Tower crane monitoring system in intelligent construction site project

Abstract

The invention discloses a tower crane monitoring system in an intelligent construction site project, which comprises a multi-modal sensor module, a digital twin modeling module, a dynamic risk map generation module and a display terminal, and is characterized in that the multi-modal sensor module covers the whole domain of a tower crane and multi-dimensional data of real-time operation of the tower crane, so that the problem of data isolation of a traditional system is solved; a tower crane three-dimensional dynamic model is constructed based on real-time multi-dimensional data, stress distribution and motion trail of a physical entity are simulated, the position prediction precision of a lifting hook is improved, and a three-dimensional risk thermodynamic diagram is generated through a dynamic risk map generation module, so that a worker can more visually check the risk level of each part of the tower crane, intelligent early warning is realized, and the working efficiency is improved. And the fault occurrence probability is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of tower crane monitoring, and in particular to a tower crane monitoring system in a smart construction site project. Background Art

[0002] Tower cranes are the most commonly used lifting equipment on construction sites. Also known as tower cranes, they are constructed in sections (referred to as "standard sections") and are used to lift materials such as steel bars, wood planks, concrete, and steel pipes. Tower cranes are an essential piece of equipment on construction sites.

[0003] Traditional tower crane monitoring systems rely primarily on manual inspections and regular maintenance, which presents the following key issues:

[0004] Delayed response: Manual inspections cannot monitor the operating status of tower cranes in real time, making it difficult to detect safety hazards in a timely manner, resulting in a high risk of accidents.

[0005] Data isolation: Sensor data is scattered and not systematically analyzed. It lacks the ability to integrate multi-source data and cannot provide effective support for decision-making.

[0006] Lack of blind spot monitoring: There is a lack of effective monitoring methods in key areas such as the top of the tower crane and the end of the boom, which makes it difficult to identify local stress concentration or mechanical failures in a timely manner.

[0007] Passive maintenance: Relying on a post-repair model, the fatigue life of key components cannot be predicted, resulting in frequent unplanned downtime and low construction efficiency. Summary of the Invention

[0008] The purpose of the present invention is to provide a tower crane monitoring system in a smart construction site project to address the above-mentioned problems. The system covers the entire tower crane area through a multimodal sensor module, and collects multi-dimensional data on the real-time operation of the tower crane to solve the data isolation problem of the traditional system. A three-dimensional dynamic model of the tower crane is constructed based on real-time multi-dimensional data to simulate the stress distribution and motion trajectory of the physical entity, thereby improving the prediction accuracy of the hook position. A three-dimensional risk heat map is generated through a dynamic risk map generation module, so that staff can more intuitively view the risk level of each part of the tower crane, realize intelligent early warning, and reduce the probability of failure.

[0009] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is as follows:

[0010] According to one aspect of the present invention, a tower crane monitoring system for a smart construction site project is provided, comprising:

[0011] Multimodal sensor module collects multi-dimensional data of tower crane operation in real time;

[0012] The digital twin modeling module builds a three-dimensional dynamic model of the tower crane based on real-time multi-dimensional data, simulating the stress distribution and motion trajectory of the physical entity;

[0013] The dynamic risk map generation module integrates real-time multi-dimensional data and the three-dimensional dynamic model of the tower crane to generate risk heat maps for each part of the tower crane and predict potential failure points;

[0014] The display terminal shows the three-dimensional dynamic model of the tower crane, the risk heat map of each part of the tower crane, and potential fault points.

[0015] Preferably, the multimodal sensor module includes a mechanical sensor, a visual sensor, an environmental sensor and a voiceprint sensor;

[0016] The mechanical sensors include load sensors, inclination sensors, flexible strain sensors and speed sensors, which are used to collect force changes and motion data of various components of the tower crane;

[0017] The visual sensor includes a binocular camera, a 360° panoramic camera and an infrared thermal imager, which are used to collect image data of the tower crane and its surroundings;

[0018] The environmental sensors include an ultrasonic anemometer, a temperature and humidity sensor, and a PM2.5 / dust sensor, which are used to collect environmental data of the tower crane and its surroundings;

[0019] The soundprint sensor includes an acoustic microphone array and a vibration soundprint sensor, which is used to collect abnormal noise and vibration data of various components of the tower crane.

[0020] Preferably, the digital twin modeling module includes a data preprocessing unit, a three-dimensional model building unit, a finite element analysis unit and a motion trajectory prediction unit;

[0021] The data preprocessing unit is used to filter, remove noise and standardize the raw data collected by the sensor;

[0022] The three-dimensional model building unit is used to build a three-dimensional model of the tower crane, establish a spatial coordinate system based on the three-dimensional model of the tower crane, and decompose the three-dimensional model of the tower crane;

[0023] The finite element analysis unit is used to calculate the stress conditions of various components of the tower crane;

[0024] The motion trajectory prediction unit is used to predict the motion trajectory of the tower crane hook, perform differential comparison with actual data, and detect abnormal deviation.

[0025] Preferably, the three-dimensional model building unit includes a geometric model generating subunit, a decomposing subunit and a coordinate system establishing subunit;

[0026] The geometric model generation subunit constructs a three-dimensional model of the tower crane based on the tower crane drawings and sensor collected data;

[0027] The decomposition subunit performs module decomposition on the three-dimensional tower crane model into multiple independent components, and assigns physical properties to each independent component;

[0028] The coordinate system establishing subunit constructs a spatial coordinate system based on the three-dimensional model of the tower crane.

[0029] Preferably, the finite element analysis unit includes a tower crane structure discretization subunit, a material property and boundary condition definition subunit, and a stress-strain calculation subunit;

[0030] The tower crane structure discretization sub-unit is meshed according to each independent component of the tower crane and discretized into a finite number of small units;

[0031] The material property and boundary condition definition subunit is used to assign material properties and constraint conditions to a finite number of small units;

[0032] The stress-strain calculation subunit is used to solve the stress-strain relationship of each small unit and simulate the stress state of the overall structure of the tower crane.

[0033] Preferably, the motion trajectory prediction unit includes a global-local coordinate system construction subunit, a prediction model construction subunit and an abnormal offset detection subunit;

[0034] The global-local coordinate system construction subunit is used to construct the global coordinate system of the tower crane and the local coordinate system of the hook;

[0035] The prediction model building subunit is used to build a dynamic model and solve the dynamic model to obtain the predicted position of the hook;

[0036] The abnormal deviation detection subunit is used to perform differential comparison with actual data to detect abnormal deviation.

[0037] Preferably, the dynamic risk map generation module includes a multi-source data fusion and feature extraction unit, a risk heat map generation unit and a fatigue life prediction unit;

[0038] The multi-source data fusion and feature extraction unit is used to align the sensor collected data in time and space, extract correlation features, establish physical and logical relationships between data, and dynamically adjust weights according to data quality and risk impact;

[0039] The risk heat map generation unit is used to generate a three-dimensional heat map of the tower crane;

[0040] The fatigue life prediction unit is used to predict the service life of key components of tower cranes.

[0041] Preferably, the multi-source data fusion and feature extraction unit includes a time synchronization subunit, a space alignment subunit and a feature correlation analysis subunit;

[0042] The time synchronization subunit is used to add a unified time stamp to the data collected by all sensors;

[0043] The spatial alignment subunit is used to map the sensor collected data into the three-dimensional coordinate system of the tower crane;

[0044] The feature association analysis subunit is used to mine cross-modal features through correlation analysis (such as Pearson correlation coefficient).

[0045] Preferably, the risk heat map generation unit includes a network model construction subunit and a dynamic update mechanism subunit;

[0046] The network model construction subunit is used for a spatiotemporal convolutional neural network model, and the spatiotemporal convolutional neural network model is used to output a three-dimensional heat map;

[0047] The dynamic update mechanism subunit is used to force the update of the output of the spatiotemporal convolutional neural network model when any sensor data suddenly changes.

[0048] Preferably, the fatigue life prediction unit includes a fatigue life prediction model construction subunit, and the fatigue life prediction model construction subunit is used to construct a fatigue life prediction model and output a remaining life report using the fatigue life prediction model.

[0049] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0050] The present invention covers the entire area of the tower crane through a multimodal sensor module, and collects multi-dimensional data on the real-time operation of the tower crane, solving the data isolation problem of the traditional system. Based on the real-time multi-dimensional data, a three-dimensional dynamic model of the tower crane is constructed to simulate the stress distribution and motion trajectory of the physical entity, thereby improving the prediction accuracy of the hook position. A three-dimensional risk heat map is generated through a dynamic risk map generation module, allowing staff to more intuitively view the risk levels of various parts of the tower crane, realize intelligent early warning, and reduce the probability of failure. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a functional structure diagram of the present invention;

[0052] Figure 2 It is a functional structure diagram of the digital twin modeling module of the present invention;

[0053] Figure 3 It is a functional structure diagram of the dynamic risk map generation module of the present invention. DETAILED DESCRIPTION

[0054] To make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention is further described below with reference to the accompanying drawings and by way of preferred embodiments. However, it should be noted that many of the details listed in this specification are merely provided to help the reader gain a thorough understanding of one or more aspects of the present invention, and these aspects of the present invention can be practiced even without these specific details.

[0055] See also Figures 1 to 3 The present invention provides a tower crane monitoring system for a smart construction site project. The technical solution is as follows:

[0056] A tower crane monitoring system for smart construction site projects includes a multimodal sensor module, a digital twin modeling module, a dynamic risk map generation module, and a display terminal. The multimodal sensor module includes a mechanical sensor, a visual sensor, an environmental sensor, and a voiceprint sensor, which collects multi-dimensional data on tower crane operation.

[0057] Mechanical sensors include load sensors, inclination sensors, flexible strain sensors and speed sensors. The load sensor is fixed at the hook connection or the wire rope tension monitoring point to monitor the weight of the hoisted object in real time to prevent overloading. The inclination sensors are fixed in the middle of the tower crane body, the root of the boom and the end of the balance arm to detect the overall inclination angle of the tower crane and the horizontality of the boom to prevent structural imbalance. The flexible strain sensor is fixed at the key stress points of the boom (such as the main chord or the web) to monitor the micro-strain of the metal structure and identify deformation caused by fatigue cracks or overload. The speed sensor is fixed on the base of the cockpit and the slewing support to collect the vibration frequency of the tower crane and judge the abnormality of the slewing mechanism or the risk of foundation settlement.

[0058] The visual sensors include binocular cameras, 360° panoramic cameras and infrared thermal imagers. The binocular cameras are fixed at the end of the boom and on the top of the cab to capture the hook swing trajectory, the status of the hoisted object and surrounding obstacles with stereo vision, and assist in anti-collision warning. The 360° panoramic camera is fixed at the top of the tower crane (above the balance arm) to monitor the overall operating environment of the tower crane and cover blind spots (such as behind the tower body and under the boom). The infrared thermal imager is fixed near heat-generating components such as motors and gearboxes to detect overheating of electrical equipment and prevent fire or mechanical failure.

[0059] Environmental sensors include ultrasonic anemometers, temperature and humidity sensors, and PM2.5 / dust sensors. The ultrasonic anemometer is fixed at the highest point of the tower crane (top of the balance arm) to measure wind speed and direction in real time, and automatically trigger limit protection in strong winds (such as prohibiting hoisting and lowering the boom height). Temperature and humidity sensors are fixed inside the cockpit and outside the electrical control box to monitor the temperature and humidity of the operating environment to prevent electronic components from getting damp or shutting down due to high temperatures. The PM2.5 / dust sensor is fixed in the middle of the tower (10-15 meters above the ground) to detect dust concentration on the construction site and to link the spray system to reduce dust pollution.

[0060] The soundprint sensor includes an acoustic microphone array and a vibration soundprint sensor. The acoustic microphone array is fixed to the outer ring of the slewing bearing and the hoist motor housing to collect the sounds of gear meshing and motor operation, and identify abnormal noise (such as bearing wear and gear tooth breakage) through spectrum analysis. The vibration soundprint sensor is fixed to the wire rope drum and pulley bracket to monitor the friction sound and vibration frequency of the wire rope, providing early warning of wire breakage or rope jumping risks.

[0061] Multiple sensors are installed throughout the tower crane's entire area (boom, tower, cockpit, and counter-jib). Through multi-tiered installation, comprehensive monitoring of the crane's structure, operating environment, and mechanical status is achieved. The sensors complement each other's functions. For example, visual sensors complement the mechanical sensors' limited spatial perception, while voiceprint sensors assist the mechanical sensors in identifying hidden faults (such as premature gear wear). Through the precise layout and coordinated operation of multimodal sensors, holographic perception of the tower crane's operating status is achieved, providing a reliable data foundation for proactive safety defenses.

[0062] Sensors deployed at key locations on the tower crane collect real-time mechanical, environmental, and visual data. This data is then transmitted to the digital twin modeling module, which includes a data preprocessing unit, a 3D modeling unit, a finite element analysis unit, and a motion trajectory prediction unit.

[0063] The data preprocessing unit uses edge computing nodes to filter, denoise, and standardize the raw data to eliminate noise interference and ensure the accuracy of the input data. Specifically, the filtering process uses three filtering methods, namely, Kalman filtering: applicable to dynamic systems (such as the motion state of tower cranes), reducing random noise through a prediction-correction mechanism, for example, used to process instantaneous jitter in tilt sensor data; low-pass filtering: filtering out noise that is higher than the normal vibration frequency of the tower crane (such as electromagnetic interference), applicable to acceleration sensor data; median filtering: for impulse noise (such as occasional abnormal values of the sensor), the median within the sliding window is used to replace the original value. After filtering, high-frequency noise and interference signals in the sensor data can be eliminated, retaining valid information. There are two types of noise, one is high-frequency noise, which comes from electrical interference or environmental vibration. The other is sensor drift, which is the zero point offset caused by long-term operation (such as the zero point drift of the load sensor). Denoising is done in three ways: wavelet denoising: separating noise frequency bands through wavelet transform, suitable for non-stationary signals (such as voiceprint sensor data); sliding average: smoothing short-term fluctuations, used for trend extraction of environmental sensors (such as temperature and humidity); adaptive filtering: dynamically adjusting parameters based on historical data to solve sensor drift problems. Data standardization is done in two ways: one is Z-score standardization, which converts data into a distribution with a mean of 0 and a standard deviation of 1. The specific formula is:

[0064]

[0065] Among them, x is the original data point, that is, the original value collected by a single sensor (such as the measurement value of a load sensor at a certain moment); μ is the arithmetic mean of all samples in the data set; σ is the degree of dispersion of the data distribution.

[0066] The Z-score standardization is applicable to sensors with different dimensions (e.g., load in tons, inclination in degrees).

[0067] The other is Min-Max scaling, which maps data to the interval [0, 1]. The specific formula is:

[0068]

[0069] Where x is the original data point (such as the measurement value of the temperature and humidity sensor at a certain moment); x min is the minimum value of all samples in the data set; x max is the maximum value of all samples in the dataset.

[0070] Min-Max scaling is used to normalize pixel values in vision sensors.

[0071] The above two data standardization processes eliminate dimensional differences and ensure the stability and convergence speed of machine learning model training.

[0072] Edge computing nodes are embedded devices (such as NVIDIA Jetson) that deploy filtering algorithms. The processing flow is as follows: Raw data reception: Raw data streams are acquired from sensors via LoRa or CAN buses. Real-time filtering and denoising: Preprocessing algorithms are run on edge nodes (with latency <50ms). Standardized output: Data packets are generated in a unified format and transmitted to the cloud or a local digital twin engine.

[0073] The 3D model construction unit constructs a high-precision 3D digital twin model based on the tower crane's CAD drawings and standardized real-time sensor data, and dynamically updates the model status (such as boom angle, load distribution, and base displacement). Specifically, 3D modeling software (such as SolidWorks or AutoCAD) is used to import the tower crane's CAD drawings to generate a basic geometric model. The tower crane is decomposed into independent components, such as the boom, counter-arm, and slewing bearing, and physical properties (such as material density and elastic modulus) are assigned to each component. A model coordinate system is established and calibrated with the sensor data coordinate system based on the actual installation location. Data communication between the sensor and the model uses the MQTT or OPC UA protocol to achieve real-time communication between sensor data and the model. A sensor ID is also set and bound to the model component; for example, inclination sensor ID = 001 corresponds to the boom root. In this embodiment, a dynamic update mechanism is provided. This dynamic update mechanism includes two processing methods: one is to set the update frequency: sensor data is received every 200ms, triggering a model status refresh. The other is exception handling: if data is lost, it is temporarily filled using historical data interpolation or model prediction values. A high-precision model calibration mechanism is also provided. The high-precision model calibration mechanism uses a 3D laser scanner to regularly obtain the actual shape of the tower crane, compares the obtained actual shape of the tower crane with the digital model, and corrects the deviation. The constructed model is transmitted to the display terminal, and the display terminal is used to present the three-dimensional model of the tower crane, allowing staff to view the model status more intuitively. In this embodiment, the dynamic parameters of the visual model are displayed as follows: boom angle: the model rotation angle is driven by the inclination sensor data. Load distribution: the force on each section of the boom is displayed in a color gradient (red = high stress, green = normal). Base displacement: combined with the displacement sensor data, the tower crane base position offset is dynamically rendered. Interactive function: supports zooming and rotating the view angle, and clicking on the component to view the real-time sensor value. The three-dimensional tower crane model constructed based on the tower crane drawings and sensor data makes the geometric error between the model and the actual tower crane less than 0.5%. The three-dimensional model can also be displayed to facilitate staff to view the tower crane status in real time.

[0074] Finite element analysis (FEM) is used to simulate the stress and strain distribution of the structure, calculate the forces acting on key components such as the boom and tower in real time, and identify areas of local stress concentration. By discretizing the tower crane structure into a finite number of small elements (such as tetrahedral or hexahedral meshes), the stress-strain relationship for each element is solved, combining material properties with boundary conditions to simulate the stress state of the entire structure. Specifically, the structure is first divided into a high-precision mesh (element size ≤ 10mm), ensuring a higher mesh density in critical areas (such as the boom joints). Steel parameters such as elastic modulus, Poisson's ratio, and yield strength are input. Static and dynamic loads are then applied. Static loads, such as the weight of the hoisted structure, can be input in real time via load sensors. Dynamic loads, such as wind load and inertial force, are applied. Wind loads calculate wind pressure based on anemometer data, while inertial forces are measured by the kinematic state of accelerometers. Constraints are then set, such as fixing the tower crane base and setting the balance arm hinge points as rotational degrees of freedom. Transient analysis is performed using the LS-DYNA dynamics algorithm to calculate stresses for each component. The stress distribution is displayed using a color gradient (red: >80% yield strength, yellow: 60%-80%, green: <60%). A gradient threshold method is used to locate areas of sudden stress changes (e.g., stress differences between adjacent elements >15%). If the calculated stress in a region exceeds the threshold three times in a row, it is marked as a high-risk point and an alarm is triggered.

[0075] The motion trajectory prediction unit combines the tower crane operating instructions (such as rotation speed, lifting height) and environmental parameters (wind speed, ground subsidence) to predict the motion trajectory of the hook in the next 5 seconds, and performs differential comparison with the actual data to detect abnormal offsets. Specifically, a local coordinate system is established based on the global coordinate system. For example, the hook position uses the base of the boom as a reference point to establish a local coordinate system and update it in real time with the rotation motion. A tower crane multi-body dynamics model is established based on the Newton-Euler equation:

[0076]

[0077] Where τ is the torque of the rotary motor (obtained from the operating instructions); J is the moment of inertia of the boom; ω is the angular velocity of rotation; F is the resultant force acting on the hook (including wind resistance and inertia); C(v) is the Coriolis force and centrifugal force; and G is gravity.

[0078] Based on the above dynamic model, the trajectory prediction process is as follows:

[0079] (1) Input parameter integration: such as input parameters such as rotation speed (rad / s), lifting height (m), amplitude angle (°), wind speed (m / s), wind direction (°), and ground settlement (mm).

[0080] (2) Prediction model construction: The fourth-order Runge-Kutta method (RK4) is used to solve the dynamic equations and predict the hook position within the next 5 seconds. Wind load is calculated based on wind speed and the projected area of the hoisted object, and the trajectory offset is corrected. Finally, the base coordinates are adjusted using displacement sensor data to avoid cumulative prediction errors.

[0081] (3) Trajectory output: The predicted three-dimensional coordinates (X, Y, Z) of the hook are output every 0.1 seconds. The predicted path and the actual path are displayed in the three-dimensional model. The predicted path can be represented by a blue dotted line, and the actual path can be represented by a red solid line.

[0082] Furthermore, the method also includes an abnormal deviation detection method, the steps of which are as follows:

[0083] Real-time data alignment: Match predicted coordinates with actual coordinates (obtained by GPS / visual sensors) by timestamp.

[0084] Deviation calculation: The deviation is calculated using the Euclidean distance deviation formula, as follows:

[0085]

[0086] Among them, X pre , Y pre , Z pre are the predicted coordinate values of the hook in three-dimensional space; X real , Y real , Z real They are respectively the current three-dimensional coordinates of the hook actually measured by the sensor.

[0087] Threshold setting: Set a static threshold (Δd>0.5m) or a dynamic threshold (based on the standard deviation of historical data, such as Δd>3σ). If the deviation exceeds the limit for five consecutive times, it is determined to be an abnormal deviation and triggers emergency braking.

[0088] The dynamic risk map generation module includes a multi-source data fusion and feature extraction unit, a risk heat map generation unit and a fatigue life prediction unit, which are used to generate a risk heat map.

[0089] The multi-source data fusion and feature extraction unit is used to align the sensor data in time and space, extract correlation features, establish physical and logical relationships between the data, and dynamically adjust the weights based on data quality and risk impact. The multi-source data fusion and feature extraction unit aligns heterogeneous data from different sensors (mechanical, visual, and environmental) in time and space, extracts correlation features, and establishes physical and logical relationships between the data. Specifically, it includes the following steps:

[0090] Time synchronization: All sensor data is timestamped to the millisecond, eliminating time skew through edge node clock synchronization protocols (such as NTP). For example, the 30 frames per second image from a binocular camera and the 10Hz sampling rate data from an inclinometer are aligned according to a time window.

[0091] Spatial alignment: Mapping sensor data to the crane's 3D coordinate system. For example, converting the hook's visual offset data to spatial coordinates relative to the crane base. Use a calibration plate or laser rangefinder to calibrate the spatial relationship between the camera and force sensor.

[0092] Feature correlation analysis: Cross-modal features are mined through correlation analysis (such as the Pearson correlation coefficient). For example, when the wind speed is greater than 8 m / s, the hook swing amplitude and the tilt angle change show a nonlinear positive correlation (modeled using polynomial regression).

[0093] Dynamic risk weight allocation dynamically adjusts the weight based on data quality (confidence) and risk impact. The specific steps are as follows:

[0094] Confidence assessment: Calculate the confidence level (0-1) based on the sensor's historical error rate (e.g., the load sensor's calibration error is ±2%). Formula:

[0095]

[0096] Among them, w conf is the confidence level.

[0097] Risk impact coefficient: Safety engineers score the degree of impact of various types of data (e.g., abnormal inclination may lead to collapse, with a score of 0.6).

[0098] Dynamic weight calculation: comprehensive weight W = w conf ×w risk Example: The confidence level of the tilt sensor is 0.9, and the risk factor is 0.6 → the final weight is 0.9 × 0.6 = 0.54.

[0099] The risk heat map generation unit is used to build a risk heat map generation model. Specifically, it builds a spatiotemporal convolutional neural network (ST-CNN) architecture. The spatiotemporal convolutional neural network (ST-CNN) architecture has the following structure:

[0100] Input layer: The input data includes time series data and spatial data. The time series data consists of sensor data (such as load, tilt angle, and wind speed) from the past 60 seconds, sampled at 10 Hz to form a 60×10 matrix. The spatial data consists of the 3D grid coordinates of the tower crane (dividing the boom into 100×100×100 voxels).

[0101] Network structure: The network structure includes a temporal convolution layer, a spatial convolution layer, and a feature fusion layer. The temporal convolution layer uses a one-dimensional convolution kernel (length 5) to extract temporal features (such as the vibration frequency trend). The spatial convolution layer uses a three-dimensional convolution kernel (3×3×3) to extract spatial correlation features (such as the pattern of stress transmission from the base of the boom to the end). The feature fusion layer combines temporal and spatial features and generates a risk probability value (0-1) through a fully connected layer.

[0102] Output layer: The output layer outputs a three-dimensional heat map, in which each voxel is marked as red (high risk), yellow (medium risk), and green (low risk).

[0103] Red: Stress > 80% of material yield limit or fatigue life < 7 days.

[0104] Yellow: Stress is between 60% and 80% or lifespan is between 7 and 30 days.

[0105] Green: stress <60% and lifetime >30 days.

[0106] The dynamic update mechanism subunit is used to force an update of the spatiotemporal convolutional neural network model's output when any sensor data changes suddenly. Specifically, this trigger condition is: forced update every 10 seconds or when any sensor data changes suddenly (e.g., load change rate > 5% / second). Its optimization strategy is: incremental learning: only recalculates the changed areas to reduce computational effort. GPU acceleration: utilizing CUDA parallel computing, a single update takes less than 1 second.

[0107] The fatigue life prediction unit is used to predict the service life of key components of tower cranes. Specifically, a fatigue life prediction model (LSTM) is constructed. The steps are as follows:

[0108] Data Preparation: Input features include historical load peak times, average operating hours, ambient temperature and humidity, and maintenance records. Label data includes the actual replacement time of key components (such as wire ropes).

[0109] Model training: The network structure is a two-layer LSTM (128 units) + a dropout layer (0.2) + a fully connected layer. The loss function is mean squared error (MSE), and the optimizer is Adam. Validation metric: The error between predicted and actual lifespan is <3 days (test set accuracy 92%).

[0110] Output: Remaining life report example: "Remaining life of wire rope: 5 days (confidence 85%), recommended to replace within 48 hours."

[0111] The following scenarios can be realized through the risk heat map generation unit:

[0112] Scenario 1: The ST-CNN detects that the stress in the middle of the boom has reached 75% (yellow warning). The LSTM model predicts that the remaining life of this part is 20 days. The system recommends: "Schedule ultrasonic flaw detection within this week."

[0113] Scenario 2: Risk simulation indicates that wind speeds will rise to 10 m / s in the next 15 minutes, potentially causing the hook to swing significantly. The system triggers the "auto hook lowering" command in advance to avoid a collision.

[0114] Through multi-source data fusion, ST-CNN heat map generation and predictive fault analysis, the present invention achieves comprehensive perception, dynamic assessment and proactive maintenance of the safety status of tower cranes, providing closed-loop management capabilities from data to decision-making for smart construction sites.

[0115] The display terminal receives data from the digital twin modeling module and the dynamic risk map generation module respectively, and is used to display the three-dimensional dynamic model of the tower crane, the hook operation trajectory, the risk heat map of each part of the tower crane and the remaining life report of each important component, which is convenient for staff to view.

[0116] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A tower crane monitoring system in a smart construction site project, characterized in that: include: Multimodal sensor module collects multi-dimensional data of tower crane operation in real time; The digital twin modeling module builds a three-dimensional dynamic model of the tower crane based on real-time multi-dimensional data, simulating the stress distribution and motion trajectory of the physical entity; The dynamic risk map generation module integrates real-time multi-dimensional data and the three-dimensional dynamic model of the tower crane to generate risk heat maps for each part of the tower crane and predict potential failure points; The display terminal shows the three-dimensional dynamic model of the tower crane, the risk heat map of each part of the tower crane, and potential fault points.

2. The tower crane monitoring system in a smart construction site project according to claim 1, characterized in that: The multimodal sensor module includes a mechanical sensor, a visual sensor, an environmental sensor and a voiceprint sensor; The mechanical sensors include load sensors, inclination sensors, flexible strain sensors and speed sensors, which are used to collect force changes and motion data of various components of the tower crane; The visual sensor includes a binocular camera, a 360° panoramic camera and an infrared thermal imager, which are used to collect image data of the tower crane and its surroundings; The environmental sensors include an ultrasonic anemometer, a temperature and humidity sensor, and a PM2.5 / dust sensor, which are used to collect environmental data of the tower crane and its surroundings; The soundprint sensor includes an acoustic microphone array and a vibration soundprint sensor, which is used to collect abnormal noise and vibration data of various components of the tower crane.

3. The tower crane monitoring system in a smart construction site project according to claim 1 is characterized in that: The digital twin modeling module includes a data preprocessing unit, a three-dimensional model building unit, a finite element analysis unit and a motion trajectory prediction unit; The data preprocessing unit is used to filter, remove noise and standardize the raw data collected by the sensor; The three-dimensional model building unit is used to build a three-dimensional model of the tower crane, establish a spatial coordinate system based on the three-dimensional model of the tower crane, and decompose the three-dimensional model of the tower crane; The finite element analysis unit is used to calculate the stress conditions of various components of the tower crane; The motion trajectory prediction unit is used to predict the motion trajectory of the tower crane hook, perform differential comparison with actual data, and detect abnormal deviation.

4. The tower crane monitoring system in a smart construction site project according to claim 3 is characterized by: The three-dimensional model building unit includes a geometric model generating subunit, a decomposing subunit and a coordinate system establishing subunit; The geometric model generation subunit constructs a three-dimensional model of the tower crane based on the tower crane drawings and sensor collected data; The decomposition subunit performs module decomposition on the three-dimensional tower crane model into multiple independent components, and assigns physical properties to each independent component; The coordinate system establishing subunit constructs a spatial coordinate system based on the three-dimensional model of the tower crane.

5. The tower crane monitoring system in a smart construction site project according to claim 4 is characterized in that: The finite element analysis unit includes a tower crane structure discretization subunit, a material property and boundary condition definition subunit, and a stress-strain calculation subunit; The tower crane structure discretization sub-unit is meshed according to each independent component of the tower crane and discretized into a finite number of small units; The material property and boundary condition definition subunit is used to assign material properties and constraint conditions to a finite number of small units; The stress-strain calculation subunit is used to solve the stress-strain relationship of each small unit and simulate the stress state of the overall structure of the tower crane.

6. The tower crane monitoring system in a smart construction site project according to claim 5 is characterized by: The motion trajectory prediction unit includes a global-local coordinate system construction subunit, a prediction model construction subunit and an abnormal offset detection subunit; The global-local coordinate system construction subunit is used to construct the global coordinate system of the tower crane and the local coordinate system of the hook; The prediction model building subunit is used to build a dynamic model and solve the dynamic model to obtain the predicted position of the hook; The abnormal deviation detection subunit is used to perform differential comparison with actual data to detect abnormal deviation.

7. The tower crane monitoring system in a smart construction site project according to claim 1, characterized in that: The dynamic risk map generation module includes a multi-source data fusion and feature extraction unit, a risk heat map generation unit and a fatigue life prediction unit; The multi-source data fusion and feature extraction unit is used to align the sensor collected data in time and space, extract correlation features, establish physical and logical relationships between data, and dynamically adjust weights according to data quality and risk impact; The risk heat map generation unit is used to generate a three-dimensional heat map of the tower crane; The fatigue life prediction unit is used to predict the service life of key components of tower cranes.

8. The tower crane monitoring system in a smart construction site project according to claim 7, characterized in that: The multi-source data fusion and feature extraction unit includes a time synchronization subunit, a space alignment subunit and a feature correlation analysis subunit; The time synchronization subunit is used to add a unified time stamp to the data collected by all sensors; The spatial alignment subunit is used to map the sensor collected data into the three-dimensional coordinate system of the tower crane; The feature association analysis subunit is used to mine cross-modal features through correlation analysis (such as Pearson correlation coefficient).

9. The tower crane monitoring system in a smart construction site project according to claim 8, characterized in that: The risk heat map generation unit includes a network model construction subunit and a dynamic update mechanism subunit; The network model construction subunit is used for a spatiotemporal convolutional neural network model, and the spatiotemporal convolutional neural network model is used to output a three-dimensional heat map; The dynamic update mechanism subunit is used to force the update of the output of the spatiotemporal convolutional neural network model when any sensor data suddenly changes.

10. The tower crane monitoring system in a smart construction site project according to claim 9, characterized in that: The fatigue life prediction unit includes a fatigue life prediction model construction subunit, which is used to construct a fatigue life prediction model and output a remaining life report using the fatigue life prediction model.

Images