Multi-source sensor data fusion large equipment transportation safety threshold dynamic decision system

Abstract

The application discloses a large-scale equipment transportation safety threshold dynamic decision system based on multi-source sensing data fusion, comprising: a sensor integration layer; a space-time preprocessing layer; a working stage inference layer; a threshold evolution management layer; a disturbance identification and elimination layer, which is used for distinguishing short-time non-structural disturbance based on positioning trajectory change rate, attitude differential change and video image matching degree, and stripping the disturbance influence from the fusion index to avoid false alarm; a safety decision layer; a visual decision interface, which is used for synchronously presenting a digital twin model, fusion results, threshold evolution trajectory, working stage and video image, and displaying abnormal contribution degree, sensor health inference result and disposal suggestion; a communication and recording layer, which is used for uploading fusion data, alarm events, threshold evolution information and video clips to a command center, and storing complete operation logs. The application has strong anti-disturbance ability, the multi-source disturbance identification model effectively filters short-time disturbance, and reduces misjudgment; the application has prediction ability, and short-time trajectory prediction gives safety early warning in advance.

Description

Technical Field

[0001] This invention relates to the technical field of large equipment transportation, and in particular to a dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion. Background Technology

[0002] Large equipment (such as large components, bridge sections, dock modules, heavy machinery parts, etc.) exhibits multi-dimensional, nonlinear, and strongly coupled characteristics in its attitude, stress, displacement, and path during transportation. Traditional transportation monitoring systems typically rely on single sensors or distributed sensing devices and employ a "fixed threshold + linear judgment" approach for safety monitoring. Such systems generally suffer from the following technical problems: 1) Thresholds lack adaptive capability and cannot adapt to changes in mechanical characteristics at different operational stages. Existing monitoring systems typically set fixed "safety thresholds, warning thresholds, and limit thresholds." However, the dynamic responses of large equipment vary greatly during Top-Up, Translation, Crossing, Boarding, and Landing. Fixed thresholds can lead to: misinterpretations of displacement exceeding limits during the Top-Up stage; misjudgments of attitude changes during the Translation stage; short-term disturbances being treated as dangerous during the Crossing stage; and numerous false alarms during stage transitions. 2) Non-structural disturbances during transportation may falsely trigger the threshold. When large equipment operates through ramps, track joints, short-term braking, and turns, it may experience: instantaneous peaks in tilt angle; sudden jumps in displacement; camera shake; and instantaneous attitude changes caused by ramps. These disturbances do not pose a danger, but traditional systems cannot distinguish between "short-term disturbances" and "structural risks," ultimately leading to a high false alarm rate.

[0003] To address these issues, we propose a dynamic decision-making system for safety thresholds in the transportation of large equipment, based on multi-source sensor data fusion. Summary of the Invention

[0004] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0005] In view of the problems existing in the dynamic decision-making system for safety thresholds in large equipment transportation based on the fusion of multi-source sensor data, this invention is proposed.

[0006] Therefore, the purpose of this invention is to provide a dynamic decision-making system for safety thresholds in the transportation of large equipment based on multi-source sensor data fusion. This system has strong anti-disturbance capabilities, and its multi-source disturbance identification model effectively filters short-term disturbances, reducing misjudgments. It also has predictive capabilities, providing early safety warnings through short-term trajectory prediction.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion, comprising: The sensor integration layer includes tilt sensors, horizontal / vertical displacement sensors, strain / stress sensors, laser displacement sensors, positioning terminals, and video acquisition units, used to collect raw multi-source data characterizing the device's attitude, deformation, force, and environment. The spatiotemporal preprocessing layer is used to perform unified time base calibration, outlier removal, noise suppression and dimensional normalization on the raw data, and generate a temporal feature matrix that can participate in fusion. A hierarchical credibility fusion layer fuses the temporal feature matrices. The operation phase inference layer, based on the trend characteristics of fused data, acceleration inference, and spatial geometric features generated by the positioning trajectory, uses the phase identification rule set to infer whether the current operation phase is lifting, translation, over-jumping, boarding, or rail dropping. The threshold evolution management layer dynamically generates safety thresholds, warning thresholds, and extreme thresholds based on the inferred operational stage, short-term fluctuations in fused data, and historical statistical baselines through an adaptive threshold evolution map model, and establishes a record of the sources of threshold changes. The disturbance identification and resolution layer is used to identify short-term unstructured disturbances based on the matching degree between the positioning trajectory change rate, attitude differential change and video image, and to separate the disturbance impact from the fusion index to avoid false alarms. At the security decision-making level, the integrated indicators are compared with the dynamic threshold set, and a hierarchical alarm is generated according to the composite triggering rules. An abnormal contribution back-inference mechanism is used to identify the key variables that trigger the alarm. A visual decision-making interface is used to simultaneously present the digital twin model, fusion results, threshold evolution trajectory, operation stage and video images, and display the anomaly contribution, sensor health inference results and handling suggestions; The communication and logging layer is used to upload fused data, alarm events, threshold evolution information and video clips to the command center, and to store complete operation logs.

[0008] As a preferred embodiment of the dynamic decision-making system for large equipment transportation safety thresholds based on multi-source sensor data fusion described in this invention, the fusion of the time-series feature matrices specifically includes the following methods: A dynamic confidence vector is constructed based on sensor health, data stability, historical consistency, and environmental interference factors. The attitude fusion vector, displacement fusion vector, and stress fusion vector are generated through a hierarchical credibility fusion model of "class-by-class index differential consistency verification + global weighted summation". In this process, each type of fusion result is calculated by at least two types of basic measurement data, and the positioning data and attitude data are used together to identify abnormal disturbances.

[0009] As a preferred embodiment of the dynamic decision-making system for large equipment transportation safety thresholds based on multi-source sensor data fusion described in this invention, the dynamic confidence vector is generated in the following manner: Health factors are calculated using a sensor-based health inference network. Volatility factor is calculated using window stability analysis; The communication quality factor is calculated using the delay estimation module; The deviation consistency factor is calculated through historical consistency analysis. After normalization, each factor is fused according to a preset weight to form the dynamic confidence level of each type of sensor; Furthermore, when the health factor of any sensor drops below the threshold, the confidence level of that sensor automatically drops to the lowest level and triggers the intervention of the backup channel.

[0010] As a preferred embodiment of the dynamic decision-making system for large equipment transportation safety thresholds based on multi-source sensor data fusion described in this invention, the disturbance identification and resolution layer adopts a "trajectory geometry and attitude difference joint disturbance model," comprising: Calculate the short-term path curvature change based on the positioning coordinate sequence; The intensity of abrupt attitude changes is calculated based on the temporal difference of the attitude fusion vector. Environmental change factors are determined by comparing the degree of keyframe matching in video footage. The disturbance level was determined by a joint assessment of the three factors. When the disturbance level is a short-term unstructured disturbance, the system will remove the corresponding disturbance impact from the fused attitude vector and send a "disturbance suppression signal" to the threshold evolution management layer to avoid misjudgment.

[0011] As a preferred embodiment of the dynamic decision-making system for large equipment transportation safety thresholds based on multi-source sensor data fusion described in this invention, the operation phase inference layer identifies the operation status in the following manner: The trend of velocity change is obtained by utilizing the trend of the displacement fusion vector; Utilize the geometric features of the positioning trajectory, including path inclination, turning points, and the shape of the area it crosses; Load response patterns utilizing attitude vectors and stress vectors; By combining three types of features, the system automatically determines whether the equipment is in the lifting, translation, over-jump, boarding, or track-dropping stage, with an error recognition rate lower than a preset threshold.

[0012] As a preferred embodiment of the dynamic decision-making system for large-scale equipment transportation safety thresholds based on multi-source sensor data fusion described in this invention, the adaptive threshold evolution graph model includes: "Local response subgraphs" generated based on short-term statistics from fused data; "Long-term stable subgraph" generated based on historical stability statistics; "Stage feature sub-graph" generated based on existing operation stages; The three types of subgraphs are assembled into a threshold evolution graph according to the node association method. The safety threshold, warning threshold and limit threshold are updated through the node weight diffusion mechanism in the graph structure, so that the threshold adjustment at different stages is interpretable and continuous.

[0013] As a preferred embodiment of the dynamic decision-making system for large equipment transportation safety thresholds based on multi-source sensor data fusion described in this invention, the sensor redundancy switching is performed according to a "three-factor switching rule," namely: If any two of the health factor, deviation consistency factor, and communication quality factor fall below the threshold, the main sensor will be deactivated and the redundant sensor will be activated. After the switchover, a consistency check is performed on the primary / backup data within 50 sampling periods to determine whether to restore the primary channel; This redundancy strategy can reduce frequent switching caused by a single anomaly and improve system stability.

[0014] As a preferred embodiment of the dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion described in this invention, the safety decision-making layer further includes a short-term trajectory prediction module, which uses fused positioning vectors and displacement trend vectors to predict the motion trend over several future periods. When the predicted value is about to reach a future threshold, a "progressive warning" is triggered in advance, forming an early warning capability that is earlier than traditional threshold judgment.

[0015] As a preferred embodiment of the dynamic decision-making system for large equipment transportation safety thresholds based on multi-source sensor data fusion described in this invention, the video image is bidirectionally verified with the displacement fusion vector through frame difference and object displacement estimation. When the deviation between the two continuously exceeds a threshold: Update the video confidence factor; Adjust the fusion weight distribution; Mark the abnormal situation as a visual aid abnormality; To achieve dynamic control of the credibility of visual information participation in fusion.

[0016] A dynamic decision-making method for transportation safety thresholds based on multi-source sensor data fusion includes the following steps: Step 1: Multi-source data acquisition and preprocessing; Perform time calibration and normalization on attitude, displacement, stress, positioning and video data to generate a temporal feature matrix; Step 2: Hierarchical credibility fusion; The temporal feature matrix is ​​sequentially weighted and fused based on the dynamic confidence vector, and the fusion result is synchronized to the stage inference layer. Step 3: Task Phase Inference; Determine the current operation stage using trend quantities, geometric quantities, and load response patterns; Step 4: Threshold evolution; A dynamic threshold set is generated by combining the threshold evolution graph model with the operation stage and historical baseline; Step 5: Disturbance identification and removal; Identify short-term disturbances and remove their impact to obtain stable fused data; Step 6: Multi-trigger decision-making and contribution back-calculation; The alarm level is determined based on the dynamic threshold set, and the contribution of each variable is calculated. Step 7: Visual presentation and recording; Display digital twins, threshold evolution, anomaly contribution, and video information, and record various data; Step 8: Closed-loop update and health assessment; The confidence vector is updated based on the sensor health analysis results to complete the next cycle of closed loop.

[0017] The beneficial effects of this invention are: This invention is highly adaptive, based on stage recognition and threshold evolution, enabling the system to automatically adapt to different transportation processes; High reliability and robustness: The four-factor confidence model plus redundancy mechanism results in fewer fusion value anomalies. It has strong anti-disturbance capabilities, and the multi-source disturbance identification model effectively filters short-term disturbances, reducing false judgments; It has predictive capabilities, providing early safety warnings through short-term trajectory prediction; Multimodal fusion is more comprehensive, and bidirectional verification between video information and sensor data improves system reliability. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic diagram of the system structure of the dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion, as described in this invention.

[0019] Figure 2 This is a schematic diagram illustrating the method steps of the dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion, as described in this invention.

[0020] Figure 3 This is a real-time attitude diagram of a gantry crane for the dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion, as described in this invention. Detailed Implementation

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0024] Secondly, the present invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of the present invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not according to the usual scale. Furthermore, the schematic diagrams are merely examples and should not limit the scope of protection of the present invention. In addition, actual fabrication should include three-dimensional spatial dimensions of length, width, and depth.

[0025] Reference Figures 1-3 It provides a dynamic decision-making system for safety thresholds in the transportation of large equipment based on multi-source sensor data fusion, including: The sensor integration layer includes tilt sensors, horizontal / vertical displacement sensors, strain / stress sensors, laser displacement sensors, positioning terminals, and video acquisition units, used to collect raw multi-source data characterizing the device's attitude, deformation, force, and environment. The spatiotemporal preprocessing layer is used to perform unified time base calibration, outlier removal, noise suppression and dimensional normalization on the raw data, and generate a temporal feature matrix that can participate in fusion. A hierarchical credibility fusion layer fuses the temporal feature matrices. The operation phase inference layer, based on the trend characteristics of fused data, acceleration inference, and spatial geometric features generated by the positioning trajectory, uses the phase identification rule set to infer whether the current operation is in the lifting, translation, over-jump, boarding, or rail-dropping phase. The threshold evolution management layer dynamically generates safety thresholds, warning thresholds, and extreme thresholds based on the inferred operational stage, short-term fluctuations in fused data, and historical statistical baselines through an adaptive threshold evolution map model, and establishes a record of the sources of threshold changes. The disturbance identification and resolution layer is used to identify short-term unstructured disturbances based on the matching degree between the positioning trajectory change rate, attitude differential change and video image, and to separate the disturbance impact from the fusion index to avoid false alarms. At the security decision-making level, the integrated indicators are compared with the dynamic threshold set, and a hierarchical alarm is generated according to the composite triggering rules. An abnormal contribution back-inference mechanism is used to identify the key variables that trigger the alarm. A visual decision-making interface is used to simultaneously present the digital twin model, fusion results, threshold evolution trajectory, operation stage and video images, and display the anomaly contribution, sensor health inference results and handling suggestions; The communication and logging layer is used to upload fused data, alarm events, threshold evolution information and video clips to the command center, and to store complete operation logs.

[0026] The fusion of the temporal feature matrices specifically includes the following methods: A dynamic confidence vector is constructed based on sensor health, data stability, historical consistency, and environmental interference factors. The attitude fusion vector, displacement fusion vector, and stress fusion vector are generated through a hierarchical credibility fusion model of "class-by-class index differential consistency verification + global weighted summation". In this process, each type of fusion result is calculated by at least two types of basic measurement data, and the positioning data and attitude data are used together to identify abnormal disturbances.

[0027] The dynamic confidence vector is generated as follows: Health factors are calculated using a sensor-based health inference network. Volatility factor is calculated using window stability analysis; The communication quality factor is calculated using the delay estimation module; The deviation consistency factor is calculated through historical consistency analysis. After normalization, each factor is fused according to a preset weight to form the dynamic confidence level of each type of sensor; Furthermore, when the health factor of any sensor drops below the threshold, the confidence level of that sensor automatically drops to the lowest level and triggers the intervention of the backup channel.

[0028] The dynamic confidence vector is used to describe the confidence level of different sensors within a certain sampling period for the same type of monitored quantity (such as attitude, displacement, and stress). Its function is to: Automatically identify sensors with poor quality before data fusion; The weights are automatically reduced when there is sensor failure, communication anomaly, or increased noise fluctuation. Ensure that the fusion results are not contaminated by a single abnormal source.

[0029] Therefore, this system constructs a four-factor dynamic confidence model for each sensor, with values ​​ranging from 0 to 1.

[0030] Specifically, the overall structure of the dynamic confidence vector is as follows: Construct a dynamic confidence vector for N sensors:

[0031] in, This represents the confidence level of the i-th sensor at time t; ; Each Calculated from four types of factors:

[0032] in: Indicates health factor; Indicates the stability factor; Indicates the communication quality factor; Indicates the deviation consistency factor; + + + =1, which is a configurable weight; Typically, the following is selected: =0.4, =0.3, =0.2, =0.1; Define a primary / standby switchover threshold:

[0033] The system will automatically switch over when the following conditions are met: The triggering conditions must meet at least two of the following:

[0034]

[0035]

[0036] Switching rules: like If the current channel is marked as "degraded", the backup channel is started and upgraded to the default trust level of 0.6; the fusion layer automatically adjusts the weight matrix. Steady-state recovery mechanism: If within 50 periods Only then was the main channel restored.

[0037] Furthermore, the disturbance identification and resolution layer adopts a "trajectory geometry and attitude difference joint disturbance model", including: Calculate the short-term path curvature change based on the positioning coordinate sequence; The intensity of abrupt attitude changes is calculated based on the temporal difference of the attitude fusion vector. Environmental change factors are determined by comparing the degree of keyframe matching in video footage. The disturbance level was determined by a joint assessment of the three factors. When the disturbance level is a short-term unstructured disturbance, the system will remove the corresponding disturbance impact from the fused attitude vector and send a "disturbance suppression signal" to the threshold evolution management layer to avoid misjudgment.

[0038] Specifically, the overall framework of the perturbation recognition model: The disturbance identification and resolution layer is based on time. Multiple types of data serve as input, and the output is the perturbation level. and the fused attitude vector after perturbation removal , Input data definition: make: : Locate coordinates (x, y, z) Attitude fusion vector (tilt / yaw / pitch) Video displacement estimation; : The pose fusion vector from the previous time step; Sampling period time; Output: Disturbance level:

[0039] 0: No disturbance; 1: Slight disturbance; 2: Moderate disturbance; 3: High-intensity disturbance; Disturbance glass attitude ; when And when the probability of non-structural disturbances is high: The threshold evolution management layer receives "perturbation suppression signals". Temporarily postpone threshold overflow determination Avoid false alarms triggered by short-term spikes; when : Transformed into a "real suspicious event" Proceed directly to the alarm determination process.

[0040] The job phase inference layer identifies the job status in the following manner: The trend of velocity change is obtained by utilizing the trend of the displacement fusion vector; Utilize the geometric features of the positioning trajectory, including path inclination, turning points, and the shape of the area it crosses; Load response patterns utilizing attitude vectors and stress vectors; By combining three types of features, the system automatically determines whether the equipment is in the lifting, translation, over-jump, boarding, or track-dropping stage, with an error recognition rate lower than a preset threshold.

[0041] The operational phases of this system include: Phase S1: Climb Phase S2: Translation Phase S3: Overjump Phase S4: Boarding Phase S5: Landing on Rails Assignment Stage Tags: ; Specifically, trend data is mainly used to identify "movement up / down / forward / acceleration / deceleration": Displacement velocity / acceleration trend: Horizontal speed: The adaptive threshold evolution map model includes: "Local response subgraphs" generated based on short-term statistics from fused data; "Long-term stable subgraph" generated based on historical stability statistics; "Stage feature sub-graph" generated based on existing operation stages; The three types of subgraphs are assembled into a threshold evolution graph according to the node association method. The safety threshold, warning threshold and limit threshold are updated through the node weight diffusion mechanism in the graph structure, so that the threshold adjustment at different stages is interpretable and continuous.

[0042] Displacement velocity / acceleration trend: Horizontal speed:

[0043] Vertical velocity:

[0044] Examples of trend judgment are shown in Table 1 below: Table 1

[0045] The trend score can be represented as:

[0046] Specifically, the sensor redundancy switching is performed according to the "three-factor switching rule," that is: If any two of the health factor, deviation consistency factor, and communication quality factor fall below the threshold, the main sensor will be deactivated and the redundant sensor will be activated. After the switchover, a consistency check is performed on the primary / backup data within 50 sampling periods to determine whether to restore the primary channel; This redundancy strategy can reduce frequent switching caused by a single anomaly and improve system stability.

[0047] The safety decision layer also includes a short-term trajectory prediction module, which uses the fusion of positioning vector and displacement trend vector to predict the motion trend for several future cycles. When the predicted value is about to reach a future threshold, it triggers a "progressive warning" in advance, forming a warning capability earlier than the traditional threshold judgment.

[0048] Furthermore, the video image undergoes bidirectional verification through frame differencing and object displacement estimation, along with the displacement fusion vector. When the deviation between the two continuously exceeds a threshold: Update the video confidence factor; Adjust the fusion weight distribution; Mark the abnormal situation as a visual aid abnormality; To achieve dynamic control of the credibility of visual information participation in fusion.

[0049] The dynamic decision-making method for transportation safety thresholds based on multi-source sensor data fusion includes the following steps: Step 1: Multi-source data acquisition and preprocessing; Perform time calibration and normalization on attitude, displacement, stress, positioning and video data to generate a temporal feature matrix; Step 2: Hierarchical credibility fusion; The temporal feature matrix is ​​sequentially weighted and fused based on the dynamic confidence vector, and the fusion result is synchronized to the stage inference layer. Step 3: Task Phase Inference; Determine the current operation stage using trend quantities, geometric quantities, and load response patterns; Step 4: Threshold evolution; A dynamic threshold set is generated by combining the threshold evolution graph model with the operation stage and historical baseline; Step 5: Disturbance identification and removal; Identify short-term disturbances and remove their impact to obtain stable fused data; Step 6: Multi-trigger decision-making and contribution back-calculation; The alarm level is determined based on the dynamic threshold set, and the contribution of each variable is calculated. Step 7: Visual presentation and recording; Display digital twins, threshold evolution, anomaly contribution, and video information, and record various data; Step 8: Closed-loop update and health assessment; The confidence vector is updated based on the sensor health analysis results to complete the next cycle of closed loop.

[0050] Specific examples are as follows: This embodiment uses a self-propelled modular transporter (SPMT) as the transport vehicle to describe in detail the specific implementation of the system of the present invention for the overall translation and loading and transfer operation of a 1200t large gantry crane in a port. The SPMT possesses multi-axle hydraulic independent suspension, modular assembly, synchronous lifting, and precise steering capabilities, making it a core piece of equipment for the overall transfer of heavy equipment. However, issues such as attitude synchronization errors, road disturbance transmission, and load redistribution during multi-module collaborative operation place extremely high demands on the real-time performance and accuracy of safety monitoring.

[0051] 1. System hardware deployment and sensor integration layer configuration For the SPMT + gantry crane transfer system, the sensor integration layer adopts a two-dimensional distributed deployment scheme of "SPMT vehicle body + gantry crane body", with a total of 68 sensor nodes deployed, as detailed below: SPMT vehicle body sensing unit: One pressure sensor (48 in total) is deployed at each suspension cylinder of the 4 groups of SPMT modules (12 axes per group). One dual-axis tilt sensor (8 in total) and one laser displacement sensor (8 in total) are deployed at the front and rear ends of each group of modules. One RTK-GNSS positioning terminal (positioning accuracy ±1cm) and one inertial measurement unit (IMU) are deployed in the SPMT main control vehicle to collect SPMT attitude, suspension load, driving displacement and positioning information.

[0052] The gantry crane body sensing unit consists of eight strain gauges (one at the top and one at the bottom of each of the four outriggers), one three-dimensional displacement sensor and one tilt sensor at the mid-span of the main beam, and four vertical displacement sensors at the saddles where the outriggers connect to the SPMT. These sensors are used to collect data on the structural stress, deformation, and overall posture of the gantry crane.

[0053] Video acquisition unit: One high-definition industrial camera is deployed at each of the four corners of the SPMT (4 in total), and one panoramic camera is deployed at each end of the main beam of the gantry crane (2 in total) to collect information on road conditions, saddle connection status and surrounding environment.

[0054] All sensors are connected to the edge computing gateway via industrial Ethernet, with a uniform sampling frequency of 50Hz and a video acquisition frame rate of 25fps to ensure data time synchronization.

[0055] 2. Spatiotemporal preprocessing layer data processing flow The edge computing gateway performs the following preprocessing operations on the raw multi-source data: Unified time base calibration: Based on the PPS second pulse signal of RTK-GNSS, all sensor data are timestamped and aligned, with the maximum time error controlled within 1ms; Abnormal sample removal: The 3σ criterion is used to remove abrupt data that exceeds the physical reasonable range (such as instantaneous over-range values ​​of pressure sensors and jump values ​​of displacement sensors). Noise suppression: Low-frequency signals such as pressure and strain are filtered by moving average (window size 5), and high-frequency signals such as tilt angle and acceleration are filtered by Kalman filter to suppress mechanical vibration and electromagnetic interference; Dimensional normalization: Map all physical quantities to the [0,1] interval to generate a time-series feature matrix with a dimension of 68×T (T is the number of sampling periods), which is then input to the hierarchical credibility fusion layer.

[0056] 3. Specific implementation of the hierarchical credibility fusion layer In this embodiment, the hierarchical reliability fusion layer fuses 68 channels of raw data into four core indices: attitude fusion vector, displacement fusion vector, stress fusion vector, and load fusion vector. The specific process is as follows: Dynamic confidence vector construction: For each type of sensor, a health factor is calculated through a sensor health inference network. Window stability analysis to calculate volatility factor The delay estimation module calculates the communication quality factor. Historical consistency analysis calculates the deviation consistency factor. By weight =0.4, =0.3, =0.2, =0.1 Calculate the confidence level of a single sensor For example, when a certain SPMT suspension pressure sensor experiences fluctuations in hydraulic oil, resulting in a change in its volatility factor... When the confidence level is 0.3, its confidence level is automatically adjusted to... =0.4×1+0.3×0.3+0.2×1+0.1×1=0.79, and the fusion weight is reduced accordingly.

[0057] Hierarchical fusion calculation: The method of "differential consistency verification of each category of index + global weighted summation" is adopted, and at least two valid data sources of the same category of index participate in the fusion. For example, the attitude fusion vector is generated by fusing data from SPMT tilt sensor, IMU and gantry crane tilt sensor, and the displacement fusion vector is generated by fusing data from RTK-GNSS, laser displacement sensor and three-dimensional displacement sensor, to ensure the redundancy and reliability of the fusion results.

[0058] Sensor redundancy switching: When any two of the following parameters of a primary sensor—health, communication quality, and deviation consistency factor—falls below 0.5, the "three-factor switching rule" is triggered, and the backup sensor automatically takes over. For example, when RTK-GNSS communication quality factor is lowered due to obstruction... =0.4, Deviation Consistency Factor When the value is 0.3, the system automatically switches to the positioning mode dominated by the laser displacement sensor and performs a re-examination of the RTK signal within 50 sampling cycles. When the signal is recovered and the confidence level is higher than 0.6, it smoothly switches back to the main channel.

[0059] 4. Stage Identification Process of the Operation Stage Inference Layer In this embodiment, the gantry crane relocation operation is divided into five stages: jacking (S1), site translation (S2), jump crossing (S3), boarding (S4), and track lowering (S5). The operation stage inference layer achieves automatic identification through joint matching of three types of features: Trend characteristics: Horizontal velocity is calculated using the displacement fusion vector. With vertical velocity For example, the peaking phase. ≈0、 >0 (vertically upward), translation phase >0.1m / s ≈0, track-setting phase ≈0、 <0 (vertically downward); Geometric features: The path inclination angle, turning points, and shape of the crossing area are calculated using RTK positioning trajectory. For example, during the over-jump phase, the trajectory exhibits obvious local curvature abrupt changes (corresponding to ground joints or elevation differences), and during the boarding phase, the trajectory changes from a straight line to a curve perpendicular to the dock shoreline; Load response characteristics: The stress changes in the structure are analyzed by combining the stress fusion vector and the load fusion vector. For example, during the lifting phase, the stress in all four outriggers increases synchronously, while during the boarding phase, the stress in one outrigger increases significantly due to the tilting of the hull.

[0060] In this embodiment, the stage identification error rate is less than 0.5% and the stage switching response time is less than 200ms, providing an accurate premise for dynamic threshold adjustment.

[0061] 5. Dynamic threshold generation of adaptive threshold evolution map model The threshold evolution management layer dynamically generates three levels of thresholds based on the current operational stage, combined with the local response subgraph, the long-term stability subgraph, and the stage characteristic subgraph. Taking the tilt angle threshold as an example, the threshold settings for different stages are as follows: Lifting stage (S1): The gantry crane slowly rises vertically with a gradual change in attitude. The safety threshold is set at 0.1°, the warning threshold at 0.2°, and the limit threshold at 0.3°. Translation phase (S2): SPMT travels at a constant speed, allowing small attitude fluctuations. The safety threshold is set at 0.2°, the warning threshold at 0.4°, and the limit threshold at 0.6°. Over-jump phase (S3): When the SPMT crosses the ground elevation difference, it will generate an instantaneous tilt angle peak. The system will automatically temporarily relax the safety threshold to 0.5°, the warning threshold to 0.8°, and the extreme threshold to 1.0° for a duration of 1.5 times the over-jump process. Boarding phase (S4): The hull is affected by waves and rolls. The threshold is adjusted in real time according to the hull attitude. The safety threshold evolves dynamically between 0.3° and 0.6°. Track lowering stage (S5): The gantry crane is precisely aligned, with strict posture requirements. The safety threshold is restored to 0.1°, the warning threshold is 0.2°, and the limit threshold is 0.3°.

[0062] All threshold adjustments are recorded from their source (e.g., "temporary relaxation during over-jump phase" and "adaptive adjustment of hull roll"), ensuring complete traceability. 6. The security decision-making layer compares the fused indicators after removing disturbances with the dynamic threshold set, and generates tiered alarms based on composite triggering rules: Level 1 Alarm (Warning): When the fusion index exceeds the safety threshold for 3 consecutive sampling cycles, an audio-visual alert is triggered. Level 2 Alarm (Warning): When the fused indicator exceeds the warning threshold for two consecutive sampling cycles, an alarm message is sent to the command center; Level 3 Alarm (Emergency): When the fusion indicator exceeds the limit threshold or multiple indicators exceed the limit simultaneously, the SPMT emergency shutdown command is immediately triggered.

[0063] Meanwhile, the safety decision-making level employs an anomaly contribution back-calculation mechanism to calculate the contribution ratio of each sensor's data to the alarm. For example, when the tilt angle of the left side of the gantry crane exceeds the limit, the system automatically identifies the tilt sensor of the left SPMT module and the strain sensor of the left outrigger as key variables, assisting maintenance personnel in quickly locating the problem.

[0064] The visual decision-making interface simultaneously displays the digital twin models of the SPMT and the gantry crane, real-time attitude curves, threshold evolution trajectories, work phase progress, and multiple video feeds. For example... Figure 3As shown, the interface clearly displays the operating status of each sensor, the current displacement distance, and the project progress. When an abnormality occurs, the abnormal location is automatically highlighted and a handling suggestion pops up (such as "Adjust the suspension pressure of the left SPMT to maintain a horizontal attitude").

[0065] This embodiment applies the system of the present invention to the relocation operation of a large gantry crane using SPMT (Special Power Transporter), solving the problems of high false alarm rate and poor adaptability of traditional fixed threshold monitoring systems. Actual testing shows that the system's false alarm rate has been reduced from over 30% in traditional systems to below 2%, with an early warning lead time of 5-10 seconds, effectively ensuring the safety of the 1200t gantry crane relocation process. Simultaneously, the system's multi-source data fusion and redundant design ensure stable operation even when some sensors fail, significantly improving the reliability and intelligence level of SPMT intelligent relocation.

[0066] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion, characterized in that, include: The sensor integration layer includes tilt sensors, horizontal / vertical displacement sensors, strain / stress sensors, laser displacement sensors, positioning terminals, and video acquisition units, used to collect raw multi-source data characterizing the device's attitude, deformation, force, and environment. The spatiotemporal preprocessing layer is used to perform unified time base calibration, outlier removal, noise suppression and dimensional normalization on the raw data, and generate a temporal feature matrix that can participate in fusion. A hierarchical credibility fusion layer fuses the temporal feature matrices; The operation phase inference layer, based on the trend characteristics of fused data, acceleration inference, and spatial geometric features generated by the positioning trajectory, uses the phase identification rule set to infer whether the current operation phase is lifting, translation, over-jumping, boarding, or rail dropping. The threshold evolution management layer dynamically generates safety thresholds, warning thresholds, and extreme thresholds based on the inferred operational stage, short-term fluctuations in fused data, and historical statistical baselines through an adaptive threshold evolution map model, and establishes a record of the sources of threshold changes. The disturbance identification and resolution layer is used to identify short-term unstructured disturbances based on the matching degree between the positioning trajectory change rate, attitude differential change and video image, and to separate the disturbance impact from the fusion index to avoid false alarms. At the security decision-making level, the integrated indicators are compared with the dynamic threshold set, and a hierarchical alarm is generated according to the composite triggering rules. An abnormal contribution back-inference mechanism is used to identify the key variables that trigger the alarm. A visual decision-making interface is used to simultaneously present the digital twin model, fusion results, threshold evolution trajectory, operation stage and video images, and display the anomaly contribution, sensor health inference results and handling suggestions; The communication and logging layer is used to upload fused data, alarm events, threshold evolution information and video clips to the command center, and to store complete operation logs.

2. The dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion according to claim 1, characterized in that: The fusion of the time-series feature matrices specifically includes the following methods: A dynamic confidence vector is constructed based on sensor health, data stability, historical consistency, and environmental interference factors. The attitude fusion vector, displacement fusion vector, and stress fusion vector are generated through a hierarchical credibility fusion model of "class-by-class index differential consistency verification + global weighted summation". In this process, each type of fusion result is calculated by at least two types of basic measurement data, and the positioning data and attitude data are used together to identify abnormal disturbances.

3. The dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion according to claim 2, characterized in that: The dynamic confidence vector is generated as follows: Health factors are calculated using a sensor-based health inference network. Volatility factor is calculated using window stability analysis; The communication quality factor is calculated using the delay estimation module; The deviation consistency factor is calculated through historical consistency analysis. After normalization, each factor is fused according to a preset weight to form the dynamic confidence level of each type of sensor; Furthermore, when the health factor of any sensor drops below the threshold, the confidence level of that sensor automatically drops to the lowest level and triggers the intervention of the backup channel.

4. The dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion according to claim 3, characterized in that: The disturbance identification and resolution layer adopts a "joint disturbance model of trajectory geometry and attitude difference", including: Calculate the short-term path curvature change based on the positioning coordinate sequence; The intensity of abrupt attitude changes is calculated based on the temporal difference of the attitude fusion vector. Environmental change factors are determined by comparing the degree of keyframe matching in video footage. The disturbance level was determined by a joint assessment of the three factors. When the disturbance level is a short-term unstructured disturbance, the system will remove the corresponding disturbance impact from the fused attitude vector and send a "disturbance suppression signal" to the threshold evolution management layer to avoid misjudgment.

5. The dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion according to claim 1, characterized in that: The job phase inference layer identifies the job status in the following manner: The trend of velocity change is obtained by utilizing the trend of the displacement fusion vector; Utilize the geometric features of the positioning trajectory, including path inclination, turning points, and the shape of the area it crosses; Utilizing the load response pattern of attitude vector and stress vector; By combining three types of features, the system automatically determines whether the equipment is in the lifting, translation, over-jump, boarding, or track-dropping stage, with an error recognition rate lower than a preset threshold.

6. The dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion according to claim 5, characterized in that: The adaptive threshold evolution map model includes: "Local response subgraphs" generated based on short-term statistics from fused data; "Long-term stable subgraph" generated based on historical stability statistics; "Stage feature sub-graph" generated based on existing operation stages; The three types of subgraphs are assembled into a threshold evolution graph according to the node association method. The safety threshold, warning threshold and limit threshold are updated through the node weight diffusion mechanism in the graph structure, so that the threshold adjustment at different stages is interpretable and continuous.

7. The dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion according to claim 1, characterized in that: The sensor redundancy switching is performed according to the "three-factor switching rule", namely: If any two of the health factor, deviation consistency factor, and communication quality factor fall below the threshold, the main sensor will be deactivated and the redundant sensor will be activated. After the switchover, a consistency check is performed on the primary / backup data within 50 sampling periods to determine whether to restore the primary channel; This redundancy strategy can reduce frequent switching caused by a single anomaly and improve system stability.

8. The dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion according to claim 1, characterized in that: The safety decision-making layer also includes a short-term trajectory prediction module, which uses the fusion of positioning vector and displacement trend vector to predict the motion trend for several future periods. When the predicted value is about to reach a future threshold, it triggers a "progressive warning" in advance, forming a warning capability that is earlier than the traditional threshold judgment.

9. The dynamic decision-making system for safety thresholds in large equipment transportation based on multi-source sensor data fusion according to claim 1, characterized in that: The video image is bidirectionally verified by frame difference and object displacement estimation, and then by the displacement fusion vector. When the deviation between the two continuously exceeds a threshold: Update the video confidence factor; Adjust the fusion weight distribution; Mark the abnormal situation as a visual aid abnormality; To achieve dynamic control of the credibility of visual information participation in fusion.

10. A method for dynamic decision-making on transportation safety thresholds based on multi-source sensor data fusion of a large equipment transportation safety threshold dynamic decision-making system according to any one of claims 1 to 9, characterized in that, Includes the following steps: Step 1: Multi-source data acquisition and preprocessing; Perform time calibration and normalization on attitude, displacement, stress, positioning and video data to generate a temporal feature matrix; Step 2: Hierarchical credibility fusion; The temporal feature matrix is ​​fused by a hierarchical weighting based on the dynamic confidence vector, and the fusion result is synchronized to the stage inference layer. Step 3: Task Phase Inference; Determine the current operation stage using trend quantities, geometric quantities, and load response patterns; Step 4: Threshold evolution; A dynamic threshold set is generated by combining the threshold evolution graph model with the operation stage and historical baseline; Step 5: Disturbance identification and removal; Identify short-term disturbances and remove their impact to obtain stable fused data; Step 6: Multi-trigger decision-making and contribution back-calculation; The alarm level is determined based on the dynamic threshold set, and the contribution of each variable is calculated. Step 7: Visual presentation and recording; Display digital twins, threshold evolution, anomaly contribution, and video information, and record various data; Step 8: Closed-loop update and health assessment; The confidence vector is updated based on the sensor health analysis results to complete the next cycle of closed loop.

Images