bridge piers and road sections into grid cells according to spatial position and time dimension. Each cell generates a unique space-time code as a data correlation reference. The coding rules integrate Beidou satellite positioning coordinates, design model component center point coordinates, and sensor physical coordinates. When processing crack detection tasks, small-size grid cells are automatically used to capture millimeter-level changes, and settlement monitoring switches to large-size grid cells to improve processing efficiency. If there are two types of detection needs in a certain area, small-size grids are preferentially retained and large-size monitoring data is interpolated and fused. The grid division process monitors the coordinate deviation in real time. When the measured coordinates of the fixed sensor deviate from the grid center point by more than the normal positioning error, the grid boundary is recalculated based on the high-precision sensor, the new grid inherits the original space-time code and adds a version identifier, ensuring that historical data can be traced.

[0085] The multi-modal feature vector fusion engine classifies and processes input data, including: time series data such as vibration sensors calculate mean value according to fixed window to reflect stable state, variance to represent fluctuation intensity, and trend slope to indicate evolution direction. When the trend slope continues to increase, the analysis window is automatically shortened to capture sudden abnormalities. The crack image input light-weight convolutional network extracts edge features through three specially designed convolutional layers. The first layer captures the general outline, the second layer enhances the crack texture, and the third layer suppresses environmental noise, outputting a feature vector representing the crack morphology. The artificial inspection text is processed by a special entity recognition model, which extracts key information such as crack width and spalling position to generate a semantic vector. When the text description conflicts with the image recognition result, the system automatically increases the image feature weight and triggers the scene review process.

[0086] The feature vector is input into a dual-branch shared weight network for cross-modal alignment, and the similarity is calculated with space-time grid coding as a hard constraint condition. The similarity result is converted into a unified reliability evaluation, including: high-reliability data directly generates fusion results; medium-reliability data starts a secondary matching mechanism, focusing on crack location, width, and other core attributes compared with historical records; low-reliability data is automatically discarded and marked as grid anomaly. The output vector is accompanied by an attribute verification list. When the position coordinate deviation exceeds the normal measurement error or the crack width difference is greater than the detection instrument accuracy, the reliability evaluation is forced to be reduced and returned to the feature extraction link.

[0087] The physical rule engine presets material behavior models, including concrete crack daily expansion threshold, steel structure stress amplitude limit, and other key parameters. When the fused data exceeds the threshold, the output stream is immediately frozen, and a red or yellow warning is generated according to the risk level of the structure part. Through dynamic confrontation verification of the physically checked data input network, the generator simulates the associated results based on the historical maintenance case library, and the discriminator compares the difference amplitude between real-time data and historical normal state. Only when the credibility evaluation reaches a high confidence standard and the physical check passes, the green output channel is activated. Data that fails to pass the verification carries an error code back to the processing engine, data that fails the physical rule reextracts the original sensor readings, and data with insufficient credibility adjusts the feature fusion weight and aligns again.

[0088] The self-evolution feedback module continuously monitors the verification results, including: when a specific grid unit continuously triggers a physical alarm, the area grid is automatically subdivided into one-fourth of the original size and the data acquisition frequency is increased; if the credibility of a certain type of data fusion continues to decline, the system reduces the weight coefficient of the conflict mode by gradient, and optimizes the recent data stream verification effect after reprocessing. All optimization parameters are archived and stored by structure type, and the optimal configuration parameters are automatically loaded to initialize the system when a similar infrastructure is built.

[0089] The visual terminal receives green channel data, highlights the risk area in the three-dimensional model, and associates the historical treatment scheme. The distributed database stores data with time and space grid encoding as the main index key, and synchronously records the data source and processing version. When the user queries, the edge node hot data is preferentially called, and the complete analysis needs to access the central library, which automatically prompts the bandwidth demand. For grid units that have failed multiple verifications, coordinate calibration records and physical rule alarm logs are forcibly associated and displayed to ensure that the decision-making information is complete and traceable.

[0090] A traffic infrastructure construction management and maintenance integrated data management method, comprising the following steps:

[0091] Step S1: The dynamic space-time grid division engine loads the BIM model, initializes the grid unit size according to the structure type, receives the Beidou coordinates, sensor coordinates and design coordinates, and generates a unique space-time code through nonlinear mapping; when the deviation between high-precision sensor coordinates and grid center exceeds the normal bit error, local grid redivision is automatically triggered and a version identifier is added.

[0092] Step S2: The multi-modal feature processing engine divides the input data, including: sliding window extraction of mean, variance and trend features for time series data; edge feature vectors are extracted from crack images through a lightweight convolutional network; disease entities are generated by directional extraction of disease entities from artificial inspection text to generate semantic vectors; when image and text features conflict, image data is preferred and review is started.

[0093] Step S3: The shared weight twin network is aligned across modalities with spatio-temporal coding as a constraint, the feature similarity is calculated, and a fusion vector with a confidence evaluation is output; when the position or size attribute deviation exceeds the detection instrument accuracy, the confidence is forced to degrade and the backflow process is processed.

[0094] Step S4: The physical rule engine preferentially checks the fusion vector, including: freezing the data stream when the super material behavior threshold is exceeded and warning according to the risk level of the structural part; the vector input dynamic adversarial verification network is passed.

[0095] Step S5: The adversarial network generator simulates the associated results based on the historical case library, and the discriminator compares real-time data with historical normal state; only when the high confidence standard and physical verification are met at the same time, the green output channel is activated; the data that does not pass is shunted back according to the error type.

[0096] Step S6: The self-evolution module monitors the verification results, including: automatically subdividing the size of the grid unit and increasing the sampling rate when the physical alarm is continuously triggered; when the fusion confidence of a specific modality continuously decreases, the gradient adjusts the feature weight and verifies the optimization effect.

[0097] Step S7: The visualization terminal receives green channel data, highlights the risk area in the three-dimensional model, and associates the historical disposal scheme; the distributed database stores the whole link blood relationship information with spatio-temporal coding index.

[0098] Through the dynamic spatio-temporal grid coding engine, the spatio-temporal benchmark misalignment problem of multi-source heterogeneous data is solved, and the precise spatial correlation of design models, construction records and monitoring data is realized; based on the cross-modal fusion mechanism of lightweight feature extraction and twin network hard constraint, the intelligent alignment problem of image, text and time series data is solved, and the real-time and accuracy of disease state perception are improved. The physical rule engine preferentially intercepts and cooperates with the double verification architecture, eliminating the decision risk caused by data errors and solving the delay hidden danger of structure safety warning.

[0099] The self-evolution feedback module gives the system dynamic optimization capability, and the grid subdivision and weight adjustment mechanism can adapt to the long-period evolution characteristics of infrastructure, reducing the dependence on manual parameter adjustment; the whole link blood relationship tracing under the unified index of spatio-temporal coding makes the whole cycle data of construction, management and maintenance traceable, verifiable and credible. And realize the change from passive response to active prediction of maintenance decision, prolong the service life of facilities.

[0100] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and implementations, it is to be understood that the terminology used is for the purpose of descriptive clarity and that it should be taken in a descriptive sense and not a limiting sense.

[0101] While the embodiments of the application have been shown and described herein, it is to be understood that the application is not limited to these embodiments. Rather, many modifications, changes, substitutions, and alterations can be made to the embodiments of the application without departing from the spirit and scope of the application as defined by the appended claims and their equivalents.

Claims

1. A traffic infrastructure construction management and maintenance integrated data management system, characterized by, The system comprises a dynamic space-time grid coding engine, a multi-modal feature vector fusion engine, a double verification module and a self-evolution feedback module connected in sequence. The dynamic space-time grid coding engine divides the bridge and road into grid units with unique space-time coding, and integrates Beidou positioning data, BIM component coordinates and sensor physical coordinates through a nonlinear coordinate mapping algorithm. The multi-modal feature vector fusion engine comprises a structured data processing pipeline and an unstructured data processing pipeline, which respectively generate statistical feature vectors, visual feature vectors and semantic feature vectors, and outputs a unified state description vector after alignment by a twin network. The double verification module is composed of a physical rule engine and a dynamic adversarial verification network in series, the physical rule engine presets a crack propagation model and a material fatigue equation, and the dynamic adversarial verification network comprises a generator and a discriminator. The self-evolution feedback module adjusts the grid size and fusion weight according to the verification result. The twin network alignment process comprises: inputting the visual feature vector and the semantic feature vector into a double-branch network with shared weights; using space-time grid coding as a similarity calculation constraint; outputting a state description vector with a 0-1 confidence score; the dynamic adversarial verification network performs: the generator simulates the associated results based on historical maintenance records; the discriminator compares the simulation results with the actual maintenance effect data; only when the output confidence is not lower than the preset confidence threshold and passes the physical rule check, the data output interface is activated; the connection relationship of the double verification module is: the output end of the physical rule engine is connected to the input end of the dynamic adversarial verification network; unverified data is returned to the fusion engine through a feedback loop.

2. The traffic infrastructure construction management and maintenance integrated data management system according to claim 1, characterized by: The dynamic space-time grid coding engine performs grid dynamic adjustment operation, the crack detection area uses grid units with a preset side length of 5-10 cm, and the settlement monitoring area uses grid units with a preset side length of 1-2 m.

3. The traffic infrastructure construction management and maintenance integrated data management system according to claim 2, characterized by: The nonlinear coordinate mapping algorithm comprises: establishing a rigid transformation matrix of Beidou coordinates and BIM component center points; when the sensor coordinates deviate from the grid center point by more than a preset spatial deviation threshold, triggering grid redivision based on sensor coordinates.

4. The traffic infrastructure construction management and maintenance integrated data management system according to claim 1, characterized by: The structured data processing pipeline uses a time series feature extractor to calculate mean, variance and linear regression slope; In the unstructured data processing pipeline: image data is extracted by a three-convolution layer lightweight network to extract edge features; text data is extracted by a named entity recognition model to extract crack and spalling disease entities.

5. The traffic infrastructure construction management and maintenance integrated data management system according to claim 1, characterized by: The physical rule engine execution process comprises: when the crack daily propagation amount in the state description vector exceeds the preset propagation rate threshold, it is determined that the concrete crack propagation equation is violated; an alarm instruction is triggered and data output is interrupted.

6. The traffic infrastructure construction management and maintenance integrated data management system according to claim 5, characterized by: The self-evolution feedback module performs: according to the alarm frequency of the physical rule engine, the corresponding grid is subdivided by a preset scaling ratio of the original size; based on the confidence decrease amplitude of the discriminator, the twin network fusion weight is updated by gradient.

7. The traffic infrastructure construction management and maintenance integrated data management system according to claim 1, characterized by: The traffic infrastructure construction, management and maintenance integrated data management system further comprises: a visual decision terminal connected to the output end of the double verification module; a distributed database for storing associated data according to space-time grid coding.

Images