Mobile jig intelligent monitoring method and system based on digital twinning and multi-modal perception

Abstract

The application provides a mobile formwork intelligent monitoring method and system based on digital twinning and multi-modal perception, and belongs to the field of intelligent building construction technology, and comprises the following steps: establishing a multi-modal sensor network on a mobile formwork entity; establishing a digital twinning platform; constructing a BIM model mapped with the mobile formwork entity, pre-burying virtual monitoring points in the BIM model, matching and calibrating the virtual monitoring points with sensors of the multi-modal sensor network, and fusing multi-source sensor data collected by the multi-modal sensor network in real time; calculating a construction health index SHI in real time based on the multi-source sensor data; predicting an instability probability in real time based on the multi-source sensor data through an AE-LSTM hybrid model; simultaneously predicting a weight factor of the construction health index SHI; and jointly evaluating whether to generate a mobile formwork adjustment instruction and issuing a warning according to the SHI and the instability probability. The application can control the safety state of the mobile formwork in real time, and solves the problem that instability trends cannot be monitored in real time.

Description

Technical Field

[0001] This invention belongs to the field of intelligent building construction technology, and in particular relates to an intelligent monitoring method and system for mobile formwork based on digital twin and multimodal perception that integrates Building Information Modeling (BIM), Internet of Things (IoT), and Artificial Intelligence (AI). Background Technology

[0002] Mobile formwork cast-in-place box girder technology is an advanced bridge construction technique. Its core lies in using a movable formwork system to cast concrete box girder structures on-site at the bridge site. This technology combines the advantages of mechanized construction with cast-in-place processes and is widely used in modern bridge construction. Although the construction of mobile formwork cast-in-place box girders is a relatively mature technology, many problems still exist during construction, such as:

[0003] 1. The moving formwork structure has high risks, complex stress and deformation conditions, and cannot monitor instability trends in real time;

[0004] 2. In complex working conditions (such as through holes with small curve radii), template configuration adjustment relies on manual experience, and the accumulation of errors leads to linear deviations;

[0005] 3. Lack of real-time monitoring of concrete compaction and formwork displacement leads to frequent quality defects; long advance movement time of movable formwork makes it impossible to control the stress state of the formwork in real time, which can easily cause problems such as honeycomb and pitting.

[0006] 4. Data from the design, construction, and supervision parties is fragmented, there is no unified standard for risk assessment, decision-making is delayed, and it is difficult to achieve collaboration among all parties. Summary of the Invention

[0007] The purpose of this invention is to propose an intelligent monitoring method and system for mobile formwork based on digital twins and multimodal perception. It integrates Building Information Modeling (BIM), Internet of Things (IoT), and Artificial Intelligence (AI) to achieve real-time acquisition of various data of cast-in-place box girders using mobile formwork. The collected data is used to monitor the safety status of the mobile formwork in real time, thereby enhancing the construction safety of mobile formwork, accelerating the construction efficiency of cast-in-place box girders, and improving the construction quality of cast-in-place box girders.

[0008] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0009] A method for intelligent monitoring of mobile formwork based on digital twin and multimodal perception, comprising:

[0010] S1. Establish a multimodal sensor network on the moving mold frame entity;

[0011] S2. Establish a digital twin platform: Construct a BIM model that maps to the moving formwork entity, embed virtual monitoring points in the BIM model, match and calibrate with the sensors of the multimodal sensor network, and fuse multi-source sensor data collected by the multimodal sensor network in real time.

[0012] S3. Based on the multi-source sensor data, calculate the construction health index SHI in real time;

[0013] S4. Based on the multi-source sensor data, the instability probability is predicted in real time using the AE-LSTM hybrid model; at the same time, the weighting factors of each parameter of the construction health index SHI are predicted for the next calculation of the construction health index SHI.

[0014] S5. Based on the construction health index SHI and the predicted instability probability, jointly assess whether to generate a moving formwork adjustment instruction and issue an early warning.

[0015] Furthermore, in step S1, the establishment of the multimodal sensor network includes:

[0016] S101. Fiber optic grating sensors are installed at the mid-span and support points of the main beam of the movable formwork to monitor the stress of the main beam.

[0017] S102. Install tilt sensors at the four corners of the top plate of the movable mold frame to monitor the lateral tilt of the mold frame;

[0018] S103. An array of acoustic sensors is installed inside the formwork of the movable formwork to monitor the compaction of the concrete.

[0019] S104. Install laser displacement gauges at the bottom of the support legs of the movable formwork to monitor the settlement of the support legs;

[0020] S105. Install a temperature, humidity and anemometer on the top of the movable mold frame to monitor environmental parameters;

[0021] S106. Industrial cameras are installed at the critical joints of the moving formwork to monitor misalignment of the formwork joints.

[0022] Furthermore, the method for pre-embedding virtual monitoring points in the BIM model in step S2 includes:

[0023] S201. Establish the association between the GIS coordinates of each sensor in the multimodal sensor network and the location of the moving module components of the BIM model, so as to realize the spatial binding of monitoring data and BIM model;

[0024] S202. Create monitoring point families for different parameters, mark and distinguish different sensor types, and label key sensor information;

[0025] S203. Use the MQTT protocol to associate sensor data streams with BIM model moving module components in real time.

[0026] Furthermore, the calculation method for the construction health index SHI mentioned in step S3 includes:

[0027] ;

[0028] SHI represents the construction health index; Represents the stress safety factor. Weighting factors representing the stress safety factor; Represents the degree of deformation deviation. Weighting factors representing the degree of deformation deviation; Represents environmental risk value, Weighting factors representing environmental risk values; , , Calculated in real time based on data from multiple sensor sources; Real-time updates are achieved through predictions using an AE-LSTM hybrid model.

[0029] Furthermore, step S4, which involves real-time prediction of the instability probability using the AE-LSTM hybrid model, includes:

[0030] S401: A fixed LSTM module is used to train the AE module with historical multi-source sensor data, enabling the AE module to effectively extract key features from the multi-source sensor data.

[0031] S402. Load the AE module, connect the LSTM module, train it using historical data containing instability events, and output the instability probability. The instability event is the instability value of the corresponding historical multi-source sensor data at the next k time steps.

[0032] S403: Based on real-time received multi-source sensor data, the instability probability is predicted through the AE module and LSTM module.

[0033] Furthermore, the training of the AE-LSTM hybrid model also includes:

[0034] Using historical multi-source sensor data as input data, feature engineering is performed on the input data corresponding to various parameters of the Construction Health Index (SHI). From the historical multi-source sensor data, the data corresponding to the SHI stress safety factor parameter is used to calculate the temporal rate of change of its stress value; the data corresponding to the SHI deformation deviation parameter is used to form a spatial gradient distribution based on its deformation data; and the data corresponding to the SHI environmental risk value parameter is used to calculate the coupling oscillation frequency of the environmental parameters. The temporal rate of change of the stress value, the spatial gradient distribution, and the coupling oscillation frequency of the environmental parameters are used as feature vectors for training.

[0035] The AE-LSTM hybrid model introduces an attention layer to dynamically adjust the weights of each feature vector;

[0036] The output is the weighting factors of each parameter of the Construction Health Index (SHI).

[0037] Furthermore, step S5 includes:

[0038] Set an instability probability threshold and a SHI threshold. When the instability probability predicted by the AE-LSTM hybrid model is greater than the instability probability threshold, and the SHI calculated based on the weighting factor predicted by the AE-LSTM hybrid model is greater than the SHI threshold, a moving formwork adjustment command is generated and an early warning is issued.

[0039] Furthermore, generating movement mold adjustment commands and issuing warnings includes:

[0040] Risk areas are identified on the digital twin platform, a hydraulic correction scheme for the mobile formwork actuator is developed, and emergency plan alerts are pushed out.

[0041] Furthermore, it also includes: the AE-LSTM hybrid model makes predictions every 30 seconds, and the SHI calculation is performed once every 1 second. Each SHI calculation uses the weight factors predicted by the current AE-LSTM hybrid model until the weight factors are updated after the next AE-LSTM hybrid model predicts the weight factors.

[0042] In another aspect, this invention proposes an intelligent monitoring system for mobile formwork based on digital twins and multimodal perception, comprising:

[0043] Multimodal sensor network: Establish a multimodal sensor network on the moving frame entity;

[0044] Digital twin platform: Construct a BIM model that maps to the physical moving formwork, embed virtual monitoring points in the BIM model, match and calibrate with sensors of a multimodal sensor network, and fuse multi-source sensor data collected by the multimodal sensor network in real time;

[0045] SHI Calculation Module: Based on the multi-source sensor data, calculates the Construction Health Index (SHI) in real time;

[0046] AE-LSTM hybrid model prediction module: Based on the multi-source sensor data, it predicts the instability probability in real time through the AE-LSTM hybrid model; at the same time, it predicts the weighting factors of each parameter of the construction health index SHI for the next calculation of the construction health index SHI.

[0047] Early warning assessment module: Based on the construction health index SHI and the predicted instability probability, jointly assess whether to generate a moving formwork adjustment instruction and issue an early warning.

[0048] Furthermore, multimodal sensor networks include:

[0049] Fiber grating sensors are installed at the mid-span and support points of the main beam of the moving formwork to monitor the stress of the main beam.

[0050] Inclination sensors are installed at the four corners of the top plate of the movable mold frame to monitor the lateral tilt of the mold frame;

[0051] An array of acoustic sensors is installed inside the formwork of the movable formwork to monitor the compaction of the concrete during vibration.

[0052] Laser displacement gauges are installed at the bottom of the support legs of the movable formwork to monitor the settlement of the support legs;

[0053] A temperature, humidity and anemometer is installed on the top of the mobile mold frame to monitor environmental parameters;

[0054] Industrial cameras are installed at key joints of the moving formwork to monitor misalignment of the formwork seams.

[0055] Furthermore, digital twin platforms include:

[0056] Spatial binding unit: Establish the association between the GIS coordinates of each sensor in the multimodal sensor network and the position of the moving module components of the BIM model, so as to realize the spatial binding of monitoring data and BIM model;

[0057] Monitoring and marking unit: Creates families of monitoring points for different parameters, marks different sensor types, and labels key sensor information;

[0058] Real-time association unit: The sensor data stream is associated with the BIM model moving module components in real time using the MQTT protocol.

[0059] Furthermore, the SHI calculation module includes:

[0060] ;

[0061] SHI represents the construction health index; Represents the stress safety factor. Weighting factors representing the stress safety factor; Represents the degree of deformation deviation. Weighting factors representing the degree of deformation deviation; Represents environmental risk value, Weighting factors representing environmental risk values; , , Calculated in real time based on data from multiple sensor sources; Real-time updates are achieved through predictions using an AE-LSTM hybrid model.

[0062] Furthermore, the AE-LSTM hybrid model prediction module includes:

[0063] First training unit: The LSTM module is fixed, and the AE module is trained using historical multi-source sensor data, so that the AE module can effectively extract key features from the multi-source sensor data;

[0064] Second training unit: Load the AE module, connect the LSTM module, train using historical data containing instability events, and output the instability probability. The instability event is the instability value of the corresponding historical multi-source sensor data at the next k time steps.

[0065] The prediction unit predicts the instability probability based on real-time received multi-source sensor data through the AE module and LSTM module.

[0066] Furthermore, the AE-LSTM hybrid model prediction module also includes:

[0067] Using historical multi-source sensor data as input data, feature engineering is performed on the input data corresponding to various parameters of the Construction Health Index (SHI). From the historical multi-source sensor data, the data corresponding to the SHI stress safety factor parameter is used to calculate the temporal rate of change of its stress value; the data corresponding to the SHI deformation deviation parameter is used to form a spatial gradient distribution based on its deformation data; and the data corresponding to the SHI environmental risk value parameter is used to calculate the coupling oscillation frequency of the environmental parameters. The temporal rate of change of the stress value, the spatial gradient distribution, and the coupling oscillation frequency of the environmental parameters are used as feature vectors for training.

[0068] The AE-LSTM hybrid model introduces an attention layer to dynamically adjust the weights of each feature vector;

[0069] The output is the weighting factors of each parameter of the Construction Health Index (SHI).

[0070] Furthermore, the early warning assessment module includes:

[0071] Set an instability probability threshold and a SHI threshold. When the instability probability predicted by the AE-LSTM hybrid model is greater than the instability probability threshold, and the SHI calculated based on the weighting factor predicted by the AE-LSTM hybrid model is greater than the SHI threshold, a moving formwork adjustment command is generated and an early warning is issued.

[0072] Furthermore, the early warning assessment module includes:

[0073] Risk areas are identified on the digital twin platform, a hydraulic correction scheme for the mobile formwork actuator is developed, and emergency plan alerts are pushed out.

[0074] Furthermore, the early warning assessment module also includes: the AE-LSTM hybrid model makes predictions every 30 seconds, and the SHI calculation is performed every 1 second. Each SHI calculation uses the weight factors predicted by the current AE-LSTM hybrid model until the weight factors are updated after the next AE-LSTM hybrid model predicts the weight factors.

[0075] Compared with the prior art, the present invention has the following beneficial effects:

[0076] 1. This invention enables synchronous real-time monitoring of multiple key parameters in the construction of cast-in-place box girders using a moving formwork by setting up a multimodal sensor network, and provides necessary data support for digital twin and AI intelligent decision-making;

[0077] 2. This invention establishes a digital twin platform to integrate multi-source sensor data collected by a multimodal sensor network in real time, dynamically linking the BIM model with measured data to achieve intelligent management of the entire construction lifecycle and provide data integration for further AI decision-making;

[0078] 3. This invention proposes a specific design scheme for the Construction Health Index (SHI) and its calculation method, unifies the risk assessment standard, integrates all information into a comprehensive indicator, and can be dynamically adjusted through its own weighting factors, so that the comprehensive indicator results are more in line with the actual working conditions, and solves the problems of decision-making lag and the efficiency of coordination among all parties.

[0079] 4. This invention proposes an AE-LSTM hybrid model as a prediction model for instability probability. AE achieves data denoising and information fusion by reducing the dimensionality and extracting features from the original data of multiple sensors, providing key low-dimensional features for the LSTM model to analyze the temporal patterns of the moving formwork construction process; the LSTM model predicts the instability probability, solving the problems of not being able to monitor instability trends in real time and frequent quality defects.

[0080] 5. This invention proposes a scheme to jointly assess whether to trigger a moving formwork adjustment command based on the Construction Health Index (SHI) and the predicted instability probability. By enhancing the SHI and the prediction model in both directions, a two-way driven intelligent decision-making system is formed, which can realize early and accurate warning of structural instability, reduce false alarm rate, and improve the response speed of risk emergency. Attached Figure Description

[0081] Figure 1 This is a flowchart of Embodiment 1 of the present invention.

[0082] Figure 2 This is a schematic diagram of the intelligent monitoring architecture of Embodiment 1 of the present invention.

[0083] Figure 3 This is a schematic diagram illustrating the correlation and synergy between SHI and the prediction model in Embodiment 1 of the present invention. Detailed Implementation

[0084] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0085] The main design concept of this invention is to realize intelligent all-round monitoring of cast-in-place box girders with mobile formwork facing complex working conditions (such as small radius curves passing through holes) or extreme environments (such as typhoon environments). Based on multimodal perception fusion and AI decision-making, it solves the problems of quality defect prevention and real-time safety early warning.

[0086] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0087] Example 1:

[0088] This embodiment proposes an intelligent monitoring method for mobile formwork based on digital twins and multimodal perception, such as... Figure 1 The steps shown are to establish as follows: Figure 2 The architecture shown specifically includes:

[0089] S1. Establish a multimodal sensor network on the moving mold body.

[0090] To achieve comprehensive monitoring of cast-in-place box girders using a mobile formwork, a multimodal sensor network needs to be established within the mobile formwork.

[0091] The type of multimodal sensor network and related parameters established in this embodiment are shown in the table below:

[0092] ;

[0093] Based on the aforementioned multimodal sensor network, collecting multi-source sensor data from the moving formwork can help improve construction accuracy and real-time monitoring, enhance safety and risk warning, and promote the digital transformation of bridge construction, thus contributing to the realization of intelligent construction.

[0094] S2. Establish a digital twin platform: Construct a BIM model that maps to the moving formwork entity, embed virtual monitoring points in the BIM model, match and calibrate with the sensors of the multimodal sensor network, and fuse multi-source sensor data collected by the multimodal sensor network in real time.

[0095] The purpose of this step is to establish a digital twin platform, which specifically includes:

[0096] 1. Construct a BIM model of the moving formwork and map it 1:1 to the physical entity of the moving formwork. Establish a parametric BIM family library for the model, including adaptive components with small radius curves.

[0097] 2. Embed virtual monitoring points in the BIM model.

[0098] 2.1 Establish the association between the GIS coordinates of each sensor in the multimodal sensor network and the location of the moving module components of the BIM model, so as to realize the spatial binding of monitoring data and BIM model;

[0099] A coordinate system transformation matrix is ​​established, and a 4×4 homogeneous transformation matrix (including rotation matrix R and translation vector T) is constructed to realize the mathematical transformation from the sensor's local coordinate system to the BIM global coordinate system. The calibration parameters can be optimized by the least squares method to solve the coordinate offset problem caused by sensor installation deviation, ensuring that the coordinate transmission error is ≤±2mm. The coordinate drift caused by construction vibration is handled by the Kalman filter algorithm, and the transformation parameters are updated in real time.

[0100] 2.2 Create monitoring point families for different parameters, mark and distinguish different sensor types, and label key sensor information;

[0101] Assign unique codes (such as RFID tags) to sensors and establish a bidirectional index relationship with BIM component IDs; extend the SensorEntity entity class representing sensor devices in the IFC standard to add dynamic attribute fields such as coordinate offset and installation angle; and use a graph database (Neo4j) to store the sensor network topology to realize the visualization of the data link between "sensor-acquisition terminal-digital twin platform".

[0102] 2.3. The MQTT protocol is used to associate sensor data streams with BIM model moving module components in real time;

[0103] Sensor data packets are transmitted based on the MQTT protocol, including timestamps, coordinate system identifiers, and measurement triplet sets. Delaunay triangulation can be used to complete the data for sparse monitoring points, improving the mapping coverage of irregular structures. Then, dynamic coloring of the BIM model driven by monitoring data can be achieved through shader programming in the Unity3D / Unreal engine.

[0104] S3. Based on the multi-source sensor data, calculate the construction health index SHI in real time.

[0105] The Construction Health Index (SHI) essentially uses a single number to quantify the health status of cast-in-place box girder construction using mobile formwork. In traditional engineering projects, safety officers have to simultaneously monitor stress tables, deformation data, and weather forecasts, which can easily lead to overlooking crucial information. SHI, on the other hand, integrates all this information into a single comprehensive indicator.

[0106] Specifically, the Construction Health Index (SHI) was established to address three key issues:

[0107] The first issue is the lack of standardized risk assessment criteria. For example, in the same extreme environment, such as a typhoon, the same amount of structural deformation may result in completely different risk levels. The Construction Health Index (SHI) can be dynamically adjusted through its own weighting factors, making the overall indication results more consistent with actual working conditions.

[0108] The second issue is the problem of decision-making lag. By the time on-site personnel discover a problem, such as excessive stress, and report it up the chain of command, the optimal time for handling the situation may have already passed. However, by calculating the SHI in real time and combining it with the instability probability predicted by the forecasting model, an early warning can be issued at a certain time (e.g., 30 minutes), which is sufficient time to activate the emergency response plan.

[0109] Thirdly, there's the issue of collaborative efficiency. Supervision units focus on safety, construction units are concerned with progress, and design units prioritize alignment control. SHI uses intuitive, unified indicators to help all parties reach a consensus on risk levels.

[0110] The calculation methods for the Construction Health Index (SHI) include:

[0111] ;

[0112] SHI represents the construction health index; Represents the stress safety factor. Weighting factors representing the stress safety factor; Represents the degree of deformation deviation. Weighting factors representing the degree of deformation deviation; Represents environmental risk value, Weighting factors representing environmental risk values; , , Calculated in real time based on data from multiple sensor sources; Real-time updates are achieved through predictions using an AE-LSTM hybrid model.

[0113] in, , , In this embodiment, the calculation can be performed in the following manner:

[0114] ; Indicates the real-time maximum stress; This represents the allowable stress value, which can be a fixed threshold or dynamically adjusted based on historical fatigue data using the prediction model (AE-LSTM hybrid model) described below.

[0115] ; Represents deformable variables; Indicates the allowable deformation value;

[0116] ; T represents the concrete temperature; represents the temperature at the time of pouring into the mold; represents the critical temperature rise value, which is a non-linear parameter; when the internal temperature gradient of the concrete exceeds 15°C this term will grow exponentially and can accurately capture the cracking risk.

[0117] The proposal of the construction health index SHI and its calculation method converts the originally vague risk perception into a quantifiable index, which can achieve unified and accurate risk control. However, in this embodiment, the risk perception of the cast-in-place box girder with a movable formwork is not only based on SHI, but SHI is associated and coordinated with the instability prediction model described below to form a two-way driven AI intelligent decision-making system.

[0118] S4. Based on the multi-source sensor data, the instability probability is predicted in real time through the AE-LSTM hybrid model; at the same time, the weight factors of each parameter of the construction health index SHI are predicted for the next calculation of the construction health index SHI.

[0119] In this embodiment, the instability prediction model adopted is the AE-LSTM hybrid model. The AE-LSTM hybrid model adopts a cascaded structure of "autoencoder (AE) + long short-term memory network (LSTM)", and its core functions are: multi-source sensor data feature extraction - time series dependence modeling - instability trend prediction. The specific architecture is as follows:

[0120] [[ID=第十九]] The autoencoder (AE) module: includes an encoder (Encoder) and a decoder (Decoder). The encoder consists of 2 - 3 fully connected layers, inputs the original multi-source sensor data x (with a dimension of d), and outputs a low-dimensional feature vector h (with a dimension of d′, d′ < d); the decoder reconstructs the reconstructed data obtained by reconstructing h through symmetric fully connected layers , for self-supervised learning of feature extraction. For example, the original data is the image data (pixel matrix) of the misalignment of the formwork joints, and d is the number of pixels. AE extracts the geometric features (such as width, position) of the misalignment as h, reducing the dimension while retaining the misalignment information.

[0121] The LSTM module: the input is the low-dimensional feature sequence output by the encoder , represents the low-dimensional feature vector h at time t, captures the time series dependence relationship through 2 LSTM layers, and finally outputs the instability prediction probability for the next k time steps through a fully connected layer.

[0122] The input data for the AE-LSTM hybrid model consists of multi-source sensor data collected in real time by the multimodal sensor network, specifically including: main beam stress data monitored by fiber optic grating sensors, lateral tilt data of the formwork monitored by tilt sensors, concrete compaction data monitored by acoustic sensor arrays, leg settlement data monitored by laser displacement gauges, environmental parameter data monitored by temperature, humidity and anemometers, and template joint misalignment data monitored by industrial cameras.

[0123] The AE-LSTM hybrid model, used as an instability prediction model, is correlated and synergistic with the Construction Health Index (SHI). Therefore, the input data undergoes feature engineering based on the parameters involved in the SHI calculation, transforming it into corresponding feature vectors. The significance of feature engineering is to convert fragmented input data into physically meaningful feature vectors, which can be used to predict SHI weighting factors and improve the prediction accuracy of the AE-LSTM hybrid model. For the stress safety factor corresponding to SHI, the temporal variation rate of stress values ​​(such as main beam stress data) in multi-source sensor data is statistically analyzed. For the deformation deviation corresponding to SHI, a spatial gradient distribution is formed based on deformation data (such as lateral tilt data and outrigger settlement data) in multi-source sensor data. For the environmental risk value corresponding to SHI, the coupled oscillation frequencies of environmental parameters (wind speed, temperature, humidity) in multi-source sensor data are statistically analyzed. These coupled oscillation frequencies are obtained by calculating the resonance frequencies of wind speed, temperature, and humidity, and are direct correlation quantities of the feature values.

[0124] The AE-LSTM hybrid model also introduces an attention layer to dynamically adjust the weights of the transformed feature vectors.

[0125] The AE-LSTM hybrid model outputs multiple objectives through different fully connected layers. First, it outputs the instability probability of the moving formwork at future k time steps, with values ​​ranging from [0,1] (0 indicating stability, 1 indicating instability), used to provide early warning of structural instability trends. Second, it outputs the weighting factors of each parameter in the Construction Health Index (SHI). .

[0126] The AE-LSTM hybrid model uses a joint loss function that combines the reconstruction loss of the AE module with the prediction loss of the LSTM module, as shown in the following formula:

[0127] ;

[0128] Where L represents the loss and MSE is the mean squared error;

[0129] is a weighting coefficient used to balance the importance of feature extraction and prediction tasks; in this embodiment, a value of 0.3 is acceptable. x represents the original data. To reconstruct the data; y is the actual instability value. This represents the predicted probability of instability.

[0130] The training process of the AE-LSTM hybrid model is as follows:

[0131] 1. Pre-training phase: First, fix the LSTM module and train only the AE module. Use historical stable-state data (no unstable events) and minimize the reconstruction loss. Optimize AE parameters to ensure that the encoder can effectively extract key features from multi-source sensor data.

[0132] 2. Joint Training Phase: Load the pre-trained AE module, connect it to the LSTM module, and perform end-to-end training using historical data containing instability events (labeling the instability value y for the next k time steps). In this embodiment, the optimizer is Adam (learning rate 1e-4), the batch size is set to 32, the training epochs are 200, the loss is validated on the validation set in each epoch, and an early stopping strategy is used (stop if there is no decrease after 10 consecutive epochs).

[0133] 3. Online updates: After deployment, new monitoring data is collected regularly, and the model parameters are fine-tuned with a small learning rate (1e-5) to adapt to changes in the working conditions of the moving formwork.

[0134] The trained AE-LSTM hybrid model, when fed with real-time multi-source sensor data, can predict the instability probability and the SHI weight factor.

[0135] S5. Based on the construction health index SHI and the predicted instability probability, jointly assess whether to generate a moving formwork adjustment instruction and issue an early warning.

[0136] In this embodiment, the AE-LSTM hybrid model is configured to generate a prediction every 30 seconds, while SHI is calculated every second. The two can be decoupled through an asynchronous message queue. Within 30 seconds after the AE-LSTM hybrid model makes a prediction, each calculation of SHI uses the weight factors predicted by the AE-LSTM hybrid model in that prediction, until the next AE-LSTM hybrid model prediction after 30 seconds obtains new weight factors, at which point the weight factors are updated.

[0137] An instability probability threshold and a SHI threshold are set. When the instability probability predicted by the AE-LSTM hybrid model is greater than the instability probability threshold, and the SHI calculated based on the weighting factor predicted by the AE-LSTM hybrid model is greater than the SHI threshold, a moving formwork adjustment command is generated and an early warning is issued. In this embodiment, the instability probability threshold is set to 85%, and the SHI threshold is set to 0.8.

[0138] The generation of the mobile template adjustment command and the issuance of the early warning include three aspects of simultaneous triggering:

[0139] (1) Identify risk areas on the digital twin platform;

[0140] (2) Based on real-time multi-source sensor data, formulate a hydraulic correction scheme for the moving mold frame actuator as an emergency plan;

[0141] (3) Push emergency plan alarms.

[0142] like Figure 3 As shown in this embodiment, the AE-LSTM hybrid model dynamically adjusts the weight factors of SHI, which can increase the weight of corresponding parameters under complex working conditions. SHI provides feature engineering support for the AE-LSTM hybrid model. Weakly correlated features that are difficult to detect in the data, such as short-term fluctuations in outrigger settlement rate, are quantified through the deformation deviation component of SHI and become high-value input features of the AE-LSTM hybrid model. When hydraulic correction is triggered, the execution effect data is fed back to the AE-LSTM hybrid model for online learning.

[0143] The method in this embodiment was specifically applied to the construction of cast-in-place box girders using mobile formwork, reducing the false alarm rate from 18% to 5% and increasing the response speed by 36 times.

[0144] Under conditions of a Category 14 typhoon, normal operations would involve an emergency shutdown for a period of time. However, based on the AE-LSTM hybrid model prediction, the environmental risk value of SHI... Weighting factors The weighting was adjusted from 0.3 to 0.6 to strengthen the environmental risk weighting. Both the instability probability and SHI exceeded the corresponding thresholds, triggering the formulation of emergency plans. The counterweight requirements were automatically calculated and the counterweight was dynamically adjusted, and no work stoppage occurred.

[0145] In complex working conditions involving small-radius curves through holes, manual judgment and adjustment are usually the only options, with an error of ±10mm. By applying this method, the instability probability can be predicted and the SHI calculation results can be used to trigger the formulation of emergency plans, with an automatic correction error of ±1.5mm.

[0146] Example 2:

[0147] This embodiment proposes an intelligent monitoring system for mobile formwork based on digital twin and multimodal perception, including:

[0148] Multimodal sensor network: Establish a multimodal sensor network on the moving frame entity;

[0149] Digital twin platform: Construct a BIM model that maps to the physical moving formwork, embed virtual monitoring points in the BIM model, match and calibrate with sensors of a multimodal sensor network, and fuse multi-source sensor data collected by the multimodal sensor network in real time;

[0150] SHI Calculation Module: Based on the multi-source sensor data, calculates the Construction Health Index (SHI) in real time;

[0151] AE-LSTM hybrid model prediction module: Based on the multi-source sensor data, it predicts the instability probability in real time through the AE-LSTM hybrid model; at the same time, it predicts the weighting factors of each parameter of the construction health index SHI for the next calculation of the construction health index SHI.

[0152] Early warning assessment module: Based on the construction health index SHI and the predicted instability probability, jointly assess whether to generate a moving formwork adjustment instruction and issue an early warning.

[0153] The multimodal sensor network includes:

[0154] Fiber grating sensors are installed at the mid-span and support points of the main beam of the moving formwork to monitor the stress of the main beam.

[0155] Inclination sensors are installed at the four corners of the top plate of the movable mold frame to monitor the lateral tilt of the mold frame;

[0156] An array of acoustic sensors is installed inside the formwork of the movable formwork to monitor the compaction of the concrete during vibration.

[0157] Laser displacement gauges are installed at the bottom of the support legs of the movable formwork to monitor the settlement of the support legs;

[0158] A temperature, humidity and anemometer is installed on the top of the mobile mold frame to monitor environmental parameters;

[0159] Industrial cameras are installed at key joints of the moving formwork to monitor misalignment of the formwork seams.

[0160] Digital twin platforms include:

[0161] Spatial binding unit: Establish the association between the GIS coordinates of each sensor in the multimodal sensor network and the position of the moving module components of the BIM model, so as to realize the spatial binding of monitoring data and BIM model;

[0162] Monitoring and marking unit: Creates monitoring point families for different parameters, marks different sensor types, and labels key sensor information;

[0163] Real-time association unit: The sensor data stream is associated with the BIM model moving module components in real time using the MQTT protocol.

[0164] The SHI calculation module includes:

[0165] ;

[0166] SHI represents the construction health index; Represents the stress safety factor. Weighting factors representing the stress safety factor; Represents the degree of deformation deviation. Weighting factor representing the degree of deformation deviation; Represents environmental risk value, Weighting factors representing environmental risk values; , , Calculated in real time based on data from multiple sensor sources; Real-time updates are achieved through predictions using an AE-LSTM hybrid model.

[0167] The AE-LSTM hybrid model prediction module includes:

[0168] First training unit: Fix the LSTM module and train the AE module to enable the AE module to effectively extract key features from historical multi-source sensor data;

[0169] Second training unit: Load the AE module, connect the LSTM module, train using historical data containing instability events, and output the instability probability. The instability event is the instability value of the corresponding historical multi-source sensor data at the next k time steps.

[0170] The prediction unit predicts the instability probability based on real-time received multi-source sensor data through the AE module and LSTM module.

[0171] The AE-LSTM hybrid model prediction module also includes:

[0172] Using historical multi-source sensor data as input data, feature engineering is performed on the input data corresponding to various parameters of the Construction Health Index (SHI). From the historical multi-source sensor data, the data corresponding to the SHI stress safety factor parameter is used to calculate the temporal rate of change of its stress value; the data corresponding to the SHI deformation deviation parameter is used to form a spatial gradient distribution based on its deformation data; and the data corresponding to the SHI environmental risk value parameter is used to calculate the coupling oscillation frequency of the environmental parameters. The temporal rate of change of the stress value, the spatial gradient distribution, and the coupling oscillation frequency of the environmental parameters are used as feature vectors for training.

[0173] The AE-LSTM hybrid model introduces an attention layer to dynamically adjust the weights of each feature vector;

[0174] The output is the weighting factors of each parameter of the Construction Health Index (SHI).

[0175] The early warning assessment module includes:

[0176] Set an instability probability threshold and a SHI threshold. When the instability probability predicted by the AE-LSTM hybrid model is greater than the instability probability threshold, and the SHI calculated based on the weighting factor predicted by the AE-LSTM hybrid model is greater than the SHI threshold, a moving formwork adjustment command is generated and an early warning is issued.

[0177] The early warning assessment module includes:

[0178] Risk areas are identified on the digital twin platform, a hydraulic correction scheme for the mobile formwork actuator is developed, and emergency plan alerts are pushed out.

[0179] The early warning assessment module also includes: the AE-LSTM hybrid model makes predictions every 30 seconds, and the SHI calculation is performed once every 1 second. Each SHI calculation uses the weight factors predicted by the current AE-LSTM hybrid model until the weight factors are updated after the next AE-LSTM hybrid model predicts the weight factors.

[0180] The intelligent monitoring system for mobile formwork based on digital twin and multimodal perception proposed in this embodiment can realize the intelligent monitoring method for mobile formwork based on digital twin and multimodal perception described in Embodiment 1, and has the same technical effect as Embodiment 1.

[0181] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent monitoring of mobile formwork based on digital twin and multimodal perception, characterized in that, include: S1. Establish a multimodal sensor network on the moving mold frame entity; S2. Establish a digital twin platform: Construct a BIM model that maps to the physical moving formwork, embed virtual monitoring points in the BIM model, match and calibrate with the sensors of the multimodal sensor network, and fuse multi-source sensor data collected by the multimodal sensor network in real time. S3. Based on the multi-source sensor data, calculate the construction health index SHI in real time; S4. Based on the multi-source sensor data, the instability probability is predicted in real time using the AE-LSTM hybrid model; at the same time, the weighting factors of each parameter of the construction health index SHI are predicted for the next calculation of the construction health index SHI. S5. Based on the construction health index SHI and the predicted instability probability, jointly assess whether to generate a moving formwork adjustment instruction and issue an early warning. The method for pre-embedding virtual monitoring points in the BIM model in step S2 includes: S201. Establish the association between the GIS coordinates of each sensor in the multimodal sensor network and the location of the moving module components of the BIM model, so as to realize the spatial binding of monitoring data and BIM model; S202. Create monitoring point families for different parameters, mark and distinguish different sensor types, and label key sensor information; S203. Use the MQTT protocol to associate sensor data streams with BIM model moving module components in real time; The calculation method for the construction health index SHI mentioned in step S3 includes: ； SHI represents the construction health index; Represents the stress safety factor. Weighting factors representing the stress safety factor; Represents the degree of deformation deviation. Weighting factor representing the degree of deformation deviation; Represents environmental risk value, Weighting factors representing environmental risk values; , , Calculated in real time based on data from multiple sensor sources; Real-time updates are achieved through predictions using an AE-LSTM hybrid model. Step S4 involves real-time prediction of the instability probability using the AE-LSTM hybrid model, including: S401: A fixed LSTM module is used to train the AE module with historical multi-source sensor data, enabling the AE module to effectively extract key features from the multi-source sensor data. S402. Load the AE module, connect the LSTM module, train it using historical data containing instability events, and output the instability probability. The instability event is the instability value of the corresponding historical multi-source sensor data at the next k time steps. S403: Based on real-time received multi-source sensor data, the instability probability is predicted through the AE module and LSTM module; Training the AE-LSTM hybrid model also includes: Using historical multi-source sensor data as input data, feature engineering is performed on the input data corresponding to various parameters of the Construction Health Index (SHI). From the historical multi-source sensor data, the data corresponding to the SHI stress safety factor parameter is used to calculate the temporal rate of change of its stress value; the data corresponding to the SHI deformation deviation parameter is used to form a spatial gradient distribution based on its deformation data; and the data corresponding to the SHI environmental risk value parameter is used to calculate the coupling oscillation frequency of the environmental parameters. The temporal rate of change of the stress value, the spatial gradient distribution, and the coupling oscillation frequency of the environmental parameters are used as feature vectors for training. The AE-LSTM hybrid model introduces an attention layer to dynamically adjust the weights of each feature vector; The output is the weighting factors of each parameter of the Construction Health Index (SHI). Step S5 includes: Set an instability probability threshold and a SHI threshold. When the instability probability predicted by the AE-LSTM hybrid model is greater than the instability probability threshold, and the SHI calculated based on the weighting factor predicted by the AE-LSTM hybrid model is greater than the SHI threshold, a moving formwork adjustment command is generated and an early warning is issued.

2. The intelligent monitoring method for moving formwork based on digital twin and multimodal perception according to claim 1, characterized in that, In step S1, the establishment of the multimodal sensor network includes: S101. Fiber optic grating sensors are installed at the mid-span and support points of the main beam of the movable formwork to monitor the stress of the main beam. S102. Install tilt sensors at the four corners of the top plate of the movable mold frame to monitor the lateral tilt of the mold frame; S103. An array of acoustic sensors is installed inside the formwork of the movable formwork to monitor the compaction of the concrete. S104. Install laser displacement gauges at the bottom of the support legs of the movable formwork to monitor the settlement of the support legs; S105. Install a temperature, humidity and anemometer on the top of the movable mold frame to monitor environmental parameters; S106. Industrial cameras are installed at the critical joints of the moving formwork to monitor misalignment of the formwork joints.

3. The intelligent monitoring method for mobile scaffolding based on digital twin and multimodal perception according to claim 1, characterized in that, Generating and issuing early warning commands for moving formwork adjustment includes: Risk areas are identified on the digital twin platform, a hydraulic correction scheme for the mobile formwork actuator is developed, and emergency plan alerts are pushed out.

4. The intelligent monitoring method for mobile scaffolding based on digital twin and multimodal perception according to claim 1, characterized in that, Also includes: The AE-LSTM hybrid model makes predictions every 30 seconds, and the SHI calculation is performed every 1 second. Each SHI calculation uses the weight factors predicted by the current AE-LSTM hybrid model until the weight factors are updated after the next AE-LSTM hybrid model predicts the weight factors.

5. A mobile formwork intelligent monitoring system based on digital twin and multimodal perception, characterized in that, include: Multimodal sensor network: Establish a multimodal sensor network on the moving frame entity; Digital twin platform: Construct a BIM model that maps to the physical moving formwork, embed virtual monitoring points in the BIM model, match and calibrate with sensors of a multimodal sensor network, and fuse multi-source sensor data collected by the multimodal sensor network in real time; SHI Calculation Module: Based on the multi-source sensor data, calculates the Construction Health Index (SHI) in real time; AE-LSTM hybrid model prediction module: Based on the multi-source sensor data, it predicts the instability probability in real time through the AE-LSTM hybrid model; at the same time, it predicts the weighting factors of each parameter of the construction health index SHI for the next calculation of the construction health index SHI. Early warning assessment module: Based on the construction health index SHI and the predicted instability probability, jointly assess whether to generate a moving formwork adjustment instruction and issue an early warning; Digital twin platforms include: Spatial binding unit: Establish the association between the GIS coordinates of each sensor in the multimodal sensor network and the position of the moving module components of the BIM model, so as to realize the spatial binding of monitoring data and BIM model; Monitoring and marking unit: Creates families of monitoring points for different parameters, marks different sensor types, and labels key sensor information; Real-time association unit: Uses the MQTT protocol to associate sensor data streams with BIM model moving module components in real time; The SHI calculation module includes: ； SHI represents the construction health index; Represents the stress safety factor. Weighting factors representing the stress safety factor; Represents the degree of deformation deviation. Weighting factor representing the degree of deformation deviation; Represents environmental risk value, Weighting factors representing environmental risk values; , , Calculated in real time based on data from multiple sensor sources; Real-time updates are achieved through predictions using an AE-LSTM hybrid model. The AE-LSTM hybrid model prediction module includes: First training unit: Fix the LSTM module and train the AE module to enable the AE module to effectively extract key features from historical multi-source sensor data; Second training unit: Load the AE module, connect the LSTM module, train using historical data containing instability events, and output the instability probability. The instability event is the instability value of the corresponding historical multi-source sensor data at the next k time steps. The prediction unit predicts the instability probability based on real-time received multi-source sensor data through the AE module and LSTM module; The AE-LSTM hybrid model prediction module also includes: Using historical multi-source sensor data as input data, feature engineering is performed on the input data corresponding to various parameters of the Construction Health Index (SHI). From the historical multi-source sensor data, the data corresponding to the SHI stress safety factor parameter is used to calculate the temporal rate of change of its stress value; the data corresponding to the SHI deformation deviation parameter is used to form a spatial gradient distribution based on its deformation data; and the data corresponding to the SHI environmental risk value parameter is used to calculate the coupling oscillation frequency of the environmental parameters. The temporal rate of change of the stress value, the spatial gradient distribution, and the coupling oscillation frequency of the environmental parameters are used as feature vectors for training. The AE-LSTM hybrid model introduces an attention layer to dynamically adjust the weights of each feature vector; The output is the weighting factors of each parameter of the Construction Health Index (SHI). The early warning assessment module includes: Set an instability probability threshold and a SHI threshold. When the instability probability predicted by the AE-LSTM hybrid model is greater than the instability probability threshold, and the SHI calculated based on the weighting factor predicted by the AE-LSTM hybrid model is greater than the SHI threshold, a moving formwork adjustment command is generated and an early warning is issued.

Images