Construction monitoring system for horizontal warehouse arched roof based on multi-modal data fusion

Abstract

The invention discloses a construction monitoring system for a horizontal warehouse arched roof based on multi-modal data fusion, relates to the technical field of construction monitoring, and is used for solving the problem of poor structural construction anomaly recognition. The method comprises the steps of collecting multi-modal data and completing synchronous preprocessing, setting a sampling frequency according to modal characteristics, dividing construction stages, constructing a stage response function, dynamically updating during stage switching, fitting an actual observation value and a prediction value, extracting a modal response deviation degree, constructing a credibility function and evaluating data reliability. Further analyzing a residual trend, adjusting an abnormal score in combination with credibility, identifying a collaborative abnormal behavior between modals, systematically constructing a modal causal graph, identifying an abnormal propagation path, extracting modal conduction characteristic information, and triggering a risk response strategy according to the modal causal graph, so that dynamic tracking of the structural abnormality in the construction process of the horizontal warehouse arched roof is realized; and the accuracy of construction process monitoring identification is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of construction monitoring, more particularly, the present application relates to a multi-modal data fusion flat warehouse arch roof construction monitoring system. BACKGROUND

[0002] The flat warehouse arch roof is a kind of large-span light structure commonly used in grain storage, logistics and industrial warehouse, and the roof takes continuous arch steel truss or corrugated steel arch plate as the main force component, which has the advantages of reasonable stress, light weight, convenient construction, high space utilization rate, etc. The structure often adopts the construction technology of on-site segmented assembly and overall tensioning and sealing, and the arch rib is positioned by temporary support, and a stable stress system is gradually formed during construction. The arch roof is prone to problems such as incomplete structure connection state, complex temporary load action, various node structure forms, and uneven construction rhythm during construction, and is prone to high dynamicity of structure stress path and response form.

[0003] The existing technology has the following problems: the monitoring of the construction of the flat warehouse arch roof has the problems of incomplete structure state perception, rough abnormality identification mechanism and lack of systematicness of response decision, and only relies on single modal data (such as strain or displacement) for state judgment, and cannot fully integrate multi-source heterogeneous modalities such as inclination, temperature and image, so that in the process of frequent construction disturbance and rapid evolution of structure state, abnormal signals are easily covered or misjudged. At the same time, the existing system mainly uses fixed threshold or linear statistical model for abnormal detection, lacks the modeling ability of differentiated response behavior in the construction stage, and is difficult to adapt to the stage characteristics such as arch rib tensioning, node sealing and concrete curing, which leads to deviation of the judgment result from the true state, lack of explanation ability and diffusion path analysis means for the cause of abnormality, and cannot identify the abnormal starting modal and its propagation chain in the structure system, so as to support fine intervention, resulting in poor response effect of the dynamic grading control strategy based on risk potential evaluation, and affecting the construction progress. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the following scheme is provided to solve the problem of poor intelligent identification of structure construction abnormality in the background art.

[0005] In order to achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0006] A multi-modal data fusion flat warehouse arch roof construction monitoring system, comprising a roof construction data acquisition module, a construction stage division module, a monitoring analysis module, an abnormality analysis module and a strategy triggering module, and the modules are connected through signals;

[0007] The roof construction data collection module is used for collecting and synchronizing multi-modal data of the construction structure region, establishing a mapping relationship between the structure region and the modal collection, setting a sampling frequency according to the modal response characteristics, and pre-processing the data;

[0008] The construction phase division module is used for dividing the construction process phases, determining the modal response functions of each phase, and constructing the phase response functions to perform the response of the phase switching process.

[0009] The monitoring analysis module is used for fitting the actual observation data and the phase prediction value of each modal, determining the modal response deviation degree, and constructing a modal reliability function to evaluate the modal data.

[0010] The abnormality analysis module is used for extracting the modal response residual sequence and analyzing the change trend, identifying the mutation event, and adjusting the residual signal according to the modal reliability function to determine the abnormal situation of the cross-modal process.

[0011] The strategy triggering module is used for constructing a modal causal diagram and identifying an abnormal propagation path, obtaining and analyzing modal conduction characteristic information formed in the abnormal propagation process, and triggering a response strategy according to the analysis result.

[0012] In a preferred embodiment, the multi-modal data of the construction structure region is collected and synchronized, the mapping relationship between the structure region and the modal collection is established, the sampling frequency is set according to the modal response characteristics, and the data is pre-processed. The specific steps are as follows:

[0013] The internal strain data of the component is collected using a resistance strain gauge and a fiber Bragg grating, and the collected data is recorded as a strain modal;

[0014] The LVDT displacement gauge and the laser range finder are arranged at the center of the arch crown, the arch foot and the temporary support to collect the spatial displacement data of the component, and the collected data is recorded as a displacement modal;

[0015] The MEMS tilt angle instrument and the IMU are arranged on the upper chord and the top sealing segment of the component to collect the component attitude change data, and the collected data is recorded as a tilt angle modal;

[0016] The thermal imager is arranged in the closed area of the arch surface concrete to identify cement hydration heat anomalies or construction deviations, and the collected data is recorded as a thermal imaging modal;

[0017] The three-dimensional laser scanner is used in the overall arch surface geometry region to perform three-dimensional contour reconstruction and geometric anomaly identification, and the collected data is recorded as a laser point cloud modal;

[0018] The temperature and humidity meter and the anemometer are arranged above the arch support construction region to collect data, and the collected data is recorded as an environmental modal;

[0019] The master clock signal source is used to periodically synchronize each modal sensor and construct a unified data time axis. The synchronization nodes are aligned based on the PTP protocol or an external time reference. If some modal sensors have asynchronous sampling, a pseudo-synchronization processing mechanism is adopted.

[0020] In a preferred embodiment, the steps for dividing the construction process into stages and determining the modal response function of each stage include:

[0021] According to the construction organization design, the construction process of the arched roof is divided into a finite state set, which includes all the construction stages in the construction organization design. The stages include steel structure installation, arch rib tensioning, formwork closure, concrete pouring, and hydration curing.

[0022] Map each moment on the timeline to its corresponding stage state, serving as a label for the stage modeling entry point;

[0023] For each stage, a family of modal response functions is constructed, with inputs being modally related structural displacements, attitudes, or disturbance characteristics, and outputs being the structural response values ​​of the modes.

[0024] In a preferred embodiment, a stage response function is constructed to respond to the stage switching process. The specific steps are as follows:

[0025] The stage response function is constructed based on the structural geometric nonlinear mapping function, and the trend prediction, deviation judgment and anomaly identification of the structural response are performed based on the stage response function.

[0026] In each stage, indicators are extracted from the multimodal data of the structure, including the cumulative displacement of the crown, the maximum strain amplitude of the nodes, and the rate of change of the dip angle, and organized into a feature vector structure within the stage;

[0027] Modal response indices are constructed using third-order tensors and used for state description. The response state of each mode in the current stage is determined, and the stage transition is determined based on the changes in the response state.

[0028] In a preferred embodiment, the process of fitting actual observed data and stage prediction values ​​for each modality to determine the degree of deviation of the modal response and constructing a modal confidence function to evaluate the data of each modality includes the following steps:

[0029] Based on the stage response function, the difference between the current stage observation value and the stage prediction value of each mode is fitted to determine the trend consistency index;

[0030] Based on the fluctuation trend of modal values, the time-scale perturbation of the modal response derivative is constructed and used as a perturbation stability index to determine whether the mode is in an unstable jittering state.

[0031] Based on the modal coupling mapping path defined in the structural model, identify the set of coupled modes of the modes, determine the coupling residual for each coupling pair, and aggregate the coupling residuals as a coupling consistency index;

[0032] After extracting the three types of indicators—trend consistency, perturbation stability, and coupling consistency—for each mode, they are uniformly encapsulated and mapped to the current confidence value to construct the mode confidence function.

[0033] In a preferred embodiment, the steps for extracting the modal response residual sequence and analyzing its changing trend, identifying abrupt events, and adjusting the residual signal according to the modal confidence function to determine abnormal situations in the cross-modal process are as follows:

[0034] Obtain the current stage response function. For each mode, obtain the actual observed trajectory within the time interval and generate the theoretical predicted trajectory within the same interval. Use the difference between the two as the modal response residual sequence to represent the degree of deviation between the actual state and the expected structure.

[0035] Perform trend analysis on the modal residual sequence to determine if any anomalies exist;

[0036] Extract the rate of change and inflection point features from the residual sequence to determine whether there is a continuous shift, a sharp increase in inflection points, or amplified fluctuations. If an anomaly is found, mark the time interval in which the structural behavior deviates significantly from the expectation.

[0037] Based on the modal credibility function, the residual signal is adjusted to respond, and the time alignment relationship and trend consistency of the residual sequences between different modes are analyzed. If there are inflection points or fluctuation amplification anomalies that occur together between modes in the same period, they are judged as cooperative anomalies.

[0038] In a preferred embodiment, the specific steps for constructing a modal cause-effect graph and identifying anomaly propagation paths are as follows:

[0039] Construct a modal causal graph, which is a directed graph structure where the set of nodes represents all modes in the system and the set of edges represents potential structural or responsive causal relationships between modes.

[0040] The modal cause-effect graph construction process involves analyzing the structural topology, extracting modal pairs with rigid transmission paths, including strain and displacement, tilt angle and temperature, analyzing the modal coupling behavior in the modeled response, recording the order of mutual influence, constructing an initial graph, and assigning a directional label to each edge.

[0041] In a preferred embodiment, the modal propagation characteristic information formed during the anomaly propagation process is acquired and analyzed, and the specific steps are as follows:

[0042] Obtain the abnormal path structure in the modal cause-effect graph and analyze the modal propagation characteristics formed during the abnormal propagation process;

[0043] Modal transmission characteristics include anomaly path complexity index and anomaly source dominance index;

[0044] The anomaly path complexity index represents the complexity of the propagation structure of an anomaly event in a modal cause-effect graph.

[0045] The anomaly source dominance index indicates the control strength and concentration of the anomaly source mode in the anomaly propagation chain during structural multimodal monitoring.

[0046] Set the threshold for the dominant source of the anomaly and the threshold for the complexity of the anomaly path;

[0047] The threshold for the dominance of anomaly sources and the threshold for the complexity of anomaly paths are compared and analyzed with the index for the dominance of anomaly sources and the index for the complexity of anomaly paths, respectively.

[0048] In a preferred embodiment, a response strategy is triggered based on the analysis results, and the specific steps are as follows:

[0049] When the abnormal path complexity index is higher than the abnormal path complexity threshold, and the abnormal source dominance index is also higher than the abnormal source dominance threshold, a system-level linkage response strategy is adopted. Priority is given to investigating the identified root cause mode and its structural region, and enhanced monitoring and real-time response are carried out on the multi-path intersection mode simultaneously.

[0050] When the abnormal path complexity index is higher than the abnormal path complexity threshold, while the abnormal source dominance index is lower than the abnormal source dominance threshold, a regional partitioning monitoring strategy is adopted to perform parallel diagnosis on the abnormal path coverage area and track the changing trend of path coupling degree.

[0051] When the abnormal path complexity index is lower than the abnormal path complexity threshold, while the abnormal source dominance index is higher than the normal source dominance threshold, a fixed-point control response strategy is adopted to re-examine the structural unit where the abnormal starting point mode is located.

[0052] When the abnormal path complexity index is lower than the abnormal path complexity threshold and the abnormal source dominance index is lower than the normal source dominance threshold, no structural intervention measures are required, and the regular monitoring strategy should be maintained.

[0053] The technical effects and advantages of the multimodal data fusion construction monitoring system for the arched roof of a flat warehouse, as described in this invention:

[0054] This invention constructs a multimodal data fusion-based construction monitoring system to achieve precise identification and anomaly perception of the structural state of a flat warehouse arched roof during construction. The system collects and synchronizes multimodal structural data, establishes a mapping relationship between structural regions and modes, sets differentiated sampling frequencies based on modal response characteristics, and performs preprocessing. Combining the construction process into stages, it constructs modal response functions for each stage and dynamically reconstructs the structural response model during stage transitions. The system fits actual modal observation data with predicted values, extracts the degree of response deviation, constructs a modal credibility function, dynamically evaluates data reliability, further performs trend analysis on the modal residual sequence, adjusts residuals based on credibility, identifies structural mutation events and cross-modal collaborative anomalies, and finally, constructs a modal causal graph based on the coupling relationship between structural logic and modal response, identifies anomaly propagation paths, extracts modal transmission characteristics in the anomaly diffusion chain, forms a basis for structural risk judgment, and drives targeted response strategy output, achieving dynamic and intelligent monitoring of structural safety under complex construction conditions of flat warehouse arched roofs. Attached Figure Description

[0055] Figure 1 This is a structural schematic diagram of a construction monitoring system for a multimodal data fusion arched roof of a flat warehouse, according to the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] In order to achieve the above objectives, Figure 1 A structural schematic diagram of a construction monitoring system for a multimodal data fusion arched roof of a flat warehouse is provided. Specifically, it includes a roof construction data acquisition module, a construction stage division module, a monitoring and analysis module, an anomaly analysis module, and a strategy triggering module. The modules are connected by signals.

[0058] The roof construction data acquisition module is used to collect and synchronize multimodal data of the construction structure area, establish a mapping relationship between the structure area and the modal acquisition, set the sampling frequency according to the modal response characteristics, and perform data preprocessing.

[0059] The construction phase division module is used to divide the construction process into phases, determine the modal response function of each phase, construct the phase response function, and perform response during phase switching.

[0060] The monitoring and analysis module is used to fit the actual observed data of each mode with the stage prediction values, determine the degree of deviation of the modal response, construct the modal confidence function, and evaluate the data of each mode.

[0061] The anomaly analysis module is used to extract the modal response residual sequence and analyze the changing trend, identify abrupt events, and adjust the residual signal according to the modal confidence function to determine the anomalies in the cross-modal process.

[0062] The strategy triggering module is used to construct a modal cause-effect graph and identify the anomaly propagation path, obtain and analyze the modal transmission characteristic information formed during the anomaly propagation process, and trigger response strategies based on the analysis results.

[0063] Step 1: Conduct multimodal data acquisition and time synchronization of the construction structure area, and collect high-frequency data on the state evolution of the flat warehouse arched roof during construction. The specific steps are as follows:

[0064] Establishing a mapping relationship between the construction structure area and modal data involves mapping and matching the structural areas with the collected modal data based on the structural function division and construction risk distribution of the arched roof. An arched roof typically consists of a lower support, truss frame, arch rib connection nodes, capping section, and temporary supports. The following key responses will occur during construction:

[0065] Internal strain of components is used to capture the stress evolution of arch ribs and truss components under load. Resistance strain gauges and fiber optic gratings are used for data acquisition to determine local stress concentration and early crack development. The acquired data is recorded as strain modes.

[0066] For component spatial displacement, LVDT displacement meters and laser rangefinders are installed at the center of the arch, the arch foot, and the temporary support to determine the trend of large deformations such as arch settlement and lateral displacement, and the collected data are recorded as displacement modes.

[0067] The component attitude change (tilt angle) is detected by installing MEMS inclinometers and IMUs on the upper chord and top section components to identify the initial rotation of the support or the installation deviation of the component, and the collected data is recorded as incline mode.

[0068] Local temperature and imaging characteristics: A thermal imager is set up in the closed area of ​​the arch concrete to identify abnormal cement hydration heat or construction deviations. The collected data is used as a thermal imaging mode. A three-dimensional laser scanner is used to reconstruct the three-dimensional contour and identify geometric anomalies in the overall geometric area of ​​the arch. The collected data is recorded as a laser point cloud mode.

[0069] External environmental disturbances are investigated by installing thermometers, hygrometers, and anemometers above the arch frame construction area to determine the impact of sudden changes in wind load, humidity, or temperature on the component's condition, and the collected data is used as environmental modal data.

[0070] The data mapping strategy ensures that the sensing network can perform comprehensive monitoring of the structural physical state and environmental disturbances.

[0071] Because the physical quantity characteristics and frequency of change of different modes differ significantly, the system is configured with a reasonable sampling period and data caching mechanism for each type of mode to ensure data continuity and system resource balance:

[0072] Fast response modes such as strain / tilt are configured with high-frequency sampling of 10Hz to 20Hz to cope with sudden loads or dynamic disturbances, while slow variables such as displacement and temperature are configured with medium-low frequency sampling of 1Hz to 5Hz. Due to the large data volume, point cloud modes and image modes are configured to collect one frame every 10 minutes and dynamically scheduled in combination with structural stage triggering conditions. Environmental modes (wind speed, temperature and humidity) are configured with a stable sampling rate of once every 30 seconds to build disturbance baselines in real time.

[0073] In addition, during high-risk construction phases (such as hoisting and tensioning), the sampling frequency of key modes can be automatically increased to achieve a dynamic scheduling mechanism for sampling tasks.

[0074] Data collected by various modal sensors is processed for data time synchronization and pseudo-synchronization under a unified time base. That is, a master clock signal source (GPS / NTP) is used to periodically synchronize each sub-node and construct a unified data time axis.

[0075] Synchronization nodes are aligned based on the PTP protocol or an external time base. If some modalities (such as images) are sampled asynchronously, a pseudo-synchronization processing mechanism is adopted to construct the corresponding modal interpolation value or nearest neighbor matching value at the same time t, thereby achieving cross-modal data alignment and uniformly encapsulating the output data.

[0076] After completing the initial data acquisition and time alignment, the system uses statistical thresholds to identify and remove abrupt changes or signal spikes in each modal data. For rapidly changing modes (such as tilt angle and strain), it performs first-order low-pass filtering or sliding window mean filtering to eliminate spurious fluctuations caused by sensor jitter. For short-term packet loss segments, it performs linear or spline interpolation repair to improve continuity. In scenarios with multiple redundant sensor deployments (such as multiple displacement gauges), the system performs consistency verification on the acquired values, eliminates abnormal channels, and compresses and stores the preprocessed multimodal dataset.

[0077] Step two involves construction phase-driven feature modeling. Based on the phased and non-stationary characteristics of the arched roof structure construction process, a modal response modeling method adapted to the multi-stage evolution law is constructed. The specific content is as follows:

[0078] Because the stress boundary conditions and structural form of the arched roof change continuously in each stage of construction, the monitoring data of different stages have structural heteroscedasticity and response mode switching problems. Therefore, it is necessary to construct a set of construction stages based on the actual construction process logic, define an independent feature modeling function for each stage, and establish a cross-stage state transition mapping relationship as a benchmark reference for subsequent dynamic credibility judgment and anomaly scoring.

[0079] The construction phase set is defined and the process state is coded. Based on the construction organization design, the construction process of the arched roof is divided into a finite set of states: S = {S1, S2, ..., S...} K}, where each stage S k It represents a continuous time interval with independent structural objectives, stress characteristics and monitoring response features in the construction process. The finite state set includes all construction stages of the construction organization design, including: steel structure installation, arch rib tensioning, formwork closure, concrete pouring, hydration curing and other stages.

[0080] Define the state function: Z(t) → S k , That is, mapping each moment t on the time axis to the corresponding stage state S. k This serves as the label for the entry point in phase modeling, where K represents the total number of construction phases, determined by the construction organization's project plan; T... k For stage S k The time interval; Z(t) is the stage mapping function used to drive stage switching.

[0081] For each construction phase, the system constructs corresponding modal behavior modeling logic at the data processing layer, and models according to two core principles:

[0082] The core principle is based on structural constraint characteristics, namely, whether components are connected, fixed, or loaded at different stages, which has a decisive impact on their modal response curves. Therefore, the system determines the modal constraint direction, variable space, and response law based on the structural connection state.

[0083] The core of the evolution of intermodal coupling relationship is that as the structural integrity is enhanced, the synchronicity of the response curves between different modes (such as strain and displacement, displacement and tilt angle) is enhanced, and the system establishes an intermodal response chain as an auxiliary path for state identification.

[0084] For each stage S k Construct a family of modal response functions: N is the total number of stages, where Mode M i In stage S kThe response mapping in the diagram takes modally relevant structural displacement, attitude, or perturbation characteristics as input and outputs the structural response value for that mode. R represents the real number domain and is the set of all real numbers. Represents a d i A d-dimensional real vector space, where the input is a vector space of dimension d. i A real-valued vector;

[0085] Develop the construction phase response function based on a family of nonlinear functions: Where x is the input feature vector, such as the local node attitude vector, the displacement of neighboring nodes, and the wind load intensity; Set the weights for the structural response according to actual needs; The stage deformation compensation parameters are based on; σ geom (x) is a nonlinear mapping function based on structural geometry, representing the bending moment response of the arch surface at this stage;

[0086] The role of the construction phase response function is to establish a behavioral expectation model for the structural modes under each construction phase, so as to reflect the mapping relationship between the modal response and its input characteristics under the current structural form, force boundary and environmental disturbance conditions, thereby realizing the trend prediction, deviation judgment and anomaly identification of the structural response;

[0087] For example, during construction, the stress boundaries, connection states, and component integrity of the structure change continuously with each stage, resulting in strong non-stationarity in the modal response. By constructing stage response functions, a reference trajectory for the reasonable structural response of each mode at the current stage can be provided, enabling the system to clearly determine the amplitude, trend, and stability level of the current modal response, serving as the basis for subsequent judgments on whether the modal behavior deviates from the normal state.

[0088] It should be noted that a family of nonlinear functions refers to a set of function structures with the following characteristics: possessing nonlinear operator structures, including but not limited to hyperbolic functions such as tanh, sinh, and sigmoid, used to express nonlinear trends such as asymptotic boundary changes and smooth transitions; and power functions such as the 1.5 norm and piecewise power functions, used to amplify perturbations or compress anomalies. Structural response mapping weights map each component of the input modal feature vector to a single modal response value, used to characterize the fundamental influence strength and direction of different input factors on structural behavior. They can be initialized using unit vectors, or... A weighted approximation is constructed based on the modal projection information in the structural stiffness matrix. The stage deformation compensation parameter is used to compensate for the response offset between the stage structural state and the input modal features caused by factors such as loading, support boundaries, and assembly sequence. It corrects the modeling error caused by the non-equilibrium of modal response during structural stage evolution. The initial stage is set to 0, indicating no compensation. When the stage error residual is high, this value is adjusted to offset the structural error. It is fitted using the modal drift equilibrium point or error centroid between historical stages in the structural model. The structural geometric nonlinear mapping function maps the input feature x to a structural geometric response term. In the formula, x j For the j-th input feature, l j These are the structural geometric dimensions corresponding to the features (such as node spacing and arch rib radius, directly obtained from the CAD / BIM model), κ j This is the geometric nonlinearity adjustment factor in this modal direction, representing the degree of buckling / nonlinearity probability of the component in this direction.

[0089] In each stage, the system extracts multiple representative indicators (such as cumulative displacement of the arch, maximum strain amplitude of nodes, and rate of change of dip angle) from the multimodal data of the structure and organizes them into a feature vector structure within the stage. These feature vectors not only record the structural state of the current stage, but also form an evolution trajectory over time to describe the evolution trend of the structure from the beginning to the end of the stage. The system uses this to construct a staged structural state map, thereby providing support for dynamic monitoring and prediction.

[0090] A modal response index tensor is constructed for state description, representing the response state of each mode at the current stage, using a third-order tensor: Where i is the mode number; j is the feature channel number (e.g., displacement direction, tilt component); t is the time point; M is the input feature dimension of each mode; R(i, j, t) is the i-th mode at stage S. k The response value of the j-th channel, T k For stage S k The time interval.

[0091] The system performs dynamic responses during phase transitions; that is, when the system detects that construction has entered a new phase, it determines whether the requirements for transitioning from S... k To S k+1 Under the state transition conditions, the stage transition logic unit performs the following tasks: extracts the end feature vector of the previous stage and the initial feature vector of the current stage, analyzes whether there are abnormal jumps or lags in the evolution trend between the two, determines whether the modal response function should be reconstructed or continued (i.e., updated model or inherited model), and constructs a new response transmission diagram based on the modal coupling path to represent the modal cooperation relationship of the current stage.

[0092] Changes in response status are a key basis for judging stage transitions. After determining the current response status, the system can analyze in real time whether it has deviated from the behavior range described by the current stage model or whether it conforms to the response characteristics of the initial state of the next stage, thereby triggering the reconstruction or migration of the stage response function and realizing the stage switching and continuous adaptation of the construction state model.

[0093] Specifically, a sudden change in the trend of one or more modal responses (such as a significant change in slope), a reconstruction of the coupling relationship between multiple modes (such as a sudden drop in response synchronicity or path reconstruction), or a sudden and significant increase in discontinuous indicators (such as the number of inflection points, oscillation frequency, and frequency of positive and negative conversion of derivatives) indicates that the current structural state has changed significantly. This may be due to node behaviors such as connection completion, loading initiation, and closure completion, and is a direct physical signal for judging the stage switching.

[0094] In actual construction monitoring, the data quality collected by each mode varies significantly due to various factors. For example, the displacement mode may be affected by the loosening of the support or the offset of the laser; the thermal imaging mode may be affected by image noise caused by reflection, occlusion or rain and fog. Therefore, it is necessary to independently calculate the reliability of each mode at each moment based on the structural stage model and the behavior of the mode itself, so as to form a modal reliability vector with discernibility.

[0095] Step 3: Perform dynamic estimation of modal confidence. The specific steps are as follows:

[0096] According to the stage response function defined in step two (i.e., mode M) i S at the current stage k The response trend model is used to fit the difference between the current stage observation and the stage prediction for each mode, and to construct the trend deviation index MT. i (t), the process is as follows:

[0097] Get the current time t and its corresponding stage S k Call modal M i Stage response function at this stage For the current modal input feature vector x i (t), calculate the theoretically predicted response: The trend consistency indicator is: Among them, y i (t) represents mode M i The actual observed value at time t For the corresponding predicted value, ||·|| q Represents a nonlinear norm function (such as the 1.5 power norm);

[0098] The instability of modal response is not necessarily reflected in the magnitude of the response amplitude, but more often manifests as a sharp fluctuation or abrupt change in the direction of the derivative. Especially under conditions such as wind disturbance and construction vibration, some modal data will exhibit non-inertial oscillation characteristics, that is, the upward and downward fluctuations are asymmetrical and the direction is frequently reversed.

[0099] Therefore, based on the fluctuation trend of the modal values, a time-scale perturbation of the modal response derivative is constructed to determine whether the mode is in an unstable jittering state. The definition is as follows:

[0100] Within a sliding time window [t-τ, t], construct a discrete sequence of the first derivatives of the modal response:

[0101] Calculate the asymmetric derivative wave energy function as a perturbation stability index: Where D i (τ - ) represents a subset consisting only of terms with negative derivatives, ∈ is a small positive constant to avoid division by zero, if BD i If (t) deviates significantly from 1, then the mode is considered to have directional instability perturbation during that period.

[0102] In structural response logic, certain modes are explicitly coupled (e.g., tilt angle and displacement, strain and temperature). If the response residual between a certain mode and its coupled modes continues to increase, the response of that mode can be considered mismatched or distorted. A modal coupling residual tensor is defined, which identifies mode M based on the modal coupling mapping path defined in the structural model. i The set of coupled modes C(i), for each coupled pair (M) i M j Establish coupling residuals: XA ij (t)=CA(T(y i (t), y i (t))), where T is the dual-modal coupling trend deviation function (which can be the trend synchronization index, the sliding window curvature difference function, etc.), and CA(·) is the residual mapping after fitting.

[0103] Aggregate coupling residuals as a coupling consistency indicator: This index evaluates mode Mi The reliability of the coordination between its coupled modes is significantly reduced if it is inconsistent with the multimodal modes at the same time.

[0104] After extracting the three relevant indices—trend consistency, perturbation stability, and coupling consistency—for each mode, these three results are uniformly encapsulated and mapped to the current confidence value, thus constructing the mode confidence function C. i (t) is represented as: Where DY is a monotonically increasing compression function (such as the Sigmoid or SoftMin approximation function); γ1 and γ2 are model adjustment coefficients used to balance the relative influence of perturbation dominance and coupling error dominance; and C is guaranteed to be... i (t)∈(0,1) and is adapted to the physical properties of each mode;

[0105] The final output modal confidence function has a continuous and differentiable response, which facilitates gradient analysis in the subsequent anomaly scoring module. It also needs to ensure monotonicity. When the modal trend deviation, disturbance direction instability, or coupling mismatch is significant, the confidence is automatically compressed. Furthermore, it supports critical state labeling, which facilitates the system to prompt modal risks rather than just result risks.

[0106] Step 4: Anomaly joint scoring based on residuals. This involves combining the structural response function and modal confidence output results from the previous stage to perform multimodal residual analysis on various monitoring data during the construction of the arched roof. Based on this, an anomaly scoring mechanism for the structural state is constructed. This mechanism is not only used to identify single-modal anomaly mutations, but also focuses on deviations in the overall structural behavior caused by intermodal dysregulation, providing a scoring basis for subsequent causal path identification and risk assessment.

[0107] Modal response trajectory residuals are extracted. For each mode at the current construction stage, its structural response follows a modeled phased behavioral trend. Therefore, by comparing the actual observed trajectory with the predicted trajectory, the modal response residuals are extracted. The processing flow is as follows:

[0108] From step two, call the current stage response function to obtain the actual observed trajectory within a certain time interval for each mode; within the same interval, use the structural response function to generate the theoretical predicted trajectory, and use the difference between the two as the modal response residual sequence to represent the degree of deviation between the actual state and the structural expectation.

[0109] By constructing basic signals for anomaly identification, the actual magnitude and direction of modal deviation from the structural model can be reflected.

[0110] Anomalies are usually not caused by a single observation, but gradually emerge in the evolution trend of the residuals. Therefore, we need to analyze the trend of the modal residual sequence to determine whether there are abnormal rapid amplification, jumps, unstable segments, or other characteristics.

[0111] Extract the rate of change and inflection point features from the residual sequence to determine whether there is a continuous shift, a sharp increase in inflection points, or amplified fluctuations. If the above phenomena are found, mark the time interval in which the structural behavior deviates significantly from the model's expectations.

[0112] Based on the modality confidence function in step three, the residual signal is adjusted accordingly. That is, for modes with low confidence, the abnormal risk represented by their residuals will be increased, while for modes with high confidence, their residuals retain their original judgment power. If a mode has small residuals but extremely low confidence, the system will maintain the abnormal and suspicious state of that mode in the scoring.

[0113] Furthermore, to determine whether multiple modalities are incoordinating, multimodal anomaly cross-identification and structural scoring are performed. The specific steps are as follows:

[0114] Analyze the time alignment and trend consistency of residual sequences among different modes. If multiple modes exhibit inflection points or amplified fluctuations within the same time period, it is judged as a cooperative anomaly.

[0115] Output the anomaly score of the structure at this moment as the core risk quantification indicator of the current system. If the score exceeds the set threshold, the system will enter the causal path identification and risk classification process.

[0116] It should be noted that the structural anomaly score is a risk quantification index that reflects the overall deviation of the structure from the modeling expectation. It is formed by analyzing the response residuals, trend deviations, and intermodal mismatches of each mode at the current stage, and dynamically weighting them in conjunction with the modal credibility. It is used to represent the anomaly intensity and potential risk level of the structural state at the current moment.

[0117] Step 5 involves constructing a modal cause-effect graph and reasoning about anomaly paths. This means that when a system's structural anomaly score increases, the analysis determines which specific modality the anomaly first appears in, and further tracks whether the anomaly spreads from one modality to other modalities. The ultimate goal is to determine the starting modality and propagation path of the anomaly chain. The specific steps are as follows:

[0118] During structural response, the modes are not independent of each other. For example, strain is a local expression of the stress on the component, displacement is the geometric response of the component under stress, and tilt angle reflects the overall posture change of the component. There is a clear physical response chain among these three. In addition, during specific construction stages, such as concrete curing or tensioning, the sequence of different modal responses also has a phased pattern.

[0119] Construct a modal causal graph G = (V, E) that is a directed graph structure, where the set of nodes V = {M1, M2, ..., M} N} represents the set of all modalities and edges in the system. This indicates the existence of potential structural or responsive causal relationships between modes;

[0120] The mapping process can be as follows: analyze the topological relationships of the structure, extract mode pairs with rigid transmission paths, such as strain and displacement, tilt angle and temperature, analyze the modal coupling behavior in the response of the modeled stage, record their mutual influence order, construct an initial graph, and assign a directional label to each edge to indicate that the information disturbance is transmitted from mode i to mode j.

[0121] For example, during the construction of an arched roof, the system continuously monitors multiple modal signals, including arch foot displacement, mid-section strain, arch rib inclination angle, node temperature, arch crown vertical displacement, and image contour changes. Within a certain timeframe, the system first detects abnormal fluctuations in the mid-section strain mode, followed by a sudden change in the arch rib inclination angle, then a gradual exceedance of the arch crown vertical displacement, and simultaneously, a rapid rise in the node temperature. Ultimately, the image modality shows a significant distortion in the node contour. Based on a pre-constructed modal causal graph, the system automatically identifies multiple anomaly propagation paths with causal relationships, such as the mechanical propagation chain from strain to inclination angle to arch crown displacement, and the multimodal collaborative path from strain-induced temperature rise to ultimately leading to image anomalies. By comparing the triggering sequence and modal dependency structure of the anomaly signals, the system successfully identifies a set of anomaly paths and determines that the mid-section strain mode is the primary source of the anomaly, while the arch crown and node regions are concentrated areas of anomaly diffusion.

[0122] It should be noted that this figure is not based on observational statistical covariance, but rather on a priori coupling relationship map constructed based on structural design logic, loading path, construction sequence, and modal response stage model.

[0123] Obtain the abnormal path structure in the modal cause-effect graph and analyze the modal transmission characteristic information formed during the abnormal propagation process. The modal transmission characteristic information includes the abnormal path complexity index and the abnormal source dominance index.

[0124] The anomaly path complexity index is an indicator used to quantitatively express the complexity of the propagation structure of anomaly events in a modal causal graph. It describes the construction difficulty and information density of the path traversed by an anomaly from the initial mode to other modes in terms of topological hierarchy, branch structure, and modal penetration.

[0125] The anomaly path complexity index comprehensively analyzes attributes such as the length variation of each anomaly path, the existence of node loops within the path, and the intersection density of shared nodes between paths. It captures the potential coupling scale and structural chain complexity during anomaly propagation. A high anomaly path complexity index indicates that the anomaly state is not limited to a single mode or local disturbances between a few neighboring modes, but rather extends to different modes through multiple logical branches in the graph, forming a complex propagation network with high dimensions, multiple directions, and multiple intersections. Such anomalies typically indicate a potential systemic structural imbalance or instability trend, requiring broader monitoring and response of the entire system.

[0126] The logic for obtaining the complexity index of abnormal paths is as follows:

[0127] Identify the set of anomalous paths P = {P1, ..., P2} from the modal cause-effect graph. n}, where n is the total number of paths, and each path P k = (v1, v2, ..., v l ), is a directed node sequence;

[0128] For each path, calculate the maximum continuous depth change between nodes within the path. The calculation expression is: Among them, deg + , deg - These represent the out-degree and in-degree of a node, respectively, reflecting the path jump tension. i With v i-1 Let them be nodes i and i-1 respectively.

[0129] Obtain the number of duplicate nodes and the total number of nodes in the path. Use the ratio of the number of duplicate nodes to the total number of nodes in the path as the density value v of node overlap in the path. k Obtain the number of shared nodes and the path length, and use the ratio of the number of shared nodes to the path length as the path split value FL. k The complexity index of the abnormal path is calculated as follows:

[0130] The larger the anomaly path complexity index, the more complex and dispersed the anomaly's propagation path in the graph, and the easier it is for it to penetrate multiple modalities.

[0131] It should be noted that repeated nodes indicate that the path structure has loops or branching backflows. Out-degree refers to the number of directed edges from a certain node (modality) to other nodes, and in-degree refers to the number of directed edges from a certain node (modality).

[0132] In structural multimodal monitoring, the anomaly source dominance index indicates whether an anomaly is triggered by a specific modality node and spreads to other modalities, as well as the control strength and concentration of that source modality in the anomaly propagation chain.

[0133] The anomaly source dominance index measures dominance from three dimensions: first, the concentration of the starting mode in the anomaly path, reflecting whether the anomaly is dominated by a few modes; second, the information output intensity of the starting mode in the graph structure (high out-degree ratio), indicating that it has high propagation ability; and third, the compactness of the structural distribution among these starting modes. If the topological distance between modes is relatively close, it indicates that the anomaly may originate from the clustering effect of a specific structural sub-region.

[0134] When the anomaly source dominance index is high, it indicates that a key mode in the system is likely to be the direct cause of the risk. This mode or its region should be the primary focus of review or response, thereby effectively focusing the scope of investigation, saving diagnostic resources, and reducing response latency. Conversely, when the index is low, it indicates that the anomaly presents a state of multiple starting points, multiple paths, and low aggregation, with risk sources being more dispersed. The response strategy needs to take into account both comprehensiveness and redundancy.

[0135] The logic for obtaining the dominant index of the anomaly source is as follows:

[0136] Obtain the set R = {V1, ..., V} of modal nodes marked as starting points from the abnormal path set. m}, where m is the total number of nodes, and the node that is selected most frequently is used as the starting point of the path v. i The aggregation degree is calculated using the proportion of a mode in all paths as the starting point, and the expression is: In the formula, QP represents the starting frequency, and ZLJ represents the total number of paths;

[0137] Get the starting point v i Modal out-degree Cdeg(v i ) and modal in-degree Rdeg(v i The calculation expression for the initial structural compressive force is as follows: To obtain the average graph distance between the first nodes of a path, calculate the overlap density of the starting points. The calculation expression is: In the formula, d(v i v j Let be the shortest path distance between the starting point and the node in the graph. Calculate the anomaly source dominance index, expressed as:

[0138] A high anomaly source dominance index indicates the presence of a strong root cause mode in the system, whose anomalous behavior may dominate multimodal diffusion. A low anomaly source dominance index suggests that the anomaly may be triggered by multiple weak points and that the system has a high risk of sporadic occurrence.

[0139] An anomaly path complexity threshold is set to determine whether the anomaly propagation structure exceeds the system's acceptable complexity limit and whether a local disturbance has evolved into a system-level diffusion risk. An anomaly source dominance threshold is set to determine whether there is a centralized dominant mode as the root cause of the anomaly, so as to support whether to adopt a key localized response or a decentralized parallel monitoring strategy. The anomaly path complexity index and the anomaly source dominance index are compared and analyzed respectively.

[0140] When the anomaly path complexity index exceeds its anomaly path complexity threshold, and the anomaly source dominance index also exceeds the anomaly source dominance threshold, it indicates that not only is there a structurally complex anomaly propagation chain in the system, but the anomaly is also triggered by a few key modes. This situation usually manifests as a composite risk characteristic of single-point dominance and wide-area diffusion, meaning that the anomaly has a clear source and has penetrated multiple modal paths, which may cause systemic linkage instability. Therefore, in such scenarios, a system-level linkage response strategy should be adopted, prioritizing the investigation of the identified root cause modes and their structural regions, and simultaneously enhancing the monitoring and real-time response of multi-path intersection modes to prevent local anomalies from spreading into overall structural risks.

[0141] When the anomaly path complexity index exceeds the anomaly path complexity threshold, while the anomaly source dominance index is below the anomaly source dominance threshold, it indicates that although the anomaly has formed a complex propagation pattern within the structural scope, it does not exhibit a clear centralized starting point or dominant mode. This type of anomaly is often driven by external disturbances, overall structural fatigue, or multiple weak anomalies, and is not easily controlled through single-point investigation. In this case, a regional zoning monitoring strategy should be adopted to perform parallel diagnosis of multiple anomaly path coverage areas to prevent further spread of systemic risks, supplemented by a dynamic risk hotspot marking mechanism to track the changing trend of path coupling degree.

[0142] When the anomaly path complexity index is lower than the anomaly path complexity threshold, while the anomaly source dominance index is higher than the normal source dominance threshold, it indicates that the current anomaly has not spread widely in the structural system, but a single or a few root cause modes with significant control capabilities have been identified. This situation belongs to the state of being triggered by a local high-risk point. It is suitable to adopt a fixed-point control response strategy, concentrate efforts on accurately re-inspecting, temporarily reinforcing, or suspending construction operations on the structural unit where the anomaly starting point mode is located, block the potential anomaly spread chain, and effectively curb the risk from developing from local to global.

[0143] When both indices are below their respective thresholds, the system as a whole is in a state of controllable disturbance. The anomaly has not formed a significant structural path propagation, nor has it shown the concentrated driving characteristics of the dominant mode. Such situations are mostly local micro-disturbances or instantaneous deviations. The system does not need to take structural intervention measures immediately. At this time, the conventional monitoring strategy can be maintained to continue to track the modal response and path trend. The trend judgment mechanism can be used to dynamically assess whether the response level needs to be increased, so as to ensure efficient allocation of construction monitoring system resources and avoid false triggering.

[0144] Based on the threshold comparison results, the current risk level is determined (e.g., low risk, localized concentrated risk, dispersed and diffused risk, global high risk, etc.), and this level is mapped to the structural space to generate structural partition identifiers with risk level labels. According to the determined risk level, the corresponding set of response strategies is matched, including whether to trigger early warning prompts, whether to initiate local re-inspection or modal recalibration, whether to require suspension of construction or loading processes in a certain area, and whether to enter the full structure emergency response state. At the same time, the system needs to convert the selected strategies into executable plans.

[0145] In summary, this invention constructs a multimodal data fusion-based construction monitoring system to achieve precise identification and anomaly perception of the structural state of a flat warehouse arched roof during construction. The system collects and synchronizes multimodal structural data, establishes a mapping relationship between structural regions and modes, sets differentiated sampling frequencies based on modal response characteristics, and performs preprocessing. Combining the construction process into stages, it constructs modal response functions for each stage and dynamically reconstructs the structural response model during stage switching. The system fits actual modal observation data with predicted values, extracts the degree of response deviation, constructs a modal reliability function, dynamically evaluates data reliability, further performs trend analysis on the modal residual sequence, adjusts residuals based on reliability, identifies structural mutation events and cross-modal collaborative anomalies, and finally, constructs a modal causal graph based on the coupling relationship between structural logic and modal response, identifies anomaly propagation paths, extracts modal transmission characteristics in the anomaly diffusion chain, forms a basis for structural risk judgment, and drives targeted response strategy output, achieving dynamic and intelligent monitoring of structural safety under complex construction conditions of flat warehouse arched roofs.

[0146] The above formulas are all dimensionless calculations. The formulas are derived from software simulations using a large amount of collected data, and are the closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0147] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0148] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0149] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0150] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0151] In conclusion, the above description is only a preferred embodiment of the present invention and is 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 construction monitoring system for a flat warehouse arched roof based on multimodal data fusion, characterized in that: It includes a roof construction data acquisition module, a construction stage division module, a monitoring and analysis module, an anomaly analysis module, and a strategy triggering module. The modules are connected by signals. The roof construction data acquisition module is used to collect and synchronize multimodal data of the construction structure area, establish a mapping relationship between the structure area and the modal acquisition, set the sampling frequency according to the modal response characteristics, and perform data preprocessing. The construction phase division module is used to divide the construction process into phases, determine the modal response function of each phase, construct the phase response function, and perform response during phase switching. The monitoring and analysis module is used to fit the actual observed data of each mode with the stage prediction values, determine the degree of deviation of the modal response, construct the modal confidence function, and evaluate the data of each mode. The anomaly analysis module is used to extract the modal response residual sequence and analyze the changing trend, identify abrupt events, and adjust the residual signal according to the modal confidence function to determine the anomalies in the cross-modal process. The strategy triggering module is used to construct a modal cause-effect graph and identify the anomaly propagation path, obtain and analyze the modal transmission characteristic information formed during the anomaly propagation process, and trigger response strategies based on the analysis results.

2. The construction monitoring system for a multimodal data fusion arched roof of a flat warehouse according to claim 1, characterized in that: This method is used to collect and synchronize multimodal data of the construction structure area, establish a mapping relationship between the structure area and modal acquisition, set the sampling frequency according to the modal response characteristics, and perform data preprocessing. The specific steps are as follows: The internal strain data of the component is acquired using resistance strain gauges and fiber optic gratings, and the acquired data is recorded as strain modes. LVDT displacement meters and laser rangefinders were installed at the center of the arch, the arch foot, and the temporary supports to collect spatial displacement data of the components, and the collected data was recorded as displacement modes. MEMS inclinometers and IMUs are installed on the upper chord and the capping section to collect data on the attitude changes of the components, and the collected data is recorded as inclinometer modes. A thermal imager is installed in the closed area of ​​the arch concrete to identify abnormal cement hydration heat or construction deviations, and the collected data is used as a thermal imaging mode. A 3D laser scanner was used to reconstruct the 3D contour and identify geometric anomalies in the overall arched geometric area, and the collected data was recorded as laser point cloud modal. Temperature and humidity meters and anemometers were installed above the arch frame construction area to collect data, and the collected data was used as an environmental modality. The master clock signal source is used to periodically synchronize each modal sensor and construct a unified data time axis. The synchronization nodes are aligned based on the PTP protocol or an external time reference. If some modal sensors have asynchronous sampling, a pseudo-synchronization processing mechanism is adopted.

3. The construction monitoring system for a multimodal data fusion arched roof of a flat warehouse according to claim 2, characterized in that: The steps for dividing the construction process into stages and determining the modal response function of each stage include: According to the construction organization design, the construction process of the arched roof is divided into a finite state set, which includes all the construction stages in the construction organization design. The stages include steel structure installation, arch rib tensioning, formwork closure, concrete pouring, and hydration curing. Map each moment on the timeline to its corresponding stage state, serving as a label for the stage modeling entry point; For each stage, a family of modal response functions is constructed, with inputs being modally related structural displacements, attitudes, or disturbance characteristics, and outputs being the structural response values ​​of the modes.

4. The construction monitoring system for a multimodal data fusion arched roof of a flat warehouse according to claim 3, characterized in that: A stage response function is constructed to respond to the stage switching process. The specific steps are as follows: The stage response function is constructed based on the structural geometric nonlinear mapping function, and the trend prediction, deviation judgment and anomaly identification of the structural response are performed based on the stage response function. In each stage, indicators are extracted from the multimodal data of the structure, including the cumulative displacement of the crown, the maximum strain amplitude of the nodes, and the rate of change of the dip angle, and organized into a feature vector structure within the stage; Modal response indices are constructed using third-order tensors and used for state description. The response state of each mode in the current stage is determined, and the stage transition is determined based on the changes in the response state.

5. The construction monitoring system for a multimodal data fusion arched roof of a flat warehouse according to claim 4, characterized in that: This method is used to fit the actual observed data of each modality with the stage prediction values, determine the degree of deviation of the modal response, construct the modal confidence function, and evaluate the data of each modality, including the following steps: Based on the stage response function, the difference between the current stage observation value and the stage prediction value of each mode is fitted to determine the trend consistency index; Based on the fluctuation trend of modal values, the time-scale perturbation of the modal response derivative is constructed and used as a perturbation stability index to determine whether the mode is in an unstable jittering state. Based on the modal coupling mapping path defined in the structural model, identify the set of coupled modes of the modes, determine the coupling residual for each coupling pair, and aggregate the coupling residuals as a coupling consistency index; After extracting the three types of indicators—trend consistency, perturbation stability, and coupling consistency—for each mode, they are uniformly encapsulated and mapped to the current confidence value to construct the mode confidence function.

6. The construction monitoring system for a multimodal data fusion arched roof of a flat warehouse according to claim 5, characterized in that: This method is used to extract modal response residual sequences, analyze their changing trends, identify abrupt events, and adjust the residual signals according to the modal confidence function to determine abnormal situations in cross-modal processes. The specific steps are as follows: Obtain the current stage response function. For each mode, obtain the actual observed trajectory within the time interval and generate the theoretical predicted trajectory within the same interval. Use the difference between the two as the modal response residual sequence to represent the degree of deviation between the actual state and the expected structure. Perform trend analysis on the modal residual sequence to determine if any anomalies exist; Extract the rate of change and inflection point features from the residual sequence to determine whether there is a continuous shift, a sharp increase in inflection points, or amplified fluctuations. If an anomaly is found, mark the time interval in which the structural behavior deviates significantly from the expectation. Based on the modal credibility function, the residual signal is adjusted to respond, and the time alignment relationship and trend consistency of the residual sequences between different modes are analyzed. If there are inflection points or fluctuation amplification anomalies that occur together between modes in the same period, they are judged as cooperative anomalies.

7. The construction monitoring system for a multimodal data fusion arched roof of a flat warehouse according to claim 6, characterized in that: The specific steps for constructing modal cause-effect graphs and identifying anomaly propagation paths are as follows: Construct a modal causal graph, which is a directed graph structure where the set of nodes represents all modes in the system and the set of edges represents potential structural or responsive causal relationships between modes. The modal cause-effect graph construction process involves analyzing the structural topology, extracting modal pairs with rigid transmission paths, including strain and displacement, tilt angle and temperature, analyzing the modal coupling behavior in the modeled response, recording the order of mutual influence, constructing an initial graph, and assigning a directional label to each edge.

8. The construction monitoring system for a multimodal data fusion arched roof of a flat warehouse according to claim 7, characterized in that: The specific steps for acquiring and analyzing the modal propagation characteristics formed during anomaly propagation are as follows: Obtain the abnormal path structure in the modal cause-effect graph and analyze the modal propagation characteristics formed during the abnormal propagation process; Modal transmission characteristics include anomaly path complexity index and anomaly source dominance index; The anomaly path complexity index represents the complexity of the propagation structure of an anomaly event in a modal cause-effect graph. The anomaly source dominance index indicates the control strength and concentration of the anomaly source mode in the anomaly propagation chain during structural multimodal monitoring. Set the threshold for the dominant source of the anomaly and the threshold for the complexity of the anomaly path; The threshold for the dominance of anomaly sources and the threshold for the complexity of anomaly paths are compared and analyzed with the index for the dominance of anomaly sources and the index for the complexity of anomaly paths, respectively.

9. The construction monitoring system for a multimodal data fusion arched roof of a flat warehouse according to claim 8, characterized in that: The response strategy is triggered based on the analysis results, and the specific steps are as follows: When the abnormal path complexity index is higher than the abnormal path complexity threshold, and the abnormal source dominance index is also higher than the abnormal source dominance threshold, a system-level linkage response strategy is adopted. Priority is given to investigating the identified root cause mode and its structural region, and enhanced monitoring and real-time response are carried out on the multi-path intersection mode simultaneously. When the abnormal path complexity index is higher than the abnormal path complexity threshold, while the abnormal source dominance index is lower than the abnormal source dominance threshold, a regional partitioning monitoring strategy is adopted to perform parallel diagnosis on the abnormal path coverage area and track the changing trend of path coupling degree. When the abnormal path complexity index is lower than the abnormal path complexity threshold, while the abnormal source dominance index is higher than the normal source dominance threshold, a fixed-point control response strategy is adopted to re-examine the structural unit where the abnormal starting point mode is located. When the abnormal path complexity index is lower than the abnormal path complexity threshold and the abnormal source dominance index is lower than the normal source dominance threshold, no structural intervention measures are required, and the regular monitoring strategy should be maintained.

Images