Digital twin system and method for manned special equipment

Abstract

The invention discloses a digital twin system for manned special equipment. The digital twin system comprises a physical space, a digital space, a service space and a digital main line, the physical space is used for collecting equipment operation state data and passenger behavior data; comprising a manned special equipment body and a distributed sensor deployment unit. The digital space is used for receiving the sensing data transmitted by the physical space and establishing virtual mirror image data of the manned special equipment, and at least comprises a fault diagnosis model for reflecting the mechanical state of the equipment and an equipment maintenance model for reflecting the behavior state of passengers; the service space is used for carrying out reasoning analysis based on the sensing data collected by the virtual mirror image and the physical space so as to generate a control instruction, and at least comprises an intelligent fault diagnosis model used for evaluating the health state of equipment and a behavior recognition model used for recognizing unsafe behaviors of passengers; and the digital main line is used for realizing real-time and bidirectional data synchronization and control instruction transmission among the physical space, the digital space and the service space.

Description

Technical Field

[0001] This invention relates to the field of special equipment safety monitoring technology, and in particular to a digital twin system and method for manned special equipment. Background Technology

[0002] As a core facility operating frequently in urban public spaces, the safety of manned special equipment directly affects the safety of public life and property. In recent years, with the increase in urban population density and the surge in equipment usage, the traditional "periodic maintenance + manual inspection" operation and maintenance model has become insufficient to meet the requirements of high-reliability operation, gradually exposing its shortcomings of slow response and insufficient hazard prediction.

[0003] Current mainstream video surveillance systems are mostly limited to manual, post-event tracing. Even when simple target detection algorithms are introduced in some scenarios, their design for security applications makes it difficult to meet the dynamic behavior recognition needs of special scenarios such as escalators and elevators. On the one hand, these devices have unique spatial structures such as the fluidity of steps and physical constraints between upper and lower areas. On the other hand, passenger behavior exhibits diverse, short-term, and non-linear characteristics, significantly limiting the accuracy and stability of general-purpose visual algorithms. More importantly, existing behavior recognition systems generally lack a linkage mechanism with the underlying operating system of the equipment, failing to form a closed-loop logic from behavior recognition to risk control, directly restricting the ability to provide real-time early warning and rapid response to risks.

[0004] Digital twin technology has been widely used in aerospace, high-end manufacturing, and civil engineering in recent years, enabling real-time mapping and intelligent decision-making throughout the entire lifecycle of equipment by constructing virtual mirrors. However, in the monitoring and management of manned special equipment, there is still a lack of a general digital twin system framework that can integrate multi-source information and has dynamic feedback capabilities, making it difficult to connect the entire process of "equipment status monitoring - passenger behavior recognition - operation and maintenance decision optimization".

[0005] Therefore, there is an urgent need to build a digital twin system for identifying unsafe passenger behaviors in the operation scenarios of manned special equipment. By integrating multi-source sensing data and intelligent algorithms, it can achieve precise perception of equipment status, accurate identification of risky behaviors, and predictive maintenance throughout the entire life cycle, thus filling the current technological gap in the safety management of special equipment. Summary of the Invention

[0006] To solve the above-mentioned technical problems, the present invention adopts the following solution.

[0007] A digital twin system for manned special equipment includes: physical space, digital space, service space, and a digital mainline connecting the three.

[0008] The physical space is used to collect equipment operating status data and passenger behavior data; it includes the main body of the manned special equipment and a distributed sensor deployment unit.

[0009] The digital space is used to receive sensor data transmitted from the physical space and establish virtual mirror data of manned special equipment, including at least a fault diagnosis model reflecting the mechanical state of the equipment and an equipment maintenance model reflecting the behavior of passengers.

[0010] The service space is used to perform reasoning analysis based on the sensor data collected from the virtual image and physical space to generate control commands, including at least an intelligent fault diagnosis model for assessing the health status of the equipment and a behavior recognition model for identifying unsafe passenger behaviors.

[0011] The digital mainline is used to achieve real-time, bidirectional data synchronization and control command transmission between the physical space, digital space and service space.

[0012] Optionally, the distributed sensor deployment unit includes: vibration sensors, electrical and environmental sensors, displacement and motion sensors, hazard detection sensors, and vision sensors.

[0013] Optionally, the special equipment for carrying passengers is an escalator, a passenger elevator, or a passenger ropeway.

[0014] Optionally, the intelligent fault diagnosis model is implemented through the following process: multi-domain feature extraction of sensor signals; screening and optimization of the extracted features; and classification of the screened features using a machine learning model to diagnose equipment fault types.

[0015] Optionally, the behavior recognition model is implemented through the following process: extracting passenger posture key points from the video stream based on the posture detection model; extracting and fusing multi-dimensional behavior features from the key points; and classifying the fused features based on preset behavior recognition logic to determine the type of unsafe behavior.

[0016] Optionally, the pose detection model is a YOLOv11 model trained on a custom dataset.

[0017] Optionally, the unsafe behavior types include at least one of falling, squatting, walking backwards, crossing boundaries, and crowding.

[0018] Optionally, the service space also includes an operational fault model, which integrates data from multiple sensors in the distributed sensor system and assesses security threats to the device's operating environment and operational level based on rule threshold analysis and pattern analysis.

[0019] Optionally, the digital mainline includes: a maintenance thread for synchronizing historical maintenance data and real-time status data; an operation thread for synchronizing physical space sensor data to the digital space; and a control thread for transmitting control commands generated in the service space to the physical space.

[0020] A digital twin method for manned special equipment, the method comprising:

[0021] S1: Real-time collection of operational data and passenger behavior data of manned special equipment through distributed sensors deployed on the physical body of the manned special equipment;

[0022] S2: The collected data is transmitted to the digital space through the operation thread of the digital main line. A virtual image of the manned special equipment is established in the digital space, and the transmitted data is preprocessed and stored.

[0023] S3: Based on the real-time sensing data from step S1 and the virtual mirror data from step S2, perform inference analysis in the service space to generate control commands.

[0024] Compared with the prior art, the present invention has the following beneficial technical effects:

[0025] The four-in-one digital twin architecture achieves semantic consistency and bidirectional interaction among physical entities, virtual models, and intelligent services, filling the gap in human-machine collaborative safety monitoring. The fusion of behavior recognition and fault diagnosis modules simultaneously monitors unsafe passenger behavior and equipment mechanical failures, improving the comprehensiveness and accuracy of monitoring. Multi-source sensor system-level perception, with seven types of sensors fully covering the equipment itself and its surrounding environment, constitutes a complete perception system, enhancing the system's anti-interference capabilities. Customized escalator behavior recognition algorithms, combined with the YOLOv11 posture model and scenario-based behavior classification logic, improve the accuracy and robustness of behavior recognition in special equipment scenarios. High-precision fault classification is achieved through multi-domain feature extraction and machine learning diagnostic processes. Real-time, traceable data flow across the physical-digital-service space is realized through digital masterline full lifecycle data management, enhancing the system's closed-loop control capabilities and operational efficiency. Attached Figure Description

[0026] The accompanying drawings illustrate exemplary embodiments of the invention and, together with the description thereof, serve to explain the principles of the invention. These drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification.

[0027] Figure 1 This is a diagram of a digital twin system for manned special equipment in one embodiment of the present invention;

[0028] Figure 2This is a physical space example diagram of a digital twin system in one embodiment of the present invention;

[0029] Figure 3 This is a digital space example diagram of a digital twin system in one embodiment of the present invention;

[0030] Figure 4 This is an example diagram of the service space of a digital twin system in one embodiment of the present invention;

[0031] Figure 5 This is a flowchart of intelligent fault diagnosis of service space in one embodiment of the present invention;

[0032] Figure 6 This is a behavior analysis process for the service space in one embodiment of the present invention;

[0033] Figure 7 This is a behavior recognition model in one embodiment of the present invention;

[0034] Figure 8 This is a digital mainline framework in one embodiment of the present invention. Detailed Implementation

[0035] The following is in conjunction with the appendix Figures 1 to 8 The present invention will be further described in detail below with reference to the embodiments. It is to be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present invention are shown in the accompanying drawings.

[0036] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other. The technical solution of this invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0037] Unless otherwise stated, the exemplary embodiments / exemplifications shown are to be understood as providing exemplary features of various details that provide ways in which the technical concept of the invention can be implemented in practice. Therefore, unless otherwise stated, the features of the various embodiments / exemplifications may be additionally combined, separated, interchanged and / or rearranged without departing from the technical concept of the invention.

[0038] The use of crosshairs and / or shading in the accompanying drawings is generally used to clarify the boundaries between adjacent components. Thus, unless otherwise stated, the presence or absence of crosshairs or shading does not convey or indicate any preference or requirement for the specific material, material properties, dimensions, proportions, commonalities between the illustrated components, or any other characteristics, properties, etc., of the components. Furthermore, in the accompanying drawings, the dimensions and relative dimensions of components may be exaggerated for clarity and / or descriptive purposes. When exemplary embodiments can be implemented differently, a specific process sequence may be performed in a different order than that described. For example, two consecutively described processes may be performed substantially simultaneously or in the reverse order of their description. Furthermore, the same reference numerals denote the same components.

[0039] When a component is referred to as being "on" or "above" another component, "connected to," or "joined to" another component, the component may be directly on, directly connected to, or directly joined to the other component, or there may be intermediate components. However, when a component is referred to as being "directly on" another component, "directly connected to," or "directly joined to" another component, there are no intermediate components. Therefore, the term "connection" can refer to a physical connection, an electrical connection, etc., and may or may not have intermediate components.

[0040] For descriptive purposes, the present invention may use spatial relative terms such as “below,” “under,” “below,” “down,” “above,” “above,” “higher,” and “side (e.g., in a “sidewall”)” to describe the relationship between one component and another component as shown in the accompanying drawings. In addition to the orientations depicted in the drawings, the spatial relative terms are also intended to encompass different orientations of the device during use, operation, and / or manufacture. For example, if the device in the drawings is flipped, a component described as “below” or “under” another component or feature would subsequently be positioned “above” said other component or feature. Thus, the exemplary term “below” can encompass both “above” and “below” orientations. Furthermore, the device may be otherwise positioned (e.g., rotated 90 degrees or in other orientations), thus interpreting the spatial relative descriptive terms used herein accordingly.

[0041] The terminology used herein is for the purpose of describing particular embodiments and is not intended to be limiting. As used herein, unless the context clearly indicates otherwise, the singular forms “a” and “the” are intended to include the plural forms as well. Furthermore, when the terms “comprising” and / or “including” and variations thereof are used in this specification, it indicates the presence of the stated features, integrals, steps, operations, parts, components, and / or groups thereof, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, parts, components, and / or groups thereof. It should also be noted that, as used herein, the terms “substantially,” “about,” and other similar terms are used as approximate terms rather than as terms of degree, thus explaining the inherent biases in measurements, calculated values, and / or provided values ​​that would be recognized by one of ordinary skill in the art.

[0042] In one embodiment, the present invention provides a digital twin system for manned special equipment, comprising: a physical space, a digital space, a service space, and a digital mainline connecting the three.

[0043] The physical space is used to collect equipment operating status data and passenger behavior data; it includes the main body of the manned special equipment and a distributed sensor deployment unit.

[0044] The digital space is used to receive sensor data transmitted from the physical space and establish virtual mirror data of manned special equipment, including at least a fault diagnosis model reflecting the mechanical state of the equipment and an equipment maintenance model reflecting the behavior of passengers.

[0045] The service space is used to perform reasoning analysis based on the sensor data collected from the virtual image and physical space to generate control commands, including at least an intelligent fault diagnosis model for assessing the health status of the equipment and a behavior recognition model for identifying unsafe passenger behaviors.

[0046] The digital mainline is used to achieve real-time, bidirectional data synchronization and control command transmission between the physical space, digital space and service space.

[0047] Optionally, the distributed sensor deployment unit includes: vibration sensors, electrical and environmental sensors, displacement and motion sensors, hazard detection sensors, and vision sensors.

[0048] Optionally, the special equipment for carrying passengers is an escalator, a passenger elevator, or a passenger ropeway.

[0049] Optionally, the intelligent fault diagnosis model is implemented through the following process: multi-domain feature extraction of sensor signals; screening and optimization of the extracted features; and classification of the screened features using a machine learning model to diagnose equipment fault types.

[0050] Optionally, the behavior recognition model is implemented through the following process: extracting passenger posture key points from the video stream based on the posture detection model; extracting and fusing multi-dimensional behavior features from the key points; and classifying the fused features based on preset behavior recognition logic to determine the type of unsafe behavior.

[0051] Optionally, the pose detection model is a YOLOv11 model trained on a custom dataset.

[0052] Optionally, the unsafe behavior types include at least one of falling, squatting, walking backwards, crossing boundaries, and crowding.

[0053] Optionally, the service space also includes an operational fault model, which integrates data from multiple sensors in the distributed sensor system and assesses security threats to the device's operating environment and operational level based on rule threshold analysis and pattern analysis.

[0054] Optionally, the digital mainline includes: a maintenance thread for synchronizing historical maintenance data and real-time status data; an operation thread for synchronizing physical space sensor data to the digital space; and a control thread for transmitting control commands generated in the service space to the physical space.

[0055] In another embodiment, the present invention provides a digital twin system for manned special equipment, comprising: a physical space, a digital space, a service space, and a digital mainline; the digital mainline connects the physical space, the digital space, and the service space.

[0056] The physical space includes the manned special equipment body and a distributed sensor deployment unit; the distributed sensor deployment unit includes: vibration sensors, electrical and environmental sensors, displacement and motion sensors, hazard detection sensors and vision sensors.

[0057] The digital space is used to receive sensor data transmitted from the physical space and to establish virtual image data of manned special equipment, including: fault diagnosis model, equipment maintenance model and data management layer;

[0058] The fault diagnosis model detects and classifies anomalies based on real-time sensor data and historical operation records. This model supports fault location and provides key inputs for downstream control strategies.

[0059] The equipment maintenance model includes a crowd safety sub-model and a crowd judgment sub-model. The crowd safety sub-model is used to detect critical passenger safety incidents, such as falls. The crowd judgment sub-model is used to identify non-compliant passenger behaviors, such as pushing strollers, walking backwards, and sitting on steps.

[0060] The data management layer is responsible for ingesting data from digital threads in real time, as well as data cleaning, tagging, and long-term storage.

[0061] The service space relies on real-time sensor data from the physical space and virtual mirror data from the digital space to perform reasoning analysis and generate control commands, including: intelligent fault diagnosis model, behavior recognition model and operation fault model;

[0062] The intelligent fault diagnosis model uses a mechanical fault diagnosis method based on multi-source feature extraction and machine learning to assess the health status of equipment.

[0063] The behavior recognition model identifies unsafe behaviors of passengers based on the posture detection model. The types of unsafe behaviors identified include at least one of falling, squatting, walking backwards, crossing boundaries, and crowding.

[0064] The digital mainline is used to realize data exchange and control command transmission between physical space, digital space and service space, including: maintenance thread, operation thread, control thread and communication port;

[0065] The maintenance thread is used to continuously collect historical maintenance records and synchronize them with the real-time operating status of the equipment, providing a basis for fault trend analysis and predictive planning;

[0066] The operation thread is used to acquire sensor data from the physical space and synchronize it in real time with a virtual mirror in the digital space;

[0067] The control thread is used to transmit control commands generated in the service space back to the physical space;

[0068] The communication port provides an interface for external systems and human-computer interaction.

[0069] Furthermore, special equipment for carrying passengers includes escalators, passenger elevators, or passenger ropeways.

[0070] Furthermore, vibration sensors are installed on the motors, gearboxes, drive wheels, and bearing systems of the manned special equipment; electrical and environmental sensors include motor current sensors, ambient temperature sensors, noise sensors, and handrail temperature sensors; displacement and motion sensors include braking distance sensors and step chain displacement detectors; hazard detection sensors include water level indicators and tension chain abnormal noise detectors; and visual sensors are high-definition surveillance cameras installed at the entrances / exits and the middle side of the manned special equipment.

[0071] Furthermore, the multi-domain feature extraction of the intelligent fault diagnosis model includes: extracting frequency domain features through Fast Fourier Transform, extracting envelope spectrum features through Hilbert Transform, and generating time-frequency features through continuous wavelet transform, discrete wavelet transform, or wavelet-based singular value decomposition; feature selection uses the Random Forest algorithm to rank the extracted features by importance and select the minimum feature set; machine learning classification uses the XGBoost model for high-precision classification of fault types.

[0072] Furthermore, the posture detection model of the behavior recognition model is the YOLOv11 model, which is trained on a customized labeled dataset and used to detect passenger posture key points in real time. The behavior recognition model also includes a feature extraction module, which is used to extract structured features from human body key points and integrate posture geometry analysis, motion distance evaluation and statistical boundary modeling techniques to complete multi-dimensional feature fusion.

[0073] Furthermore, the behavior recognition model identifies unsafe behaviors in the following ways: falling or squatting behaviors are determined by calculating the vertical distance between the head and knee key points or the shoulder and knee key points; walking backwards or wandering behaviors are identified by comparing the displacement of key points between frames with the motion threshold; and out-of-bounds behaviors are determined by judging whether the coordinates of the head key point or the elbow key point exceed the range of the average plus or minus 2 standard deviations.

[0074] Furthermore, the behavior recognition model also includes a model optimization line, which combines a minimum classification error strategy, uses historical behavior data and labeled samples to optimize the classification threshold, and calculates the classification error to correct the parameters of feature extraction, thereby achieving iterative optimization of the model.

[0075] Furthermore, the operational fault model integrates data from machine room water level sensors, handrail temperature sensors, motor current sensors, ladder chain displacement sensors, and noise sensors. It employs a combination of rule-based threshold analysis and pattern analysis to assess potential safety threats such as chain derailment, belt slippage, equipment overheating, or water ingress.

[0076] Furthermore, the maintenance thread is used to collect historical maintenance records and synchronize them with the real-time operating status of the equipment, providing a basis for fault trend analysis and predictive maintenance planning; the control thread supports fault mitigation, operating mode switching, emergency intervention, and remote reset operations; and the communication port provides an interface for external systems and human-machine interaction.

[0077] Furthermore, the physical space also includes an embedded control unit and a semi-intelligent monitoring interface. The embedded control unit is used to receive control commands from the service space and execute physical actions, while the semi-intelligent monitoring interface is used for maintenance personnel to view operating status, alarm information, and diagnostic results.

[0078] In another embodiment, the present invention provides a digital twin system for manned special equipment, comprising: a physical space, a digital space, a service space, and a digital mainline; the digital mainline connects the physical space, the digital space, and the service space.

[0079] The physical space includes the manned special equipment body, a distributed sensor deployment unit, an embedded control unit, and a semi-intelligent monitoring interface used by maintenance personnel. The distributed sensor deployment unit includes: vibration sensors, electrical and environmental sensors, displacement and motion sensors, hazard detection sensors, and vision sensors.

[0080] Vibration sensors are installed on the motors, gearboxes, drive wheels, and bearing systems of the manned special equipment; electrical and environmental sensors include motor current sensors, ambient temperature sensors, noise sensors, and handrail temperature sensors; displacement and motion sensors include braking distance sensors and step chain displacement detectors; hazard detection sensors include water level indicators and tension chain abnormal noise detectors; and visual sensors are high-definition surveillance cameras installed at the entrances / exits and the middle side of the manned special equipment.

[0081] The embedded control unit is used to receive control commands from the service space and execute physical actions, while the semi-intelligent monitoring interface is used for maintenance personnel to view the operating status, alarm information, and diagnostic results.

[0082] The digital space is used to receive sensor data transmitted from the physical space and to establish virtual image data of manned special equipment, including: fault diagnosis model, equipment maintenance model and data management layer;

[0083] The fault diagnosis model detects and classifies anomalies based on real-time sensor data and historical operating records. This model supports fault location and provides key inputs for downstream control strategies. The equipment maintenance model includes a crowd safety sub-model and a crowd judgment sub-model. The crowd safety sub-model is used to detect critical passenger safety incidents, such as falls. The crowd judgment sub-model is used to identify non-compliant passenger behaviors, such as pushing strollers, walking backwards, and sitting on steps.

[0084] The data management layer is responsible for ingesting data from digital threads in real time, as well as data cleaning, tagging, and long-term storage. This component also supports digital-physical synchronization, visual state representation, and intuitive model interpretation to aid maintenance decisions.

[0085] The service space relies on real-time sensor data from the physical space and virtual mirror data from the digital space to perform reasoning analysis and generate control commands, including: intelligent fault diagnosis model, behavior recognition model and operation fault model;

[0086] The intelligent fault diagnosis model uses a mechanical fault diagnosis method based on multi-source feature extraction and machine learning to assess the health status of equipment.

[0087] The mechanical fault diagnosis method includes:

[0088] a) Data acquisition and transmission: Sensor location information and real-time signal values ​​are acquired through the digital mainline, achieving seamless integration of data acquisition and transmission, and providing basic data support for subsequent fault diagnosis;

[0089] b) Multi-source feature extraction: Multi-dimensional feature extraction is performed on real-time signals. Frequency domain features are obtained using Fast Fourier Transform to reveal the spectral energy distribution characteristics. To capture localized and non-steady-state mode features, continuous wavelet transform, discrete wavelet transform, and wavelet-based singular value decomposition are used to generate time-frequency features. At the same time, envelope spectral features are extracted using Hilbert transform. This method can effectively identify weak pulse components related to early faults.

[0090] c) Feature selection optimization: Random forest algorithm is used to rank the importance of the extracted features, then the top-ranked feature subsets are evaluated iteratively and their classification accuracy is compared with the full feature model. Finally, the smallest feature set that can achieve the highest or similar accuracy is selected to reduce the computational complexity of the model.

[0091] d) Model Training and Anomaly Detection: The XGBoost model is used for high-precision classification. To accurately locate faults, the training dataset of the XGBoost model is expanded by integrating historical alarm data and fault data from other units in the same system. At the same time, input features covering the time domain, frequency domain, time-frequency domain, and envelope domain are further selected according to the feature importance ranking of the original high-dimensional dataset, and finally a complete fault diagnosis model is constructed. The trained model is applied to anomaly detection, which can support the detection and classification of linear anomalies and has high recognition accuracy.

[0092] This fault diagnosis model undertakes the function of system health status assessment. Through the entire process of multi-source feature extraction, feature screening and machine learning classification, it realizes accurate diagnosis of mechanical system faults and health status monitoring.

[0093] The behavior recognition model identifies unsafe behaviors of passengers based on the posture detection model. The types of unsafe behaviors identified include at least one of falling, squatting, walking backwards, crossing boundaries, and crowding.

[0094] The identification of unsafe passenger behaviors based on the posture detection model includes:

[0095] 1. A YOLOv11 model trained on a custom labeled dataset is used to detect passenger posture key points in real time and generate posture data;

[0096] 2. The structured features are extracted from the key points of the human body, and after integrating posture geometry analysis, motion distance assessment and statistical boundary modeling techniques to complete multi-dimensional feature fusion, the results are output to the classification output module.

[0097] 3. The classification output module is based on the identification logic of unsafe behaviors, and can accurately classify five types of unsafe behaviors: falling, squatting, walking backwards, crossing boundaries and overcrowding.

[0098] The logic for identifying unsafe behaviors includes:

[0099] Falling or squatting behavior: determined by calculating the vertical distance between the head and the knee or between the shoulder and the knee.

[0100] Backward walking or wandering behavior: identified by comparing the displacement of key points between frames with a motion threshold;

[0101] Out-of-boundary behavior: This is determined by judging whether the coordinates of the head key points or the elbow key points exceed the range of the average plus or minus 2 standard deviations.

[0102] The specific calculation method uses the typical behavior of "squatting" as an example, introducing the head-knee distance as a key discrimination indicator. This distance is calculated in three-dimensional space using the Euclidean distance formula:

[0103]

[0104] in, The three-dimensional coordinates of the key points in the head. The x-coordinate of the key points in the head. The vertical coordinate of the key points in the head. The vertical coordinates of the key points in the head; The three-dimensional coordinates of key points of the knee joint. The x-coordinate of the key points of the knee joint The vertical coordinates of key points of the knee joint are shown. The vertical coordinates are for key points of the knee joint.

[0105] In two-dimensional images, this is achieved through scaling factors. The estimation is performed using the following formula:

[0106]

[0107] in The thresholds were determined by calibrating the fixed escalator step heights against the pixel heights of the steps in the images. A minimum classification error strategy was employed to determine the thresholds. Through analysis of 100,000 images, an appropriate scaling factor was selected, resulting in an accuracy rate exceeding 93% for identifying all five types of unsafe behaviors, meeting the engineering application requirements for real-time monitoring.

[0108] The classification results are used for behavior judgment, such as Figure 7 As shown; on the other hand, the input crowd safety sub-model provides data support for crowd judgment, and finally the crowd judgment sub-module combines the behavioral risk level to give equipment control suggestions and crowd guidance suggestions.

[0109] The behavior recognition model also includes a model optimization module. Specifically, the video stream is simultaneously input into the model optimization module. The model optimization module combines a minimum classification error strategy, uses historical behavior data and labeled samples to optimize the classification threshold, and interacts with the loss function in real time. By calculating the classification error, the parameters of the feature extraction stage are corrected in reverse. This optimization information is continuously fed back to the posture assessment and feature extraction stages, enabling the model to iteratively optimize in a closed loop. This further enhances the accuracy of posture assessment and the relevance of feature extraction, ensuring the robustness and real-time classification capability of the customized behavior judgment algorithm in complex crowd environments.

[0110] Taking escalators as an example, the iterative training involves dividing the labeled dataset into training, validation, and test sets in a 7:2:1 ratio, setting training parameters, employing stochastic gradient descent (SGD) optimization, calculating the bounding box loss using the CIoU loss function, and calculating the classification loss using the cross-entropy loss function. The behavioral logic embedding involves embedding escalator scene behavior judgment logic during training. For example, for "walking backwards," the judgment threshold is adjusted based on the escalator's running direction; for "crossing boundaries," the escalator boundary coordinate data from a virtual mirror is imported to calibrate the key point boundary crossing judgment criteria. The accuracy optimization involves analyzing the recognition accuracy of various behaviors using a confusion matrix. For weakly recognized behaviors such as "squatting" and "overcrowding," additional samples are added, and a hard example mining strategy is used to strengthen training. After optimization, the model achieves an accuracy rate exceeding 93% for all five unsafe behaviors, meeting real-time monitoring requirements.

[0111] The operational fault model integrates data from various physical space sensors, including: machine room water level sensors for detecting flood risk; handrail temperature sensors for monitoring abnormal heating; motor current sensors for detecting overload or electrical imbalance; ladder chain displacement sensors for identifying potential misalignment or mechanical loosening issues; and noise sensors for detecting unexpected acoustic anomalies indicating friction or mechanical wear. By analyzing these sensor signals in real time and running the fault model, it supports the detection of various types of faults, excluding mechanical vibration. The collected data is fused using a combination of rule-based threshold analysis and pattern analysis to assess potential safety threats such as chain derailment, belt slippage, equipment overheating, and water ingress.

[0112] The digital mainline is used to realize data exchange and control command transmission between physical space, digital space and service space, including: maintenance thread, operation thread, control thread and communication port;

[0113] The maintenance thread is used to continuously collect historical maintenance records and synchronize them with the real-time operating status of the equipment, providing a basis for fault trend analysis and predictive planning;

[0114] The operation thread is used to acquire sensor data from the physical space and synchronize it in real time with a virtual mirror in the digital space;

[0115] The control thread is used to transmit control commands generated in the service space back to the physical space;

[0116] The communication port provides an interface for external systems and human-computer interaction.

[0117] In another embodiment, the present invention provides a digital twin method for manned special equipment, the method comprising:

[0118] S1: Real-time collection of operational data and passenger behavior data of manned special equipment through distributed sensors deployed on the physical body of the manned special equipment;

[0119] S2: The collected data is transmitted to the digital space through the operation thread of the digital main line. A virtual image of the manned special equipment is established in the digital space, and the transmitted data is preprocessed and stored.

[0120] S3: Based on the real-time sensing data from step S1 and the virtual mirror data from step S2, perform inference analysis in the service space to generate control commands.

[0121] In another embodiment, the present invention provides a digital twin method for manned special equipment, comprising the following steps:

[0122] S1: Data Acquisition: Real-time acquisition of operational data and passenger behavior data of manned special equipment through distributed sensors deployed on the physical body of the manned special equipment;

[0123] The distributed sensors include: vibration sensors, electrical and environmental sensors, displacement and motion sensors, hazard detection sensors, and vision sensors;

[0124] Operational data includes vibration data, electrical data, environmental data, displacement data, and hazard detection data; passenger behavior data consists of video streams collected by visual sensors.

[0125] S2: Virtual Image Construction and Data Management: The collected data is transmitted to the digital space through the operation thread of the digital mainline. A virtual image of the manned special equipment is built in the digital space, and the transmitted data is preprocessed and stored.

[0126] S3: Intelligent Service Reasoning and Decision-Making: Based on the real-time sensor data from step S1 and the virtual mirror data from step S2, reasoning analysis is performed in the service space to generate control commands.

[0127] S31: Based on multi-source feature extraction and machine learning algorithms, perform equipment health status assessment and intelligent fault diagnosis;

[0128] In step S31, the intelligent fault diagnosis includes:

[0129] a) Multi-source feature extraction: Frequency domain features are extracted by fast Fourier transform, envelope spectrum features are extracted by Hilbert transform, and time-frequency features are generated by continuous wavelet transform, discrete wavelet transform, or wavelet-based singular value decomposition.

[0130] b) Feature selection optimization: Random forest algorithm is used to rank the extracted features by importance and select the minimum feature set;

[0131] c) Machine learning classification: The XGBoost model is used for high-precision classification of fault types.

[0132] S32: Identify unsafe passenger behaviors based on posture detection models;

[0133] The identification of unsafe passenger behaviors based on the posture detection model includes:

[0134] 1. A YOLOv11 model trained on a custom labeled dataset is used to detect passenger posture key points in real time and generate posture data;

[0135] 2. The structured features are extracted from the key points of the human body, and after integrating posture geometry analysis, motion distance assessment and statistical boundary modeling techniques to complete multi-dimensional feature fusion, the results are output to the classification output module.

[0136] 3. The classification output module is based on the identification logic of unsafe behaviors, and can accurately classify five types of unsafe behaviors: falling, squatting, walking backwards, crossing boundaries and overcrowding.

[0137] The logic for identifying unsafe behaviors includes:

[0138] Falling or squatting behavior: determined by calculating the vertical distance between the head and the knee or between the shoulder and the knee.

[0139] Backward walking or wandering behavior: identified by comparing the displacement of key points between frames with a motion threshold;

[0140] Out-of-boundary behavior: This is determined by judging whether the coordinates of the head key points or the elbow key points exceed the range of the average plus or minus 2 standard deviations.

[0141] S33: Assess security threats at the environmental and operational levels based on rule-based threshold analysis and pattern analysis;

[0142] S4: Data Synchronization and Control Command Execution: Data exchange and control command transmission between steps S1, S2, and S3 are achieved through the digital mainline;

[0143] Specifically, it includes:

[0144] S41: Synchronize the sensor data collected in step S1 to step S2 in real time through the operation thread;

[0145] S42: Collect historical maintenance records through the maintenance thread and synchronize them with the real-time operating status of the equipment to provide a basis for fault trend analysis;

[0146] S43: The control commands generated in step S3 are transmitted back to the device control unit in the physical space via the control thread to perform fault mitigation, operating mode switching, emergency intervention or remote reset operations.

[0147] In another embodiment, an escalator is used as a prototype. The core features of an escalator are step flow, vertical zoning, and the linkage between passenger behavior and steps. Compared to a box elevator, passenger behavior is relatively limited, lacking the issue of a flowing carrier. Analyzing the key aspects of the algorithm, target detection needs to consider motion compensation, behavior recognition is related to the direction of step flow, and spatial constraints require dynamic adaptation of the detection area. The overall architecture consists of four interconnected spaces: physical space, digital space, service space, and a digital mainline connecting the three, such as... Figure 1 As shown. These three parts work together to form a complete closed-loop system that senses the physical world, maps and analyzes it in the digital world, makes decisions through intelligent services, and ultimately feeds back to control the physical world. Each unit works together to achieve passenger behavior recognition and safety warnings. The specific structure and functions are as follows:

[0148] The physical space, serving as the foundational layer of the entire system, includes escalator hardware entities, distributed sensor deployment units, embedded control systems, and a semi-intelligent monitoring interface used by maintenance personnel. Figure 2 As shown. The structure of an escalator includes basic mechanical components such as the main drive unit, step chain, tensioning assembly, handrail mechanism, and step model. To achieve real-time status awareness, the system is equipped with various sensors and interfaces on the physical body of the special equipment and its surrounding environment:

[0149] a) Vibration sensors – installed in motors, gearboxes, drive wheels, and bearing systems to capture dynamic mechanical characteristics;

[0150] b) Electrical and environmental sensors – including motor current sensors, ambient temperature and noise sensors, and handrail temperature sensors;

[0151] c) Displacement and motion sensors – including braking distance sensors and step chain displacement detectors;

[0152] d) Hazard detection sensors – such as water level indicators and abnormal noise detectors for tensioned chains;

[0153] e) Visual sensors – High-definition surveillance cameras are installed at key locations such as escalator entrances and exits, and the sides of the middle steps, enabling real-time acquisition of video data on passenger posture and movement trajectory;

[0154] f) Semi-automatic visual terminal – for maintenance personnel to view operating status, alarm information and diagnostic results;

[0155] g) Embedded control unit – receives high-level instructions from the service space and executes real-time physical actions.

[0156] These components together form a complete system-level perception system, generating high-precision operational data, inputting the data into the digital space, and providing support for intelligent decision-making at the service layer.

[0157] The digital space is the core hub for system analysis and decision-making. It forms a crucial bridge between physical hardware and algorithm services, providing a platform for full lifecycle safety management and the intelligent evolution of escalators. For example... Figure 3 As shown, it is divided into the following three modules.

[0158] a) Fault diagnosis model: This model uses real-time sensor data and historical operation records to detect and classify abnormal situations. It supports fault location and provides key inputs for downstream control strategies.

[0159] b) Equipment maintenance model, which integrates two behavior-oriented sub-models:

[0160] Crowd safety model: Detects critical passenger safety incidents, such as falls. Once a fall is detected, it triggers an emergency response protocol to prevent secondary injuries.

[0161] Crowd identification models: Identify non-compliant passenger behaviors—including pushing strollers, walking backwards, and sitting on steps. These behaviors may not pose an immediate danger but increase system risk. The system issues early warnings or adaptive suggestions to guide passengers towards safer behavior. These models collectively enhance the human-computer interaction layer, enabling proactive risk detection based on pose estimation and video analytics.

[0162] c) Data Management and Visualization Layer: Responsible for real-time data ingestion from digital threads, as well as data cleaning, tagging, and long-term storage. This component also supports digital-physical synchronization, visual state representation, and intuitive model interpretation to aid maintenance decisions.

[0163] Overall, the main goal of digital space is to maintain a dynamic, real-time, and highly accurate representation of escalator systems, thereby enabling status monitoring, fault detection, risk assessment, and intelligent control.

[0164] The service space relies on real-time sensor data collected from the physical space and virtual mirror data generated in the digital space to conduct multi-dimensional reasoning analysis, and uses digital threads to issue control commands to the physical equipment control system. This space integrates three frameworks: an intelligent fault diagnosis model, and escalator maintenance and operational fault diagnosis. Figure 4 As shown, each component has a clearly defined function.

[0165] a) Fault diagnosis model:

[0166] This model is responsible for system health status assessment and is based on a mechanical fault diagnosis method using multi-source feature extraction and machine learning: multi-source signal features are extracted, and FFT and Hilbert are used to obtain frequency domain and time-frequency domain features; the optimal feature subset is selected using the random forest algorithm; finally, the XGBoost model is used for high-precision classification. Figure 5As shown, a) Data acquisition and transmission: Sensor location information and real-time signal values ​​are acquired through a digital mainline, achieving seamless data acquisition and transmission, providing basic data support for subsequent fault diagnosis. b) Multi-source feature extraction: Multi-dimensional feature extraction is performed on real-time signals. Frequency domain features are obtained using Fast Fourier Transform to reveal spectral energy distribution characteristics. To capture localized and non-steady-state mode features, continuous wavelet transform, discrete wavelet transform, and wavelet-based singular value decomposition are used to generate time-frequency features. Simultaneously, envelope spectral features are extracted using Hilbert transform, which can effectively identify weak pulse components related to early faults. c) Feature selection and optimization: The random forest algorithm is used to rank the importance of extracted features. Subsequently, the top-ranked feature subsets are iteratively evaluated, and their classification accuracy is compared with the full feature model. Finally, the smallest feature set that achieves the highest or similar accuracy is selected, reducing the computational complexity of the model. d) Model training and anomaly detection: After data extraction and feature selection, the fault diagnosis model is trained. XGBoost ensemble learning algorithm was selected as the basic model. This model has strong fitting ability for nonlinear features and anti-overfitting characteristics. (1) Initial parameter setting (2) Iterative training: Input the selected feature dataset into the model, adopt batch gradient descent, optimization strategy, and calculate the prediction error with cross-entropy loss function. Real-time monitoring of the accuracy of the validation set during training (3) Twin data linkage: Real-time simulation data of digital twin virtual mirror was connected during training. A batch of virtual fault samples was inserted every 10 rounds of iteration for reinforcement training, so that the model learns the law of fault evolution rather than single fault features. The trained model was applied to anomaly detection, which can support the detection and classification of linear anomalies and has high recognition accuracy. Experimental results show that the classification performance of this method is excellent in four states. Its precision scores are 0.97, 0.99, 0.99 and 0.94, respectively, and the recall rates are 0.99, 0.99, 0.94 and 0.96, respectively. The overall classification accuracy reached 0.97, which fully proves that the proposed diagnostic method has good robustness and reliability in real escalator fault scenarios. The overall algorithm flow is as follows. Figure 7 As shown in Figure e), the core function of the model is to assess the health status of the system. Through the entire process of multi-source feature extraction, feature selection, and machine learning classification, the model can accurately diagnose mechanical system faults and monitor their health status.

[0167] b) Behavior recognition model:

[0168] Focusing on predicting unsafe pedestrian behaviors, this study employs a YOLOv11 model, trained on a customized labeled dataset, to achieve real-time detection of key points in passenger posture, such as... Figure 6 As shown. This process revolves around escalator maintenance and is divided into two main lines: behavior analysis and model optimization.

[0169] Behavior Analysis Line: The video stream first enters the posture assessment stage, generating posture data based on human keypoint detection technology. Then, it enters the feature extraction stage, extracting structured features from human keypoints and integrating posture geometry analysis, motion distance assessment, and statistical boundary modeling techniques to complete multi-dimensional feature fusion before outputting to the classification output module. The classification output module, based on domain-adaptive classification logic tailored for escalator scenarios, achieves accurate classification of five unsafe behaviors: falling, squatting, walking backwards, crossing boundaries, and overcrowding. The classification results are used for behavior judgment, such as... Figure 7 As shown, falling and squatting are determined by calculating the vertical distance between the head and knee and shoulder and knee joints, walking backwards / wandering is identified by comparing the displacement of key points between frames with the motion threshold, and crossing the boundary is determined by whether the head / elbow coordinates exceed the average value ±2σ range; on the other hand, the input to the crowd safety module provides data support for crowd judgment, and finally the crowd judgment module combines the behavioral risk level to give equipment control suggestions and crowd guidance suggestions.

[0170] Model optimization process: The video stream is simultaneously input into the retraining module. The retraining module combines the minimum classification error strategy, uses historical behavior data and labeled samples to optimize the classification threshold, and interacts with the loss function in real time. By calculating the classification error, the parameters of the feature extraction stage are corrected in reverse. This optimization information is continuously fed back to the pose assessment and feature extraction stages, making the model iterative optimization closed loop. This further enhances the accuracy of pose assessment and the relevance of feature extraction, ensuring the robustness and real-time classification capability of the customized behavior judgment algorithm in complex crowd environments.

[0171] The model as a whole achieves intelligent analysis and continuous iteration of the maintenance model for crowd behavior in escalator scenarios through the collaboration of multiple modules.

[0172] c) Operational failure model:

[0173] In addition to fault diagnosis and crowd-related safety assessments, the service space includes an operational fault model designed to identify a range of equipment-level or environmental anomalies that could compromise escalator safety. This model integrates data from various physical space sensors, including: machine room water level sensors to detect flood risk; handrail temperature sensors to monitor abnormal heating; motor current sensors to detect overload or electrical imbalances; step chain displacement sensors to identify potential misalignment or mechanical loosening; and noise sensors to detect unexpected acoustic anomalies indicating friction or mechanical wear. By analyzing these sensor signals in real time and running the fault model, it supports the detection of various fault types beyond mechanical vibration. The collected data is fused using a combination of rule-based threshold analysis and pattern analysis to assess potential safety threats such as chain derailment, belt slippage, equipment overheating, and water ingress.

[0174] The digital mainline, as the core data channel of the DT framework, enables continuous, high-frequency data exchange between the physical space, digital space, and service space. This ensures consistency and traceability of operations throughout the entire system lifecycle and supports seamless integration of sensing, analysis, and control. Figure 8 As shown, it is divided into the following four modules:

[0175] a) Maintenance thread: Continuously collects historical maintenance records and synchronizes them with the real-time operating status of the equipment to provide a basis for fault trend analysis and predictive planning.

[0176] b) Operation thread: Acquire sensor data from the physical space, including vibration, temperature, noise, water level, and monitoring video, and synchronize this data with virtual entities in the digital space in real time.

[0177] c) Control Thread: Transmits decision outputs and control commands from service space back to physical space. This thread supports the following operations: fault mitigation, operating mode switching, emergency intervention, and remote reset commands.

[0178] d) Communication port: Provides an interface for external systems and human-computer interaction.

[0179] Compared with the prior art, the present invention has the following significant advantages:

[0180] 1. Strong applicability to various scenarios: It can dynamically adjust the recognition logic according to the spatial constraints of escalator step flow, and its adaptability is better than that of general monitoring solutions.

[0181] 2. Enhance the comprehensive monitoring efficiency and intelligent operation and maintenance level of manned special equipment: The digital twin framework proposed in this invention integrates a fault diagnosis module and an unsafe behavior monitoring module, which can simultaneously and accurately identify mechanical faults and unsafe behaviors of passengers, effectively filling the gap in the separate monitoring of equipment faults and human risks, and improving the reliability of operation and maintenance equipment.

[0182] 3. Highly efficient data linkage: Real-time linkage of multi-source data is achieved by relying on digital thread technology: Passenger behavior video streams from visual sensors, equipment operation data from vibration sensors, and water level sensor data can be synchronously transmitted to the processing unit through the operation thread. Equipment status data can be called up in real time to check whether it is caused by equipment failure and avoid misjudgment based on single data.

[0183] 4. Safety Management Closed Loop: Sensors collect behavioral and equipment data in real time. After the model identifies unsafe behaviors or equipment hazards, it outputs a tiered response. At the same time, the warning results are pushed to the operation and maintenance terminal. After the operation and maintenance personnel handle the situation, they can mark the results in the system. The data is synchronously transmitted back to the database for subsequent model optimization, realizing full-process traceability and iterative capability.

[0184] 5. Strong scalability and portability: In terms of scalability, it supports adding new functional modules without reconstructing the core architecture; in terms of portability, when migrating from escalators to passenger elevators, only the installation parameters of the vision sensors and the behavior judgment threshold need to be adjusted, without replacing the hardware.

[0185] This technology is transferable, as shown in the following ways:

[0186] The design concept of this technology is modular design, in which the physical entities and virtual models in the digital twin architecture can be replaced with corresponding special equipment entities and virtual models such as passenger elevators, passenger ropeways, and amusement facilities, covering both personnel and equipment safety monitoring and maintenance scenarios.

[0187] This digital twin technology adopts a layered architecture, divided into physical space, digital space, and service space, all connected by a digital mainline. The physical space is responsible for data acquisition, the digital space for feature extraction and analysis, the digital mainline for data transmission, and the service space for generating control commands and human-machine interaction. This architecture is not bound to the specific structure of any manned equipment. Whether it's escalator step data, elevator car data, or cable car cabin data, only the scene configuration parameters of each layer need to be adjusted to adapt to new equipment.

[0188] The core algorithm of this technology is also transferable:

[0189] Behavior recognition algorithm: Its classification logic for unsafe behaviors such as boundary crossing and abnormal postures only needs to adjust the scene boundary parameters, such as providing step boundary parameters for escalators, car door boundary parameters for elevators, and cabin guardrail boundary parameters for cableways, to adapt it;

[0190] Fault diagnosis algorithm: The algorithm uses FFT and wavelet transform for feature extraction and XGBoost for classification. This logic is universal. It is only necessary to adjust the algorithm threshold according to the corresponding fault parameters of the new equipment (such as the vibration threshold of the elevator traction machine and the tension threshold of the cableway). There is no need to redevelop the model. The diagnostic model can be optimized by comparing virtual and real data to achieve accurate identification of faults in new equipment.

[0191] This technology employs a universal virtual-real interaction interface for closed-loop control:

[0192] The closed-loop control link of the digital twin is designed with a general industrial control protocol. During migration, it is only necessary to match the execution instruction protocol of the new device in the virtual image to achieve closed-loop linkage, without relying on the specific control logic of the device.

[0193] In the description of this specification, the references to terms such as "one embodiment / mode," "some embodiments / modes," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment / mode or example is included in at least one embodiment / mode or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment / mode or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments / modes or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments / modes or examples described in this specification, as well as the features of different embodiments / modes or examples.

[0194] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0195] Those skilled in the art should understand that the above embodiments are merely for illustrating the present invention and are not intended to limit the scope of the invention. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of the present invention.

Claims

1. A digital twin system for manned special equipment, characterized in that, include: Physical space, digital space, service space, and the digital thread connecting the three; The physical space is used to collect equipment operating status data and passenger behavior data; Includes the main body of manned special equipment and distributed sensor deployment units; The digital space is used to receive sensor data transmitted from the physical space and establish virtual mirror data of manned special equipment, including at least a fault diagnosis model reflecting the mechanical state of the equipment and an equipment maintenance model reflecting the behavior of passengers. The service space is used to perform reasoning analysis based on the sensor data collected from the virtual image and physical space to generate control commands, including at least an intelligent fault diagnosis model for assessing the health status of the equipment and a behavior recognition model for identifying unsafe passenger behaviors. The digital mainline is used to achieve real-time, bidirectional data synchronization and control command transmission between the physical space, digital space and service space.

2. The system according to claim 1, characterized in that, Preferably, the distributed sensor deployment unit includes: vibration sensors, electrical and environmental sensors, displacement and motion sensors, hazard detection sensors, and vision sensors.

3. The system according to claim 2, characterized in that, The special equipment for carrying passengers is an escalator, a passenger elevator, or a passenger ropeway.

4. The system according to claim 1, characterized in that, The intelligent fault diagnosis model is implemented through the following process: multi-domain feature extraction of sensor signals; screening and optimization of the extracted features; and classification of the screened features using a machine learning model to diagnose equipment fault types.

5. The system according to claim 1 or 4, characterized in that, The behavior recognition model is implemented through the following process: extracting passenger posture key points from the video stream based on the posture detection model; extracting and fusing multi-dimensional behavioral features from the key points; The fused features are classified based on a pre-defined behavior recognition logic to determine the type of unsafe behavior.

6. The system according to claim 5, characterized in that, The pose detection model is a YOLOv11 model trained on a custom dataset.

7. The system according to claim 5, characterized in that, The unsafe behavior types include at least one of falling, squatting, walking backwards, crossing boundaries, and crowding.

8. The system according to claim 1, characterized in that, The service space also includes an operational fault model, which integrates data from multiple sensors in the distributed sensor system and assesses security threats to the device's operating environment and operational level based on rule threshold analysis and pattern analysis.

9. The system according to claim 1, characterized in that, The digital mainline includes: a maintenance thread for synchronizing historical maintenance data and real-time status data; an operation thread for synchronizing physical space sensor data to the digital space; and a control thread for transmitting control commands generated in the service space to the physical space.

10. A digital twin method for manned special equipment, characterized in that, The method includes: S1: Real-time collection of operational data and passenger behavior data of manned special equipment through distributed sensors deployed on the physical body of the manned special equipment; S2: The collected data is transmitted to the digital space through the operation thread of the digital main line. A virtual image of the manned special equipment is established in the digital space, and the transmitted data is preprocessed and stored. S3: Based on the real-time sensing data from step S1 and the virtual mirror data from step S2, perform inference analysis in the service space to generate control commands.

Images