Material intelligent transportation and safety monitoring system and method for shield construction

Abstract

The invention relates to the technical field of tunnel engineering construction, and discloses an intelligent material transportation and safety monitoring system and method for shield construction, and the system comprises a visual perception unit, a sensor network module, an AI analysis center module, a safety decision module and a human-computer interaction interface. According to the invention, data acquisition is carried out through the visual perception unit and the sensor network module, multi-target detection operation is carried out on image frames through the AI analysis center module after target identification and track prediction, target types, space coordinates, contour boundaries and confidence coefficients are identified and extracted, and safety judgment and early warning output are carried out. According to the invention, by integrating the multi-view camera equipment and the UWB, GNSS and other sensors and adopting a deep learning target detection algorithm, high-precision identification and continuous tracking can be carried out on construction site personnel, equipment, segments and other key objects, and accurate input is provided for subsequent risk analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel engineering construction, and in particular to a material intelligent transportation and safety monitoring system and method for shield construction. Background Art

[0002] During shield tunnel construction, the transportation and hoisting of large prefabricated components, such as segments and box culverts, are critical. Due to limited construction site space and a complex operating environment, material transportation often involves the coordinated efforts of gantry cranes, transport vehicles, and operators. This is characterized by high frequency, numerous intersections, and significant safety risks.

[0003] Existing material transportation safety management methods primarily rely on manual command and on-site monitoring. Although video surveillance and sensor equipment are deployed in some projects, these systems typically only have basic functions such as image acquisition and alarm triggering. Given the dynamic environment of construction sites, existing systems have the following technical limitations:

[0004] 1. The existing video system lacks the ability to intelligently identify multiple targets, including gantry cranes, segments, vehicles, and personnel, and is unable to automatically track and determine their behavior.

[0005] 2. Existing sensor systems are mostly focused on data collection and lack a linkage analysis mechanism, making it difficult to provide real-time warnings for high-risk events such as lifting path conflicts and misentering the work area.

[0006] 3. There is a lack of systematic safety inspection and intelligent confirmation process before the lifting operation, and there is a lack of automatic intervention during the lifting process;

[0007] 4. After the lifting operation is completed, the recording and analysis of related risk events rely on manual organization, lacking a structured processing mechanism, making it difficult to support subsequent safety assessments and construction optimization. Summary of the Invention

[0008] In response to the shortcomings of the existing technology, the present invention provides a material intelligent transportation and safety monitoring system and method for shield construction. The monitoring system and method can perform high-precision identification and continuous tracking of key objects such as construction site personnel, equipment, and pipe segments, providing accurate input for subsequent risk analysis.

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent material transportation and safety monitoring system for shield construction, including a visual perception unit, a sensor network module, an AI analysis center module, a safety decision module and a human-computer interaction interface;

[0010] The visual perception unit, as the core front-end perception module of the system, is deployed in the operating area of ​​the shield construction site. It is used to capture images and output video streams of dynamic targets in the operating area of ​​the shield construction site, providing raw data support for subsequent target recognition and path prediction. The operating area of ​​the shield construction site includes the hoisting operation area, the yard boundary, the component loading and unloading point, and the intersection of the transportation channel.

[0011] The sensor network module, as a component for multi-source dynamic data collection and linkage analysis, is deployed on the gantry crane structure, work area, transport vehicles, and personnel's personal equipment at the construction site. It is used to obtain real-time equipment operating status, target location information, and lifting safety parameters, ensuring the system's full-time and space-time monitoring capabilities of the construction process.

[0012] The AI ​​analysis center module is the core module for data calculation and intelligent judgment of the system, which is used to analyze, fuse and reason multi-source heterogeneous data from the visual perception unit and sensor network module to complete target detection tasks, trajectory prediction tasks, safety judgment tasks and risk output tasks;

[0013] The safety decision module is a logical execution unit that implements automated lifting safety risk identification and response control. Based on the target recognition results, trajectory prediction information, and state perception data provided by the AI ​​analysis center module, combined with the built-in rule library, it dynamically determines high-risk behaviors that may occur at the construction site and triggers corresponding control instructions or early warning mechanisms.

[0014] The human-computer interaction interface is an integral part of realizing information visualization, operation instruction interaction and data query management, and provides a unified, intuitive and real-time control and management channel for construction operators, safety management personnel and back-end dispatch personnel.

[0015] Further technical solutions of the present invention: The visual perception unit is composed of a camera device, which is a panoramic camera and a local high-definition camera. The camera device is connected to the edge server through an Ethernet interface or a wireless network, and the image data is pushed to the AI ​​analysis center module in real time in RTSP or H.265 encoding format. The image frame rate of the visual perception unit is not less than 25fps, and the transmission delay is controlled within 200ms.

[0016] A further technical solution of the present invention is as follows: the dynamic targets in the working area of ​​the shield construction site include personnel, gantry cranes, pipe segments, box culverts, and vehicles; after the image acquisition is completed, it is input into the target detection model for real-time recognition, and the recognition process is as follows:

[0017] Target recognition process formula:

[0018] Given an input image frame First, after the image preprocessing module, it is sent to the deep convolutional neural network to obtain the target candidate box set:

[0019]

[0020] in:

[0021] (x i ,y i ) represents the center coordinate of the i-th candidate box;

[0022] w i ,h i is the width and height of the candidate box;

[0023] Indicates category;

[0024] p i ∈[0,1] is the confidence of the target;

[0025] Apply the non-maximum suppression algorithm to the detection results to remove duplicate frames and generate the final recognition target set:

[0026]

[0027] Where θ is the IoU threshold, which is usually set to 0.5.

[0028] Furthermore, the sensor network module is composed of the following subsystems:

[0029] Gantry crane status monitoring subsystem: Multiple sets of industrial-grade sensors installed on the gantry crane frame and control cabinet are used to monitor the operating status of the lifting equipment, including:

[0030] Displacement sensor: obtains the running distance and direction of the crane along the track;

[0031] Load sensor: installed at the hook, lifting point, and rope to monitor the quality of the currently hoisted object in real time;

[0032] Speed ​​sensor: used to measure the speed of the hook lifting and the vehicle body movement, with a typical sampling frequency of 10Hz;

[0033] Tilt angle sensor or gyroscope: determines the shaking amplitude or posture change of the hoisting component;

[0034] Personnel Positioning and Identification Subsystem: To ensure real-time awareness of workers in the lifting area, the system requires construction workers to be equipped with UWB positioning tags or RFID / Bluetooth low-power devices, which, combined with on-site base stations, achieve two-dimensional positioning capabilities.

[0035] The positioning system is based on the TOF ranging principle and combines the TDOA algorithm to solve the position coordinates. Its basic positioning formula is:

[0036]

[0037] in:

[0038] Δt ij is the time difference between the i-th base station and the j-th base station receiving the signal;

[0039] d i , d j is the distance from the target to base station i, j;

[0040] c is the wireless signal propagation speed;

[0041] Combining the positioning geometry array composed of multiple base stations, the target position (x, y) can be obtained by least squares inversion;

[0042] Transport vehicle tracking subsystem: GNSS positioning modules and IMUs are installed on transport vehicles to monitor trajectory, speed, and acceleration information.

[0043] Multi-source data fusion and synchronization: The data collected by the sensor network module is standardized and encapsulated using a unified timestamp and coding identifier, and transmitted to the AI ​​analysis center module via the MQTT or Modbus-TCP protocol. In conjunction with the clock synchronization module, the NTP protocol is used to align the time of all terminal devices to a millisecond-level error range to ensure the timing consistency of multi-channel data.

[0044] A further technical solution of the present invention is as follows: the AI ​​analysis center module is deployed in an on-site edge server or connected to the engineering data platform through a cloud-edge collaborative architecture. Its core functions include: multi-target recognition and classification, multi-source data fusion, trajectory prediction and conflict analysis, risk rule reasoning and alarm triggering;

[0045] The multi-target recognition and classification is used to process image data and identify target objects on the scene, input image frame After feature extraction network f θ Extract the feature map F and generate the target set through the candidate box predictor Use the following formula:

[0046]

[0047] in:

[0048] b i =(x i ,y i ,w i ,h i ) represents the bounding box of the i-th target;

[0049] c iIndicates category (such as construction workers, gantry cranes, box culverts, etc.);

[0050] s i Indicates confidence;

[0051] The image targets identified by the multi-source data fusion mechanism are spatially mapped and temporally aligned with the physical coordinate data acquired by the sensor, and the Hungarian matching algorithm and Kalman filter are used for target association and tracking:

[0052] Kalman filter state model:

[0053] Let the target state be The observed value is The state update and observation update process is:

[0054] x t|t-1 =Ax t-1|t-1 ,z t =Hx t|t-1 +w t

[0055] Where: A is the state transfer matrix, H is the observation matrix,

[0056] Hungarian algorithm matching matrix calculation formula:

[0057] Constructing the cost matrix Among them C ij =d(p i ,q j ) represents the image target p i With sensor target q j Euclidean distance or IoU cost, solving the minimum total matching cost;

[0058]

[0059] The trajectory prediction and conflict analysis utilizes the AI ​​analysis center module to predict the execution trajectory of construction personnel, lifting equipment, and transport vehicles, using a multi-step linear regression model or an LSTM-based time series prediction model;

[0060] The prediction formula is as follows:

[0061]

[0062] Where: x t is the current position, v t is the current speed, k is the predicted step size;

[0063] Predicting trajectories for multiple targets Execute conflict judgment:

[0064]

[0065] Where δ is the conflict distance threshold;

[0066] The safety rule reasoning and warning output utilizes the AI ​​analysis center module to build a logic judgment rule library, and outputs risk level judgment based on target type, spatial position, speed status and prediction results.

[0067] Furthermore, the security decision module includes four parts: rule knowledge base, real-time judgment engine, response control unit, risk classification and execution interface;

[0068] The rule knowledge base is the basic logical unit of the safety decision-making module. It has a built-in set of judgment conditions and corresponding response strategies for common risk scenarios at shield construction sites. The rule form supports hard logic judgment and fuzzy matching logic.

[0069] The real-time judgment engine is responsible for matching the target detection results sent by the AI ​​analysis center module with the rule library. It uses an event trigger mechanism to poll the input cache every 100ms, matching all identified targets and their predicted trajectories with the rule conditions one by one, and generating a risk event queue.

[0070] The judgment process is as follows:

[0071] Receive the current frame target state set S t ={s1, s2, ..., s n}

[0072] For each target s i , call its status parameters;

[0073] Match all rules R j , when R is satisfied j When the trigger condition is met, add it to the risk event queue E t ;

[0074] Output event E t The highest level of risk among all the risks and transfer them to the response control unit;

[0075] The response control unit executes a hierarchical response strategy based on the risk level;

[0076] The risk event recording and backtracking mechanism automatically generates the following structured log entries after a risk event is detected. The logs are uploaded to the security management platform for responsibility backtracking, risk statistics, management assessment, and subsequent optimization.

[0077] The system stability and fault tolerance mechanism has the following fault tolerance strategies:

[0078] When the data delay exceeds 500ms, the system suspends the action response and prompts "data synchronization abnormality";

[0079] When the key target cannot be identified for three consecutive frames, the fault protection mechanism is triggered;

[0080] A manual review entry is set up to deal with the risks of false alarms or missed alarms, and corrections and confirmations are made through a secure terminal.

[0081] The present invention also provides a method for intelligent material transportation and safety monitoring for shield construction, which uses the above-mentioned intelligent material transportation and safety monitoring system for shield construction for monitoring, and specifically includes the following steps:

[0082] Step 1: Data collection, the data collection task is initiated by multiple types of perception devices deployed at the construction site, including visual perception units and sensor network modules;

[0083] Step 2: Target recognition and trajectory prediction. After the data is uploaded to the AI ​​analysis center module, the system calls the deployed target recognition model to perform multi-target detection operations on the image frame, identifying and extracting the target type, spatial coordinates, contour boundaries, and confidence level.

[0084] Step 3: Safety judgment and warning output: The system integrates the target recognition results and trajectory prediction results, calls the rule knowledge base built into the safety decision module, and matches each rule to see if there are potential risk events.

[0085] Step 4: Intervention and recording of hoisting operations. When the identification results trigger a medium- or high-level risk event, the system immediately sends a control signal through the safety decision module, interrupts the current action of the hoisting equipment through the API or PLC interface, and locks the control authority.

[0086] Step 5: Report generation and data archiving. After the lifting operation is completed, the system will structure the images, sensor data, target identification records, risk events and operation response records collected throughout the entire process to generate a lifting task report and a safety risk report.

[0087] A further technical solution of the present invention: the data collection in step 1 includes collecting dynamic target images and videos of the working area of ​​the shield construction site through the visual perception unit, and collecting the operating status, target position information and lifting safety parameters of the dynamic targets in the working area of ​​the shield construction site through the sensor network module.

[0088] A further technical solution of the present invention is as follows: in step 2, when performing the trajectory prediction operation on the historical position data of consecutive frames, a multi-step linear regression model or an LSTM neural network is used to infer the spatial path within several future time steps. The prediction expression is as follows:

[0089]

[0090] Where: x t is the current target position; v t is the estimated speed; k is the prediction step size (e.g. 1 to 5 seconds); For future location;

[0091] The prediction results are superimposed on the monitoring interface in the form of trajectory lines and passed as input to the safety judgment module.

[0092] The present invention has the following beneficial effects:

[0093] (1) The present invention integrates multi-view camera equipment with UWB, GNSS and other sensors, and adopts deep learning target detection algorithm to perform high-precision identification and continuous tracking of key objects such as construction site personnel, equipment, and pipe segments, providing accurate input for subsequent risk analysis.

[0094] (2) Before starting the lifting operation, the present invention automatically performs image recognition of the operating area and fusion judgment of positioning data to confirm whether there are people stranded or obstacles in the area, and authorizes the start only after all safety conditions are met, so as to avoid accidents caused by blind operation.

[0095] (3) The present invention performs multi-level early warning judgment on risk events such as path conflict, personnel misentry, hoisting overload, abnormal speed, etc. that occur during the lifting process, and responds by issuing sound and light alarms, platform pop-up prompts, or automatically suspending operations, effectively improving the efficiency of risk response.

[0096] (4) The present invention automatically records the status data, warning events, response actions and processing feedback during the entire lifting task, and generates structured logs and standardized report files, which are convenient for the construction unit to conduct audit filing, safety assessment and subsequent optimization.

[0097] (5) The human-computer interaction interface of the present invention supports a variety of device display forms such as large-screen terminals and mobile tablets, and combines the hierarchical authority control of operators, safety managers and platform administrators to achieve visual and standardized management of work operations and safety control.

[0098] (6) The present invention adopts a modular architecture design, supports docking with shield construction digital platforms and smart construction site systems, and is compatible with commonly used industrial interface protocols, facilitating promotion and application in different project scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Figure 1 This is a system structure diagram of the present invention;

[0100] Figure 2 This is a closed-loop flow chart of the hoisting operation of the present invention;

[0101] Figure 3 This is a closed-loop logic diagram of the entire lifting process data of the present invention;

[0102] Figure 4 This is the deployment structure diagram of the visual perception unit of the present invention

[0103] Figure 5 This is the data flow diagram of the sensor network module of the present invention

[0104] Figure 6 This is the flow chart of the safety decision module of the present invention

[0105] Figure 7 Schematic diagram of the human-computer interaction interface of the present invention

[0106] Figure 8 Schematic diagram of the risk event recording structure of the present invention

[0107] Figure 9 Schematic diagram of the construction environment perception of the present invention

[0108] Figure 10 This is a display diagram of the target recognition results of the present invention

[0109] Figure 11 Schematic diagram of path prediction and collision judgment of the present invention

[0110] Figure 12 This is the hoisting trajectory prediction and path fitting diagram of the present invention

[0111] Figure 13 This is a gridded risk heat map of the operating area of ​​the present invention. DETAILED DESCRIPTION

[0112] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention.

[0113] See also Figures 1 to 13 The present invention provides a material intelligent transportation and safety monitoring system for shield construction, including a visual perception unit, a sensor network module, an AI analysis center module, a safety decision module and a human-computer interaction interface.

[0114] Visual perception unit: As the core front-end perception module of the system of the present invention, the visual perception unit is deployed in the operating area of ​​the shield construction site, including the lifting operation area, the yard boundary, the component loading and unloading point, and the intersection of the transportation channel; this unit is mainly responsible for image acquisition and video stream output of dynamic targets at the construction site (such as personnel, gantry cranes, pipe segments, box culverts, vehicles, etc.), providing original data support for subsequent target recognition and path prediction.

[0115] The visual perception unit consists of fixed-mounted cameras, including a panoramic camera and a localized high-definition camera. The panoramic camera, typically located at a commanding height on the site with a viewing angle of at least 120°, captures global image information. The localized high-definition cameras, with a resolution of at least 2 million pixels and infrared night vision support, are primarily deployed along the lifting path and in high-frequency workstation areas to capture detailed images. In high-dust environments, the system uses image sharpening and edge enhancement algorithms to enhance recognition robustness. Furthermore, the system employs a multi-view fusion mechanism to complete and correct critical areas where the views of multiple cameras overlap. Image timestamp synchronization is controlled by the NTP protocol, ensuring time alignment between different data sources to meet the timing consistency requirements of downstream trajectory fusion analysis. The cameras connect to the edge server via Ethernet or wireless networks. Image data is pushed to the AI ​​analysis center module in real time using RTSP or H.265 encoding. The visual perception unit maintains an image frame rate of at least 25 fps, with transmission latency controlled within 200ms.

[0116] The visual perception unit is responsible for capturing images of dynamic targets at the construction site and outputting video streams, providing raw data support for subsequent target recognition and path prediction. After image capture is completed, it is input into the target detection model for real-time recognition. The recognition process is as follows:

[0117] Target recognition process formula:

[0118] Given an input image frame First, after the image preprocessing module, it is sent to the deep convolutional neural network to obtain the target candidate box set:

[0119]

[0120] in:

[0121] (x i ,y i ) represents the center coordinate of the i-th candidate box;

[0122] w i , h i is the width and height of the candidate box;

[0123] Indicates category;

[0124] p i ∈[0, 1] is the confidence of the target.

[0125] Apply the non-maximum suppression algorithm to the detection results to remove duplicate frames and generate the final recognition target set:

[0126]

[0127] Where θ is the IoU threshold, which is usually set to 0.5.

[0128] Table 1. Object categories recognized by visual perception units

[0129]

[0130] Through the collaboration of the above-mentioned visual perception structure and recognition algorithm, the system can achieve continuous monitoring and classification identification of multiple types of targets at the shield construction site, providing high-quality perception input for subsequent path prediction, behavior judgment and safety warning.

[0131] Sensor network module: The sensor network module is an integral component for multi-source dynamic data collection and linkage analysis. It is deployed on the gantry crane structure, operating area, transport vehicles, and personnel's personal equipment at the construction site. It is used to obtain real-time equipment operating status, target location information, and lifting safety parameters, ensuring the system's full-time and space-time monitoring capabilities of the construction process.

[0132] The sensor network module consists of the following subsystems:

[0133] Gantry crane status monitoring subsystem: Multiple sets of industrial-grade sensors installed on the gantry crane frame and control cabinet are used to monitor the operating status of the lifting equipment, including:

[0134] Displacement sensor: obtains the running distance and direction of the crane along the track;

[0135] Load sensor: installed at the hook, lifting point, and rope to monitor the quality of the currently hoisted object in real time;

[0136] Speed ​​sensor: used to measure the speed of the hook lifting and the vehicle body movement, with a typical sampling frequency of 10Hz;

[0137] Tilt angle sensor or gyroscope: determines the shaking amplitude or posture change of the hoisting component;

[0138] Table 2 shows the data format collected by the gantry crane condition monitoring subsystem:

[0139]

[0140] The sensor data is connected to the local communication node via RS485 or CAN bus, and then structured and uploaded synchronously through the edge server.

[0141] Personnel Positioning and Identification Subsystem: To ensure real-time awareness of workers in the lifting area, the system requires construction workers to be equipped with UWB positioning tags or RFID / Bluetooth low-power devices. Combined with on-site base stations, this system achieves two-dimensional positioning with an accuracy better than 30cm.

[0142] The positioning system is based on the TOF ranging principle and combines the TDOA algorithm to solve the position coordinates. Its basic positioning formula is:

[0143]

[0144] in:

[0145] Δt ij is the time difference between the i-th base station and the j-th base station receiving the signal;

[0146] d i , d j is the distance from the target to base station i, j;

[0147] c is the wireless signal propagation speed;

[0148] Combining the positioning geometric array composed of multiple base stations, the target position (x, y) can be obtained by least squares inversion;

[0149] Table 3 shows the format for uploading positioning data:

[0150] Personnel ID Coordinate X (m) Coordinate Y (m) state Regional identification P0123 12.65 3.87 normal Lifting danger zone A

[0151] The system sets geographic fences and operating boundaries. When it identifies a person's trajectory crossing the boundary of a high-risk area, the linked AI module triggers an early warning.

[0152] Transport vehicle tracking subsystem: GNSS positioning modules and IMUs are installed on transport vehicles to monitor trajectory, speed, and acceleration information.

[0153] Table 4 shows the data format of the transport vehicle tracking subsystem.

[0154]

[0155] Vehicle path data is matched with the digital map of the construction road to determine potential interference factors such as path congestion, deviation, and intersection.

[0156] Multi-source data fusion and synchronization: The data collected by the sensor network module is standardized and encapsulated using a unified timestamp and coding identifier, and transmitted to the AI ​​analysis center module via the MQTT or Modbus-TCP protocol. In conjunction with the clock synchronization module, the NTP protocol is used to align the time of all terminal devices to a millisecond-level error range to ensure the timing consistency of multi-channel data.

[0157] The data is formatted and processed in the AI ​​center. The AI ​​module performs target registration, conflict identification and path prediction based on the above data, and then generates the lifting risk judgment results and drives the safety decision-making module.

[0158] As the basis of the multimodal monitoring capability of the present invention, this module can effectively support real-time and accurate lifting safety identification and dynamic early warning tasks in complex construction scenarios, providing comprehensive and reliable data protection for construction management.

[0159] AI Analysis Center Module: The AI ​​Analysis Center module is the data computing and intelligent judgment core of the system. It is responsible for parsing, fusing, and reasoning multi-source heterogeneous data from the visual perception unit and sensor network module to complete target detection, trajectory prediction, safety assessment, and risk output tasks.

[0160] The AI ​​analysis center module is deployed in the on-site edge server or connected to the engineering data platform through the cloud-edge collaborative architecture. Its core functions include: multi-target recognition and classification, multi-source data fusion, trajectory prediction and conflict analysis, risk rule reasoning and alarm triggering;

[0161] Multi-target recognition and classification is used to process image data and identify target objects on the scene. The input image frame After feature extraction network f θ Extract the feature map F and generate the target set through the candidate box predictor Use the following formula:

[0162]

[0163] in:

[0164] b i =(x i ,y i , w i , h i ) represents the bounding box of the i-th target;

[0165] c i Indicates category (such as construction workers, gantry cranes, box culverts, etc.);

[0166] s i Indicates confidence.

[0167] After applying non-maximum suppression to the identified target set to eliminate redundancy, the target list is output for subsequent tracking and fusion.

[0168] The image targets identified by the multi-source data fusion mechanism are spatially mapped and temporally aligned with the physical coordinate data acquired by the sensor. The Hungarian matching algorithm and Kalman filter are used for target association and tracking:

[0169] Kalman filter state model:

[0170] Let the target state be The observed value is The state update and observation update process is:

[0171] x t|t-1 =Ax t-1|t-1 ,z t =Hx t|t-1 +w t

[0172] Where: A is the state transfer matrix, H is the observation matrix,

[0173] Hungarian algorithm matching matrix calculation formula:

[0174] Constructing the cost matrix Among them C ij =d(p i ,q j ) represents the image target p i With sensor target q j Euclidean distance or IoU cost, solving the minimum total matching cost;

[0175]

[0176] This fusion mechanism is used to achieve unified identification and stable tracking of image targets and physical objects.

[0177] Trajectory prediction and conflict analysis utilizes the AI ​​analysis center module to predict the execution trajectories of construction personnel, lifting equipment, and transport vehicles, using a multi-step linear regression model or an LSTM-based time series prediction model.

[0178] The prediction formula is as follows:

[0179]

[0180] Where: x t is the current position, v t is the current speed, and k is the prediction step size.

[0181] Predicting trajectories for multiple targets Execute conflict judgment:

[0182]

[0183] Where δ is the conflict distance threshold.

[0184] Safety rule reasoning and warning output use the AI ​​analysis center module to build a logical judgment rule library, combining target type, spatial position, speed status and prediction results to output risk level judgment.

[0185] Table 5 is a schematic diagram of typical security risk rules

[0186]

[0187] When a rule is triggered, the system executes a processing strategy based on the level of grading, including sound and light alarms, graphical interface prompts, hoisting control interruption, background data recording and other operations. The output results are notified to the operator and the management system through the human-computer interaction interface.

[0188] Through the combination of the above structure and algorithm, this module can realize real-time recognition of multiple targets, behavior prediction and operation risk determination, and is the key foundation for the system's intelligent decision-making and early warning response.

[0189] Safety Decision-Making Module: This module is a logical execution unit that implements automated lifting safety risk identification and response control. Based on the target recognition results, trajectory prediction information, and state perception data provided by the AI ​​Analysis Center module, combined with a built-in rule library, it dynamically determines high-risk behaviors that may occur at the construction site and triggers corresponding control instructions or early warning mechanisms.

[0190] The security decision module consists of four parts: rule knowledge base, real-time judgment engine, response control unit, risk classification and execution interface.

[0191] The rule knowledge base is the basic logical unit of the safety decision-making module. It has a built-in set of judgment conditions and corresponding response strategies for common risk scenarios at shield construction sites. The rule form supports hard logic judgment and fuzzy matching logic.

[0192] A rule item consists of the following elements:

[0193] Trigger conditions: such as target type, path prediction, speed change, load value, position relationship, etc.;

[0194] Constraint parameters: such as time threshold, spatial intersection distance, speed limit, area classification, etc.;

[0195] Response level: divided into five levels: "low risk, general risk, medium risk, high risk, and serious risk";

[0196] Execution actions: including but not limited to sound and light warnings, information pop-ups, job suspension, platform push, data recording, etc.

[0197] Each rule is modeled using the following logical structure

[0198] IF condition 1 ∧ condition 2 ∧ ... condition n THEN execute the action

[0199] For example:

[0200] Rule R013: If the current load is greater than the rated upper limit × 1.1 and the lifting height is greater than 2.5m, a high-level risk warning will be output and the lifting action will be locked.

[0201] The real-time judgment engine is responsible for matching the target detection results sent by the AI ​​analysis center module with the rule base. It uses an event trigger mechanism to poll the input buffer every 100ms, matching all identified targets and their predicted trajectories with the rule conditions one by one, and generating a risk event queue.

[0202] The judgment process is as follows:

[0203] Receive the current frame target state set S t ={s1, s2, ..., s n}

[0204] For each target s i , call its status parameters;

[0205] Match all rules R j , when R is satisfied j When the trigger condition is met, add it to the risk event queue E t ;

[0206] Output event E t The highest level of risk among all the risks and transfer them to the response control unit;

[0207] The response control unit implements a hierarchical response strategy based on the risk level;

[0208] Table 6 shows the graded response strategy

[0209]

[0210] All control instructions are issued through API or PLC interface and linked with the lifting control system to ensure real-time and linkage of risk management.

[0211] After a risk event is detected, the risk event recording and backtracking mechanism automatically generates the following structured log entries, which are then uploaded to the security management platform for responsibility backtracking, risk statistics, management assessment, and subsequent optimization.

[0212] Table 7 shows the log entries.

[0213] Field content Event Number AUTO20250331-001 Timestamp 2025-03-31 08:45:12 Risk Level high Trigger Object Gantry crane V2, personnel P034 Cause Description Path intersection risk + abnormal lifting speed Response Action Suspend hoisting, red alert Confirm personnel Work area manager Processing Status Processing / Resolved

[0214] The system stability and fault tolerance mechanism has the following fault tolerance strategies:

[0215] When the data delay exceeds 500ms, the system suspends the action response and prompts "data synchronization abnormality";

[0216] When the key target cannot be identified for three consecutive frames, the fault protection mechanism is triggered;

[0217] A manual review entry is set up to deal with the risks of false alarms or missed alarms, and corrections and confirmations are made through a secure terminal.

[0218] Through the above design, the safety decision-making module can realize comprehensive judgment and intelligent response to multi-source dynamic information at the construction site, forming a closed-loop process of "identification-judgment-warning-response-recording", and providing real-time, accurate and standardized lifting safety control means for the shield construction site.

[0219] Human-computer interaction interface: The human-computer interaction interface is a component of the system of the present invention that realizes information visualization, operation instruction interaction and data query management. It aims to provide construction operators, safety management personnel and back-end dispatch personnel with a unified, intuitive and real-time control and management channel.

[0220] This module presents a graphical user interface (GUI) through local workstations, embedded terminals, mobile devices (such as tablets) and other carriers, connecting visual perception, AI analysis, security decision-making and data recording modules to achieve multi-level, multi-role and multi-dimensional information interaction.

[0221] System architecture and functional partitioning

[0222] The interactive interface adopts a browser / client-based B / S architecture design. The front-end uses HTML5 and WebGL technology to achieve dynamic visual rendering, and the back-end service connects the system database and AI inference results through RESTful API.

[0223] The interface functions are divided as follows:

[0224]

[0225]

[0226] Real-time monitoring and information display

[0227] The homepage of the interface presents a panoramic view of the hoisting area and dynamically overlays the system's recognition results. Each detected target is framed with a colored border and labeled with unique number, category, speed value, hazard level and other information. For example:

[0228] Green box: ordinary workers

[0229] Yellow box: High-risk behavior prompt (such as people approaching hoisted objects)

[0230] Red box: A serious risk target has been triggered (such as a person appearing under the hoisted object)

[0231] Click on any target to pop up a detailed attribute information card, including:

[0232] Field content Target Type construction workers serial number P0123 Current coordinates (12.64,3.87) state Stationary, 1.2 meters from the lifting path Is it a risk target? Yes (path conflict prediction) Recommended measures Guided evacuation

[0233] At the same time, the current event list is displayed in real time on the right side of the interface, automatically sorted by risk level, and accompanied by voice prompts.

[0234] Operation control and safety linkage

[0235] The interface console has the following interactive capabilities:

[0236] Before starting the operation: the system performs an area security scan, and a "security confirmation box" pops up on the interface, listing all identified targets and their status in the current operation area;

[0237] During operation: When a high-level risk is detected, the system will pop up an alarm and display handling suggestions. The operator can manually confirm "Continue / Pause / Call the on-site person in charge";

[0238] Risk handling: If the risk has been handled, the operator can submit a "risk cleared" confirmation and the system will continue to operate;

[0239] Emergency: Press the "Emergency Pause" key to immediately stop the lifting command and broadcast a high-level alarm.

[0240] The module supports linkage with lifting equipment through a programmable logic controller (PLC) or API interface, ensuring that control instructions can be quickly transmitted to the execution layer.

[0241] Historical data query and export

[0242] The system automatically records log data of all operation processes, including:

[0243] Lifting task number, time period, equipment number;

[0244] Detection target statistics and recognition accuracy;

[0245] Details of all risk events and warnings;

[0246] Operator identity and response actions;

[0247] System status and external command records.

[0248] Users can query historical records by conditional filtering (time, object, risk level) and export them to PDF, Excel and other formats to facilitate risk statistics, security assessment or management assessment.

[0249] The example record format is as follows:

[0250] Task Number time Risk Objects Risk Level Response Action Confirm personnel DZG001 2025-03-31 08:47 Personnel P034 high Suspend lifting Engineer Zhang DZG002 2025-03-31 09:10 Vehicle V007 middle Prompt to slow down System automatic

[0251] User permissions and interface adaptation

[0252] The system supports hierarchical role configuration, including:

[0253] Ordinary operators: can receive early warnings, perform manual confirmation and review;

[0254] Security managers: have the authority to review logs, modify rules, and export data;

[0255] Platform administrator: can manage user permissions, system parameters and device access configuration.

[0256] The interface supports multi-terminal adaptive layout and is compatible with a variety of commonly used equipment on construction sites, including large-screen displays, hard hat terminals, mobile tablets, etc.

[0257] Through the human-computer interaction interface of the present invention, the system realizes the visualization presentation of multi-dimensional data, the efficient transmission of risk information and the whole process control of safe operation, thereby improving the intelligent management level of the shield construction site.

[0258] The system workflow includes the following steps:

[0259] Step 1: Data collection,The system starts running, and the data collection task is initiated by various types of perception devices deployed at the construction site, including visual perception units and sensor network modules.

[0260] The panoramic camera and local high-definition camera in the visual perception unit capture overall and local image data of the work area, respectively. Images are transmitted to the AI ​​analysis center module at a frame rate of 25 frames per second using H.265 compression, ensuring real-time image acquisition and clarity.

[0261] At the same time, the sensor network module synchronously collects the following data:

[0262] Gantry crane motion status (position, speed, load, tilt angle, etc.);

[0263] UWB / RFID positioning information of construction workers;

[0264] GNSS trajectory and IMU dynamic parameters of transport vehicles;

[0265] Set geo-fences and path node status in the work area.

[0266] All collected data is timestamped and uploaded to the edge server for fusion processing through a unified data interface (such as MQTT, Modbus-TCP).

[0267] Step 2: Target Identification and Trajectory Prediction. After data is uploaded to the AI ​​Analysis Center module, the system calls the deployed target recognition model to perform multi-target detection on the image frames, identifying and extracting target type, spatial coordinates, contour boundaries, and confidence levels. Typical targets include construction workers, gantry cranes, pipe segments, box culverts, construction vehicles, and safety signs. After identification, the system combines Kalman filtering with the Hungarian matching algorithm to consistently associate targets across consecutive frames, achieving stable tracking and ID assignment.

[0268] When performing trajectory prediction operations on historical position data of consecutive frames, a multi-step linear regression model or LSTM neural network is used to infer the spatial path within several future time steps. The prediction expression is as follows:

[0269]

[0270] Where: x t is the current target position; v t is the estimated speed; k is the prediction step size (e.g. 1 to 5 seconds); For future location;

[0271] The prediction results are superimposed on the monitoring interface in the form of trajectory lines and passed as input to the safety judgment module.

[0272] Step 3: Safety judgment and warning output: The system integrates the target recognition results and trajectory prediction results, calls the rule knowledge base built into the safety decision module, and matches each rule to see if there are potential risk events.

[0273] Judgment criteria include but are not limited to:

[0274] The personnel's predicted trajectory intersects with the lifting path;

[0275] The load of the lifting equipment exceeds the rated limit;

[0276] The gantry crane's operating speed is abnormal;

[0277] Unauthorized targets appear in the operation area;

[0278] The sway angle of the hoisting posture exceeds the set threshold;

[0279] The personnel's stay in the work area has exceeded the limit.

[0280] The rule judgment engine marks events that meet the trigger conditions as pending response events, assigns corresponding risk levels (low, medium, high, severe), and transmits event information to the response module and the human-computer interaction interface.

[0281] The warning information is presented in the system interface in the form of pop-up windows, flashing prompts, sound and light alarms, etc., and records information such as the warning content, triggering object, judgment basis, and recommended response measures.

[0282] Step 4: Intervention and recording of hoisting operations. When the identification results trigger a medium- or high-level risk event, the system immediately sends a control signal through the safety decision module, interrupts the current action of the hoisting equipment through the API or PLC interface, and locks the control authority.

[0283] The system displays the following information on the interface:

[0284] Risk object number and type;

[0285] Risk level and warning level;

[0286] Speculate the time and location of occurrence;

[0287] Suggestions for handling (such as guiding personnel to evacuate and waiting for obstacle clearance);

[0288] Operation confirmation options (such as "Processed" and "Report to the person in charge").

[0289] At the same time, the system generates a structured lifting event log, the fields of which are as follows:

[0290]

[0291] The event is automatically written to the database and bound to the day's job tasks for subsequent query and audit use.

[0292] Step 5: Report generation and data archiving. After the lifting operation is completed, the system will structure the images, sensor data, target identification records, risk events and operation response records collected throughout the entire process to generate a lifting task report and a safety risk report.

[0293] The report includes:

[0294] Operation task number, time, operation equipment and person in charge;

[0295] Total statistics of detected targets and average recognition accuracy;

[0296] The number, type distribution, and handling status of risk events;

[0297] Number of lifting interruptions and operator response time;

[0298] Summary of scores and recommendations for each indicator.

[0299] The report supports online viewing, PDF export, Word download, and uploading to the construction management platform, and has multi-dimensional query capabilities (filtering by time, area, risk level, event type, etc.).

[0300] Through the above-mentioned system process, the present invention realizes the intelligent management and control of the entire process of hoisting operations in shield construction and closed-loop risk control, and has strong feasibility, scalability and practicality.

[0301] In the attached Figure 13 middle,

[0302] Horizontal axis (X-axis): represents the spatial grid number of the construction site in the horizontal direction (such as east-west direction), usually in meters (m) or grids (cel l).

[0303] Vertical coordinate (Y axis): represents the spatial grid number of the construction site in the vertical direction (such as north-south direction), and the unit is the same as above.

[0304] The entire map consists of 10×10 grids, covering key operating areas of the construction site, such as gantry crane paths, personnel passages, material yards, etc.

[0305] Color meaning (risk level)

[0306] Green area: indicates that the location is currently in a low-risk state, usually far away from the lifting path or has no historical risk events;

[0307] Yellow area: indicates that the location is a medium-risk area, where there may be temporary passage, fast-moving targets, or close proximity of people;

[0308] Red area: Indicates that the location is a high-risk area, usually an intersection of lifting paths, an area with large lifting swings, or an area where warning events have occurred.

[0309] Figure 13 Role in the system: Assisting lifting dispatchers in real-time assessment of operation safety; identifying high-risk gathering areas in advance; used for optimizing construction area layout, setting early warning thresholds, and delineating emergency response areas.

[0310] It should be noted that, in this document, relational terms such as first and second, etc. are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

Claims

1. An intelligent material transportation and safety monitoring system for shield construction, characterized by: It includes visual perception unit, sensor network module, AI analysis center module, security decision module and human-computer interaction interface; The visual perception unit, as the core front-end perception module of the system, is deployed in the operating area of ​​the shield construction site. It is used to capture images and output video streams of dynamic targets in the operating area of ​​the shield construction site, providing raw data support for subsequent target recognition and path prediction. The operating area of ​​the shield construction site includes the hoisting operation area, the yard boundary, the component loading and unloading point, and the intersection of the transportation channel. The sensor network module, as a component for multi-source dynamic data collection and linkage analysis, is deployed on the gantry crane structure, work area, transport vehicles, and personnel's personal equipment at the construction site. It is used to obtain real-time equipment operating status, target location information, and lifting safety parameters, ensuring the system's full-time and space-time monitoring capabilities of the construction process. The AI ​​analysis center module is the core module for data calculation and intelligent judgment of the system, which is used to analyze, fuse and reason multi-source heterogeneous data from the visual perception unit and sensor network module to complete target detection tasks, trajectory prediction tasks, safety judgment tasks and risk output tasks; The safety decision module is a logical execution unit that implements automated lifting safety risk identification and response control. Based on the target recognition results, trajectory prediction information, and state perception data provided by the AI ​​analysis center module, combined with the built-in rule library, it dynamically determines high-risk behaviors that may occur at the construction site and triggers corresponding control instructions or early warning mechanisms. The human-computer interaction interface is an integral part of realizing information visualization, operation instruction interaction and data query management, and provides a unified, intuitive and real-time control and management channel for construction operators, safety management personnel and back-end dispatch personnel.

2. The intelligent material transportation and safety monitoring system for shield construction according to claim 1 is characterized by: The visual perception unit is composed of a camera device, which is a panoramic camera and a local high-definition camera. The camera device is connected to the edge server through an Ethernet interface or a wireless network. The image data is pushed to the AI ​​analysis center in real time in RTSP or H.265 encoding format. The image frame rate of the visual perception unit is not less than 25fps, and the transmission delay is controlled within 200ms.

3. The intelligent material transportation and safety monitoring system for shield construction according to claim 1 is characterized by: The dynamic targets in the working area of ​​the shield construction site include personnel, gantry cranes, pipe segments, box culverts, and vehicles. After the image acquisition is completed, it is input into the target detection model for real-time recognition. The recognition process is as follows: Target recognition process formula: Given an input image frame First, after the image preprocessing module, it is sent to the deep convolutional neural network to obtain the target candidate box set: in: (x i ,y i ) represents the center point coordinates of the i-th candidate box; w i , h i is the width and height of the candidate box; Indicates category; p i ∈[0,1] is the confidence of the target; Apply the non-maximum suppression algorithm to the detection results to remove duplicate frames and generate the final recognition target set: Where θ is the IoU threshold, which is usually set to 0.

5.

4. The material intelligent transportation and safety monitoring system for shield construction according to claim 1 is characterized in that: The sensor network module consists of the following subsystems: Gantry crane status monitoring subsystem: Multiple sets of industrial-grade sensors installed on the gantry crane frame and control cabinet are used to monitor the operating status of the lifting equipment, including: Displacement sensor: obtains the running distance and direction of the crane along the track; Load sensor: installed at the hook, lifting point, and rope to monitor the quality of the currently hoisted object in real time; Speed ​​sensor: used to measure the speed of the hook lifting and the vehicle body movement, with a typical sampling frequency of 10Hz; Tilt angle sensor or gyroscope: determines the shaking amplitude or posture change of the hoisting component; Personnel Positioning and Identification Subsystem: To ensure real-time awareness of workers in the lifting area, the system requires construction workers to be equipped with UWB positioning tags or RFID / Bluetooth low-power devices, which, combined with on-site base stations, achieve two-dimensional positioning capabilities. The positioning system is based on the TOF ranging principle and combines the TDOA algorithm to solve the position coordinates. Its basic positioning formula is: in: Δt ij is the time difference between the i-th base station and the j-th base station receiving the signal; d i , d j is the distance from the target to base station i, j; c is the wireless signal propagation speed; Combining the positioning geometry array composed of multiple base stations, the target position (x, y) can be obtained by least squares inversion; Transport vehicle tracking subsystem: GNSS positioning modules and IMUs are installed on transport vehicles to monitor trajectory, speed, and acceleration information. Multi-source data fusion and synchronization: The data collected by the sensor network module is standardized and encapsulated using a unified timestamp and coding identifier, and transmitted to the AI ​​analysis center via the MQTT or Modbus-TCP protocol. In conjunction with the clock synchronization module, the NTP protocol is used to align the time of all terminal devices to a millisecond-level error range to ensure the timing consistency of multi-channel data.

5. The material intelligent transportation and safety monitoring system for shield construction according to claim 1 is characterized in that: The AI ​​analysis center module is deployed in the on-site edge server or connected to the engineering data platform through the cloud-edge collaborative architecture. Its core functions include: multi-target recognition and classification, multi-source data fusion, trajectory prediction and conflict analysis, risk rule reasoning and alarm triggering; The multi-target recognition and classification is used to process image data and identify target objects on the scene, input image frame After feature extraction network f θ Extract the feature map F and generate the target set through the candidate box predictor Use the following formula: F=f θ (I), in: b i =(x i ,y i ,w i ,h i ) represents the bounding box of the i-th target; c i Indicates category (such as construction workers, gantry cranes, box culverts, etc.); s i Indicates confidence; The image targets identified by the multi-source data fusion mechanism are spatially mapped and temporally aligned with the physical coordinate data acquired by the sensor, and the Hungarian matching algorithm and Kalman filter are used for target association and tracking: Kalman filter state model: Let the target state be x t =[x,y,v x , v y ] T , the observed value is zt = [x, y] T , then the state update and observation update process is: x t|t-1 =Ax t-1|t-1 ,z t =Hx t|t-1 +w t Where: A is the state transfer matrix, H is the observation matrix, Hungarian algorithm matching matrix calculation formula: Constructing the cost matrix Among them C ij =d(p i ,q j ) represents the image target p i With sensor target q j Euclidean distance or IoU cost, solving the minimum total matching cost; The trajectory prediction and conflict analysis utilizes the AI ​​analysis center module to predict the execution trajectory of construction personnel, lifting equipment, and transport vehicles, using a multi-step linear regression model or an LSTM-based time series prediction model; The prediction formula is as follows: Where: x t is the current position, v t is the current speed, k is the predicted step size; Predicting trajectories for multiple targets Execute conflict judgment: Where δ is the conflict distance threshold; The safety rule reasoning and warning output utilizes the AI ​​analysis center module to build a logic judgment rule library, and outputs risk level judgment based on target type, spatial position, speed status and prediction results.

6. The intelligent material transportation and safety monitoring system for shield construction according to claim 1 is characterized in that: The security decision module includes four parts: rule knowledge base, real-time judgment engine, response control unit, risk classification and execution interface; The rule knowledge base is the basic logical unit of the safety decision-making module. It has a built-in set of judgment conditions and corresponding response strategies for common risk scenarios at shield construction sites. The rule form supports hard logic judgment and fuzzy matching logic. The real-time judgment engine is responsible for matching the target detection results sent by the AI ​​analysis center with the rule library. It uses an event trigger mechanism to poll the input cache every 100ms, matching all identified targets and their predicted trajectories with the rule conditions one by one, and generating a risk event queue. The judgment process is as follows: Receive the current frame target state set S t ={s1, s2, ..., s n } For each target s i , call its status parameters; Match all rules R j , when R is satisfied j When the trigger condition is met, add it to the risk event queue E t ; Output event E t The highest level of risk among all the risks and transfer them to the response control unit; The response control unit executes a hierarchical response strategy based on the risk level; The risk event recording and backtracking mechanism automatically generates the following structured log entries after a risk event is detected. The logs are uploaded to the security management platform for responsibility backtracking, risk statistics, management assessment, and subsequent optimization. The system stability and fault tolerance mechanism has the following fault tolerance strategies: When the data delay exceeds 500ms, the system suspends the action response and prompts "data synchronization abnormality"; When the key target cannot be identified for three consecutive frames, the fault protection mechanism is triggered; A manual review entry is set up to deal with the risks of false alarms or missed alarms, and corrections and confirmations are made through a secure terminal.

7. A method for intelligent material transportation and safety monitoring for shield construction, using the intelligent material transportation and safety monitoring system for shield construction according to any one of claims 1 to 6 for monitoring, characterized in that: The specific steps include: Step 1: Data collection, the data collection task is initiated by multiple types of perception devices deployed at the construction site, including visual perception units and sensor network modules; Step 2: Target recognition and trajectory prediction. After the data is uploaded to the AI ​​analysis center module, the system calls the deployed target recognition model to perform multi-target detection operations on the image frame, identifying and extracting the target type, spatial coordinates, contour boundaries, and confidence level. Step 3: Safety judgment and warning output: The system integrates the target recognition results and trajectory prediction results, calls the rule knowledge base built into the safety decision module, and matches each rule to see if there are potential risk events. Step 4: Intervention and recording of hoisting operations. When the identification results trigger a medium- or high-level risk event, the system immediately sends a control signal through the safety decision module, interrupts the current action of the hoisting equipment through the API or PLC interface, and locks the control authority. Step 5: Report generation and data archiving. After the lifting operation is completed, the system will structure the images, sensor data, target identification records, risk events and operation response records collected throughout the entire process to generate a lifting task report and a safety risk report.

8. The method for intelligent material transportation and safety monitoring for shield construction according to claim 7, characterized in that: The data collection in step 1 includes collecting dynamic target images and videos of the working area of ​​the shield construction site through the visual perception unit, and collecting the operating status, target position information and lifting safety parameters of the dynamic targets in the working area of ​​the shield construction site through the sensor network module.

9. The method for intelligent material transportation and safety monitoring for shield construction according to claim 7, characterized in that: In step 2, when performing trajectory prediction on the historical position data of consecutive frames, a multi-step linear regression model or LSTM neural network is used to estimate the spatial path within several future time steps. The prediction expression is as follows: Where: x t is the current target position; v t is the estimated speed; k is the prediction step size (e.g. 1 to 5 seconds); For future location; The prediction results are superimposed on the monitoring interface in the form of trajectory lines and passed as input to the safety judgment module.

Images