Web-based digital twin framework

Abstract

A web-based digital twin framework that implements a digital twin in a virtual space based on the state, events, and physical behavior of process equipment includes a authoring layer that defines a data structure including state information, event information, physical constraint information, and synchronization information based on data collected from real-world process equipment, and generates an execution package including a visualization model, a physical model, an event graph, a state machine, and synchronization metadata; an event-based simulation engine that receives the execution package as input, manages state transitions following the occurrence of events, and selectively performs only physical operations corresponding to states activated by events; and a real-time synchronization and correction engine that aligns the predicted state generated by the simulation engine with the actual measurement data received from the real-world process equipment on a time basis, calculates the error between the two, and selectively corrects parts of the state based on the error. The event-based simulation engine reduces the computational load by limiting the states subject to computation based on event occurrences and excluding physical operations for inactive states, and the real-time synchronization and correction engine corrects only the partial states corresponding to the states where errors occurred without updating the entire state.

Description

Technology Field

[0001] The present invention relates to a framework for building a process-specific digital twin in which actual process equipment and a virtual space are organically linked, overcoming the limitations of the simple 3D visualization (Viewer) function provided by existing manufacturing process monitoring systems. More specifically, it relates to a web-based digital twin framework that integrally provides a Twin Authoring Layer that defines processes, assets, physical actions, and events, an Event-driven Simulation Engine that reflects and simulates the state in real time in a web browser, etc., and a Real-time Sync & Correction Engine that accurately matches the data of the actual site with the virtual state. Background Technology

[0003] Recently, the adoption of digital twin technology is expanding in the manufacturing industry to improve process efficiency and enhance the precision of quality control. Digital twins are a technology that replicates real-world equipment or processes in a virtual space to perform simulations, monitoring, and predictions, playing a key role in smart factories and automated processes.

[0004] However, most existing digital twin systems are limited to simply visualizing the shape and state of equipment, resulting in restricted dynamic simulation capabilities based on events and state transitions occurring in the actual process. In particular, when using a structure that calculates the physical movements of all equipment collectively, the computational load increases excessively, making real-time processing in web-based environments difficult.

[0005] Furthermore, existing technologies lack a structure to effectively manage the difference between simulation results generated in a virtual space and sensor data collected from the actual process, leading to the accumulation of discrepancies between the virtual and actual states over time. This issue can degrade the reliability of digital twins and limit the utilization of real-time monitoring and control functions. Prior art literature

[0007] Republic of Korea Registered Patent No. 10-25628288 Heat treatment process system based on digital twin / virtual physical system (July 28, 2023) The problem to be solved

[0008] Therefore, the objective of the present invention is to provide a digital twin framework that can maintain consistency with the actual process while ensuring real-time performance even in a web environment, through a structure that selectively performs physical calculations based on process state-based event processing and corrects only partial states based on the error between the predicted state and actual measured data. means of solving the problem

[0010] A web-based digital twin framework according to the present invention for implementing a digital twin in a virtual space based on the state, events, and physical operations of process equipment to achieve the above objective comprises: an authoring layer that defines a data structure including state information, event information, physical constraint information, and synchronization information based on data collected from real process equipment, and generates an execution package including a visualization model, a physical model, an event graph, a state machine, and synchronization metadata; an event-based simulation engine that receives the execution package, manages state transitions according to the occurrence of an event, and selectively performs only physical operations corresponding to the state activated by the event; and a real-time synchronization and correction engine that aligns the predicted state generated by the simulation engine with the actual measurement data received from the actual process equipment based on time, calculates the error between the two, and selectively corrects a part of the state based on the error, wherein the event-based simulation engine reduces the computational load by limiting the state to be computed according to the occurrence of an event and excluding physical operations for inactive states, and the real-time synchronization and correction engine corrects only the partial state corresponding to the state where the error occurred without updating the entire state. An execution package containing state information, event information, physical constraint information, and synchronization information defined in the authoring layer is directly delivered to the simulation engine and operates in a state-based execution structure without separate additional coding. By adopting a structure in which physical calculations are selectively performed only on states activated by the occurrence of events, and simultaneously selectively correcting only partial states based on the error between the predicted state and actual process data, the amount of computation is substantially reduced compared to the conventional method of performing batch calculations on the entire facility, while continuously maintaining consistency between the virtual state and the actual state.In particular, since dynamic simulations at the process or equipment level can be performed in real time even in environments with limited computational resources, such as web-based environments, it has the effect of enabling stable digital twin operation even in processes involving large-scale facilities.

[0011] Here, since the execution package includes a state machine containing transition conditions between states and an event graph defining the connection relationships between state nodes based on events, the inclusion of both the state machine and the event graph in the execution package allows the state changes and event flows of the process equipment to be defined as an executable structure rather than in the form of simple data. Furthermore, the transition conditions between each state and the connection relationships between events are explicitly expressed, thereby structurally embedding the process logic. Consequently, the state transition rules and event flows defined by the process designer can be executed directly within the simulation engine, minimizing the need for separate program logic implementation or interface mapping. Additionally, changes to the process can be reflected immediately by modifying only the state machine and event graph, which is desirable as it improves system maintenance efficiency.

[0012] Furthermore, since the aforementioned event-based simulation engine includes a state transition manager that determines the active state based on whether state transition conditions are satisfied upon the occurrence of an event, it enables state transitions that simultaneously consider the logical relationship between the current state and the event, rather than relying solely on whether an event occurs. Accordingly, this prevents incorrect state transitions or abnormal process flows, and is desirable as it ensures that interdependent state changes among multiple facilities are maintained consistently, thereby improving the stability and reliability of the entire process.

[0013] Here, the event-based simulation engine includes an optional physics executor that performs only physics operations corresponding to active states and excludes physics operations corresponding to inactive states from the computation schedule. By including an optional physics executor that performs only physics operations corresponding to active states and does not perform operations for inactive states, the computation targets are limited compared to the conventional method of calculating the physics of all equipment every frame, thereby drastically reducing the amount of computation. In particular, even when the number of equipment increases due to an increase in the scale of the process, computation is concentrated only on some equipment that affects actual operation, so the system load can be effectively controlled, and stable simulation can be performed without frame drops even in a web browser environment, which is desirable.

[0014] Furthermore, the above-mentioned selective physics executor primarily selects the state to be computed based on state dependency information representing the dependency relationship between states relative to the state in which the event occurred. By selecting the state to be computed using state dependency information representing the dependency relationship between states relative to the state in which the event occurred, it is possible to form a group of computed targets that includes not only the individual equipment where the event occurred but also equipment logically connected to that equipment. This enables a more realistic simulation that reflects the sequence of processes or kinematic connections, and at the same time, equipment unrelated to the event is excluded from the compute targets, which is desirable as it can improve computational efficiency.

[0015] Here, the selective physics executor secondarily selects the states to be subject to physics calculation based on spatial proximity or collision probability regarding the states selected first by state dependency. By secondarily selecting the targets for calculation based on spatial proximity or collision probability regarding the states selected first by state dependency, it is possible to eliminate the inefficiency of calculating even for facilities that are logically related but do not actually cause physical interference. In particular, by determining whether there is an intersection based on a spatial boundary area or an axis-aligned bounding box, the targets for calculation can be limited only to facilities within the actual physical influence range, which is desirable as it allows for the simultaneous securing of accuracy and efficiency in physics calculation.

[0016] Furthermore, the aforementioned real-time synchronization and correction engine calculates a composite error that includes a spatial error based on the spatial difference between the predicted state and the measured data, and a temporal error based on the temporal difference. By calculating the difference between the predicted state and the measured data as a composite error that considers both spatial and temporal errors rather than limiting it to a simple positional difference, it becomes possible to perform a comprehensive state evaluation that reflects not only the positional accuracy of the equipment but also the consistency of operation timing. Accordingly, it is desirable because it allows for the simultaneous consideration of processes where time delay is critical, such as high-speed transfer equipment, and processes where positional accuracy is critical, such as precision machining equipment, thereby improving the accuracy of the digital twin's reflection of reality.

[0017] Here, the real-time synchronization and correction engine is configured to perform state correction only when the composite error exceeds a preset threshold value, thereby avoiding unnecessary correction for minute errors within an acceptable range, which can prevent discontinuities in state changes or visual jumps. Additionally, it is desirable to reduce the system processing load and improve the stability of the overall simulation by reducing unnecessary data updates.

[0018] Furthermore, since the above reference value is a dynamic threshold value that is variably set according to the operating speed or process precision of the equipment, configuring the reference value as a dynamic threshold value that is variably set according to the operating speed or process precision allows for flexible adjustment of correction sensitivity depending on the process conditions. For example, it is desirable to be configured to allow relatively large errors in high-speed transfer sections and to react sensitively to small errors in precision work sections, thereby enabling the implementation of a synchronization policy optimized for process characteristics.

[0019] Here, the real-time synchronization and correction engine performs interpolation or partial updates only on data corresponding to the state where an error occurred, rather than updating the entire state during state correction. By selectively correcting only the data corresponding to the error state instead of updating the entire state collectively, it is possible to ensure consistency with the actual process while maintaining the continuity of the simulation. In particular, this is desirable because it prevents abrupt state changes or unstable operation that may occur when the entire state is overwritten, and simultaneously improves the efficiency and stability of the system by making minimal modifications only to necessary parts.

[0020] In addition, the above execution package separates the visualization model and the physical model, and since rendering processing and state calculation processing are performed independently of each other, mutual interference between visualization performance and calculation performance can be minimized. Accordingly, physical calculation performance is not degraded even when high-resolution visualization is required, and conversely, the responsiveness of the user interface can be maintained even when complex physical calculations are performed, which is desirable.

[0021] Here, operating in a web browser environment, the main execution flow performing visualization processing and the separate execution flow performing physical calculations and synchronization processing operate asynchronously. By configuring these two flows asynchronously, user interface rendering and background calculations can be separated. Consequently, smooth visualization is maintained without screen freezing even under high computational load, while continuous simulation and synchronization processing are simultaneously enabled in the background. This is desirable as it improves the overall system's real-time processing performance and user experience. Effects of the invention

[0023] According to the present invention, an execution package including state information, event information, physical constraint information, and synchronization information defined in the authoring layer is directly transmitted to a simulation engine and operates in a state-based execution structure without separate additional coding. By adopting a structure in which physical calculations are selectively performed only on states activated by the occurrence of events, and simultaneously selectively correcting only partial states based on the error between the predicted state and actual process data, the amount of computation is substantially reduced compared to the conventional method of performing batch calculations on the entire facility, while continuously maintaining consistency between the virtual state and the actual state. In particular, dynamic simulations at the process or facility level can be performed in real time even in environments with limited computational resources, such as web-based environments, thereby enabling stable digital twin operation even in processes involving large-scale facilities.

[0024] Furthermore, by including state machines and event graphs together in the execution package, state changes and event flows of process equipment are defined as executable structures rather than simple data. Transition conditions between states and relationships between events are explicitly expressed, thereby structurally embedding process logic. Consequently, state transition rules and event flows defined by the process designer can be executed directly within the simulation engine, minimizing the need for separate program logic implementation or interface mapping. Moreover, changes to the process can be reflected immediately by modifying only the state machines and event graphs, resulting in improved system maintenance efficiency.

[0025] Furthermore, by including a state transition manager that determines the active state based on whether state transition conditions are met upon the occurrence of an event, state transitions become possible that simultaneously consider the logical relationship between the current state and the event, rather than relying solely on whether an event has occurred. Consequently, this prevents incorrect state transitions or abnormal process flows, and ensures that interdependent state changes among multiple facilities are maintained consistently, thereby enhancing the stability and reliability of the entire process.

[0026] In addition, by including an optional physics executor that performs only physics operations corresponding to active states and does not perform operations for inactive states, the number of computations is drastically reduced compared to the conventional method of calculating the physics of all equipment in every frame, as the scope of computation is limited. In particular, even when the number of equipment increases due to an increase in process scale, computations are concentrated only on some equipment that affect actual operation, allowing for effective control of system load and enabling stable simulation without frame drops even in a web browser environment.

[0027] Furthermore, by selecting the states to be processed using state dependency information that indicates inter-state dependencies based on the state in which an event occurred, it is possible to form a processing target group that includes not only the individual equipment where the event occurred but also equipment logically connected to it. This enables more realistic simulations that reflect process sequences and kinematic connections, while simultaneously excluding equipment unrelated to the event from the processing targets, thereby improving computational efficiency.

[0028] In addition, by secondarily selecting computation targets based on spatial proximity or collision probability for states initially selected by state dependency, the inefficiency of performing computations on facilities that are logically related but do not actually cause physical interference can be eliminated. In particular, by determining whether there is an intersection based on spatial boundary regions or axis-aligned bounding boxes, computation targets can be limited only to facilities within the actual physical influence range, thereby ensuring both accuracy and efficiency in physical computations.

[0029] Furthermore, by calculating the difference between the predicted state and actual data as a composite error that considers both spatial and temporal errors, rather than limiting it to a simple positional difference, it becomes possible to perform a comprehensive state evaluation that reflects not only the positional accuracy of the equipment but also the alignment of operation timing. Consequently, it is possible to simultaneously consider processes where time delay is critical, such as high-speed transfer equipment, and processes where positional accuracy is critical, such as precision machining equipment, thereby improving the accuracy of the digital twin in reflecting reality.

[0030] In addition, by configuring the system to perform state correction only when the composite error exceeds a preset threshold value, unnecessary correction is not performed on minute errors within an acceptable range, thereby preventing discontinuities in state changes or visual jumps. Furthermore, by reducing unnecessary data updates, the system processing load is reduced, and the stability of the overall simulation is improved.

[0031] In addition, by configuring the reference value as a dynamic threshold that is variably set according to the equipment's operating speed or process precision, the correction sensitivity can be flexibly adjusted according to process conditions. For example, it can be configured to allow relatively large errors in high-speed transfer sections and to react sensitively to small errors in precision work sections, thereby enabling the implementation of a synchronization policy optimized for process characteristics.

[0032] Furthermore, during state correction, instead of updating the entire state in bulk, selectively correcting only the data corresponding to the state where an error occurred allows for ensuring consistency with the actual process while maintaining simulation continuity. In particular, this prevents abrupt state changes or unstable operation that may occur when overwriting the entire state, and since minimal modifications are made only to necessary parts, it has the effect of simultaneously improving system efficiency and stability.

[0033] In addition, by separating the visualization model and the physical model and performing rendering and state calculation processing independently, mutual interference between visualization performance and calculation performance can be minimized. Accordingly, physical calculation performance is not degraded even when high-resolution visualization is required, and conversely, the responsiveness of the user interface is maintained even when complex physical calculations are performed.

[0034] Furthermore, by asynchronously configuring a main execution flow that performs visualization processing in a web browser environment and a separate execution flow that performs physics calculations and synchronization processing, user interface rendering and background computations can be separated. As a result, smooth visualization is maintained without screen freezing even under high computational load, while continuous simulation and synchronization processing can be performed in the background simultaneously, thereby improving the overall system's real-time processing performance and user experience. Brief explanation of the drawing

[0036] FIG. 1 is a control block diagram of a web-based digital twin framework according to the present invention. Figure 2 is a core layer structure diagram of a web-based digital twin framework according to the present invention. FIG. 3 is a block diagram of the authoring layer and simulation engine detailed module constituting the framework of the present invention. Figure 4 is a core data flow diagram of the framework. FIG. 5 is an event graph-based optional physical operation flowchart constituting the framework of the present invention. FIG. 6 is a flowchart of a prediction-correction loop-based real-time synchronization method that constitutes the framework of the present invention. Specific details for implementing the invention

[0037] Hereinafter, a web-based digital twin framework (1) according to a preferred embodiment of the present invention will be described in detail with reference to the attached drawings.

[0038] FIG. 1 is a control block diagram of a web-based digital twin framework (1) according to the present invention, FIG. 2 is a core layer structure diagram of a web-based digital twin framework (1) according to the present invention, and FIG. 3 is a block diagram of a detailed module of a writing layer and a simulation engine constituting the framework of the present invention.

[0039] The configuration of a web-based digital twin framework (1) is explained with reference to FIGS. 1 to 3.

[0040] A web-based digital twin framework (1) is an entire system for reproducing the state, events, and physical movements of an actual manufacturing process or automated equipment in a virtual space within a web environment, and for continuously correcting the virtual state by reflecting data received from the actual site. The web-based digital twin framework (1) is largely composed of an authoring layer (10), an execution package (20), an event-based simulation engine (30), and a real-time synchronization and correction engine (40). In this case, the authoring layer (10) performs the role of defining the structure and rules of the actual process into an interpretable digital structure, the execution package (20) provides the defined results as execution units that can be directly driven by the engine, and the event-based simulation engine (30) performs state transitions and selective physical operations of the equipment within the virtual space based on the execution package (20). Additionally, the real-time synchronization and correction engine (40) maintains the consistency of the digital twin by analyzing the difference between the simulation results and the actual process data and correcting the virtual state based on the partial state where the error occurred.

[0041] The web-based digital twin framework (1) is not merely a viewer-type structure that visualizes the shape of the equipment, but has a structure in which process definition, state-based execution, event-based computation control, and real-time correction are connected as a single continuous data flow. Accordingly, the definition of the actual process can be converted into a form that can be simulated without separate manual coding, and real-time performance can be secured even under the limited computational resources of the web environment by concentrating computation only on the equipment where the event occurred or the equipment associated with it.

[0042] The authoring layer (10) is configured to organize real-world process equipment and process logic into a data structure suitable for digital twin execution, and includes a data definition module (11), a model creation module (12), a logic creation module (13), and a package creation module (14). The authoring layer (10) decomposes the shape, layout, operating range, sensor events, process status, control sequence, and synchronization criteria of the real-world equipment into independent attributes, and then reconstructs them into a structure capable of interlinking. Accordingly, the authoring layer (10) plays a role in ensuring that raw information collected from the real-world process does not remain as a simple set of files or visualization assets, but is elevated into structural metadata that takes into account state-based execution and correction.

[0043] The data definition module (11) serves as a basic information layer for process definition and includes a state information definition unit (111), an event information definition unit (112), a physical constraint information definition unit (113), and a synchronization information definition unit (114). The state information definition unit (111) defines the states that each facility or process object may have. Here, the states are not limited to simple ON / OFF values ​​and may include logical states for each process stage, such as waiting, transfer, processing, alignment, loading, stopping, error, and recovery. Additionally, the state information definition unit (111) may define attribute values, entry conditions, exit conditions, allowed actions, and associated parameters corresponding to each state. For example, in the case of a transfer device, a transfer preparation state, a transfer execution state, a transfer completion state, and a stop state may be defined, and for each state, an allowable speed range, a target position range, and subsequent state transition conditions may be set together.

[0044] The event information definition unit (112) defines various events that trigger a state transition or serve as conditions for the initiation of a physical operation. Events may include logical or physical change factors of the process, such as sensor signals, user commands, timer expiration, handshake between equipment, item arrival, completion of position alignment, and detection of collision risk. By defining the occurrence conditions, the source of occurrence, the recipient, the validity period, the priority, and the correspondence with the state transition for each event, the event information definition unit (112) provides reference information for the event graph generation and event scheduling described later.

[0045] The physical constraint information definition unit (113) defines the physical constraints under which each piece of equipment or object must operate in the virtual space. This may include movable axes, allowable rotation ranges, acceleration and deceleration ranges, collision body shapes, operating areas, stopping distances, interference prohibition areas, and relative motion relationships of connecting mechanisms. For example, in the case of a robot arm, rotation limits per joint, working radius of the end-workpiece, and collision avoidance range with adjacent equipment may be defined, and in the case of a conveyor, the conveying direction, maximum load capacity, conveying speed range, and conditions for transfer linkage between sections may be defined. This physical constraint information is not used for simple visualization purposes, but is utilized as a standard to ensure that optional physical calculations reflect actual process characteristics.

[0046] The synchronization information definition unit (114) defines reference information for comparing, aligning, and correcting virtual states and actual states. The synchronization information may include the mapping relationship between sensor channels and virtual objects, timestamp criteria, sampling period, allowable delay range, correction sensitivity per state, base threshold value per equipment, speed-based correction coefficient, etc. For example, it may be defined to apply a weighting system more sensitive to temporal errors for high-speed transfer equipment, and to apply criteria more sensitive to position errors for precision machining equipment. In this way, the synchronization information definition unit (114) provides the initial setting values ​​and control criteria for the dynamic threshold generation unit (431) and the composite error calculation unit (423) described later.

[0047] The model generation module (12) is configured to convert process information organized in the data definition module (11) into actual executable model assets and includes a visualization model generation unit (121) and a physical model generation unit (122). The visualization model generation unit (121) generates a visualization model including the external shape, color, material, layout, animation points, visualization status, etc. of the process equipment. Here, the visualization model may include mesh, texture, level of detail, object hierarchy, etc., so that it can be efficiently displayed in a web rendering environment. The visualization model generation unit (121) may include not only simple 3D external shape data but also visual rules that change the display method according to changes in state, such as color emphasis during operation, flashing of error status, and display of warning areas.

[0048] The physical model generation unit (122) generates a mathematical and structural model required for physical calculations independently of the visualization model. The physical model may include minimum units of calculation such as a center of mass, a moving axis, a velocity limit, a bounding box, a collider, a connecting link, and a radius of operation. In this case, considering the real-time web calculation environment, it is desirable for the physical model to abstract and represent only the core characteristics required for collision judgment and selective calculations, rather than reflecting all the microstructures of the actual equipment as they are. For example, even equipment with a complex external shape can be reduced to a combination of multiple simple bounding bodies or axis-aligned bounding boxes for calculation efficiency.

[0049] The logic generation module (13) is a part that configures the operation rules and state relationships of the equipment into execution logic, and includes a state machine generation unit (131) and an event graph generation unit (132). The state machine generation unit (131) defines the state set and state transition rules of each equipment or process object. The state machine may include the current state, input events, transition conditions, state after transition, and actions to be performed during the transition, and may be configured so that the states of multiple equipment having a precedence relationship according to the process sequence influence each other. For example, it may be designed so that the downstream equipment can transition to a transfer preparation state only when a processing completion event occurs at the upstream equipment.

[0050] The event graph generation unit (132) organizes the connection relationships between event occurrences and the corresponding state nodes into a graph structure. The event graph consists of nodes and edges, where the nodes represent a state or a set of states, and the edges represent an event or a transitional relationship. The event graph generation unit (132) can form a complex state network by reflecting process characteristics such as parallel processing, branching, waiting, retrying, and exception return, as well as simple sequential flow. This event graph can be used as a criterion structure for determining which group of states to bring up as a candidate for operation when an event occurs in the state dependency analysis unit (341) described later.

[0051] The package generation module (14) performs the role of bundling the results of data definition, model generation, and logic generation into a single execution unit and includes an execution package generation unit (141). The execution package generation unit (141) generates an execution package (20) by combining state definitions, event definitions, physical constraints, synchronization criteria, visualization assets, physical assets, state machines, and event graphs in a cross-referenced form. Here, it is preferable that the execution package (20) be formed as a structured data set that the engine can interpret and execute immediately upon loading, rather than a simple compressed file or asset repository. Accordingly, process changes can be modified in the authoring layer (10) and then repackaged, and the modified package can be reflected in the execution environment without separate additional coding.

[0052] The execution package (20) is an execution unit generated in the authoring layer (10) and includes an execution package module (21), and the execution package module (21) includes a visualization model (211), a physical model (212), an event graph (213), a state machine (214), and synchronization metadata (215). The execution package (20) is an intermediary structure that connects the definition result of the authoring stage and the driving criteria of the execution stage, and integrates assets and logic required in a web-based simulation environment.

[0053] The visualization model (211) is a set of external data interpreted by the visualization engine, and includes the shape, arrangement, hierarchical relationships, and display rules of equipment or objects. The physical model (212) is a set of basic physical data used for optional physical operations, and includes boundary bodies, movement ranges, connection structures, motion parameters, etc., required for collision judgment. The event graph (213) defines in a network form which state nodes are affected according to the occurrence of an event, and the state machine (214) defines the rules for logical state changes of each object. The synchronization metadata (215) includes the correspondence relationship between actual data and virtual objects, time alignment criteria, error calculation methods, correction parameters per equipment, etc. In this way, the execution package (20) integrates visual representation, logical control, physical motion, and synchronization criteria, thereby enabling the event-based simulation engine (30) and the real-time synchronization and correction engine (40), described later, to operate based on the same reference data.

[0054] The event-based simulation engine (30) is configured to perform state transitions and physical operations of a virtual process by loading an execution package (20), and includes an event processing module (31), a state transition management module (32), an optional physical operation module (33), a state selection module (34), a state dependency matrix (35), a spatial boundary area (36), an axis alignment bounding box (37), and a node cluster (38). The characteristic of the event-based simulation engine (30) is that it does not perform the same tick physical calculations for all facilities, but rather limits the states subject to calculation based on whether an event occurs, state dependency, and spatial proximity. This allows for the reduction of unnecessary calculations even when handling large-scale facilities in a web environment.

[0055] The event processing module (31) includes an event scheduler (311) as a configuration for receiving and organizing events occurring externally or internally. The event scheduler (311) collects sensor inputs, state change notifications, time-based triggers, user interactions, synchronization events between facilities, etc., and manages the occurrence time, target object, priority, and valid range of each event. For example, when a task completion event of a specific facility occurs, the event scheduler (311) can determine which state transition the event triggers, whether there is an associated subsequent facility, and whether it must be processed immediately or processed after queuing.

[0056] The state transition management module (32) performs the role of changing the state based on events and includes a state transition manager (321). The state transition manager (321) refers to the state machine (214) of the execution package (20) to determine whether the combination of the current state and the input event is an allowed transition, and if the transition is allowed, performs a change to the target state. The state transition manager (321) is not limited to changing the state of a single object but can facilitate sequential state updates when a precedence relationship between multiple objects is defined. For example, a downstream process can transition from a ready state to an operating state in response to a completion event of an upstream process.

[0057] The optional physical computation module (33) is a part that performs actual physical computations and includes an optional physical implementer (331). The optional physical implementer (331) performs position changes, speed changes, collision detection, interference review, and interlocking calculations of connecting mechanisms only for states or groups of states whose activation is confirmed by the state transition manager (321). Since the optional physical implementer (331) does not perform collective computations on the entire facility but performs computations only on targets selected by the state selection module (34), it serves as a key lightweighting means for the web-based digital twin.

[0058] The state selection module (34) is configured to select the state to be processed and includes a state dependency analysis unit (341), a spatial filtering unit (342), and an active state node group (343). The state dependency analysis unit (341) queries the state dependency matrix (35) based on the source state or source facility where the event occurred and extracts logically related states as a candidate group. The state dependency matrix (35) is a structure that expresses the dependency relationships between the entire state or facility in a matrix form, representing preceding-following process relationships, kinematic linkage relationships, control dependency relationships, etc. For example, if the source state is the lowering state of a specific loading device, the state dependency matrix (35) can be used as a criterion for extracting states related to the clamp, conveyor, and robot pickup position linked thereto as a candidate group.

[0059] The spatial filtering unit (342) filters out only the states requiring actual physical calculations from the candidate group. To do this, the spatial filtering unit (342) refers to the spatial boundary area (36) or the axis alignment bounding box (37) corresponding to each facility or state object to determine whether there is a possibility of physical interference or spatial proximity between the source object where the event occurred and the surrounding candidate objects. The spatial boundary area (36) refers to spatial limit information indicating the operational range or influence range of the facility, and the axis alignment bounding box (37) may be an orthogonal coordinate system alignment box used to simplify calculations. For example, if the operating range of the source facility and the operating range of the candidate facility do not intersect each other, they may be immediately excluded from the physical calculation target even if there is a logical dependency relationship. Conversely, if a dependency relationship exists and spatial proximity or overlap is confirmed, the corresponding state remains as a candidate for the actual calculation target.

[0060] The active state node group (343) is a set of operation target states finally selected through double filtering by the state dependency analysis unit (341) and the spatial filtering unit (342). The optional physical executor (331) performs actual operations only on the states included in the active state node group (343). In this process, the node cluster (38) can be utilized as an execution unit representing a group of interrelated states or facilities. That is, multiple nodes directly connected by a single source event can be grouped into a single cluster and activated together, and this structure is advantageous for partial simulations at the cell, line, or process block level, rather than at the single facility level.

[0061] The state dependency matrix (35), spatial boundary region (36), axis-aligned bounding box (37), and node cluster (38) support the state selection module (34) as independent data structures or auxiliary computation structures. The state dependency matrix (35) primarily provides logical connectivity, the spatial boundary region (36) and axis-aligned bounding box (37) provide spatial connectivity, and the node cluster (38) provides a grouping of execution units. By utilizing such a structure, the event-based simulation engine (30) can realize the selective computation structure required in a process-specific digital twin.

[0062] The real-time synchronization and correction engine (40) is configured to correct the state of the virtual space based on the partial state where an error occurred by comparing the predicted state generated by the simulation engine with the actual measurement data received from the actual process, and includes a data alignment module (41), an error calculation module (42), a threshold setting module (43), and a correction processing module (44). The purpose of the real-time synchronization and correction engine (40) is not simply to overwrite the entire virtual state with sensor values, but to satisfy both the continuity that the simulation must maintain and the accuracy of the actual process. Therefore, it is desirable for the engine to have a sequential structure of time axis alignment, composite error calculation, dynamic threshold determination, and partial state correction.

[0063] The data alignment module (41) performs the role of placing the predicted state and actual data on a comparable standard and includes a timestamp alignment unit (411). The timestamp alignment unit (411) compares the time information of the predicted state generated by the simulation engine with the reception time information of the sensor data to correct time axis mismatches caused by different collection cycles or communication delays. For example, since the simulation calculates the state at a fixed period but the actual sensor can transmit data asynchronously, the timestamp alignment unit (411) can unify the comparison standard time point by matching the data based on the closest time interval, or by performing interpolation or corrected time point mapping if necessary.

[0064] The error calculation module (42) quantifies the difference between the aligned predicted state and the actual state and includes a spatial error calculation unit (421), a time error calculation unit (422), and a composite error calculation unit (423). The spatial error calculation unit (421) can calculate the difference between the predicted position and the actual position, the difference between the predicted posture and the actual posture, or the deviation of the state value. For example, if the position vector of the equipment is defined, the Euclidean distance between the two vectors can be calculated and used as the spatial error. The time error calculation unit (422) calculates the difference between the time of occurrence of the predicted state and the time of reception or occurrence of the actual data; this is particularly important in cases where time delay significantly affects performance, such as with high-speed transfer equipment. The composite error calculation unit (423) integrates the spatial error and the time error at a ratio suitable for the process characteristics to generate a single judgment criterion error value. At this time, the composite error may be a simple summation, but it is desirable to configure it so that different weights are applied depending on the type of equipment.

[0065] The threshold setting module (43) provides criteria for determining what level of error requires actual correction and includes a dynamic threshold generation unit (431). The dynamic threshold generation unit (431) does not use only a fixed power threshold, but can variably set the threshold by reflecting the current operating speed of the equipment, the required level of process precision, the importance of the state, the process stage, etc. For example, in the high-speed transfer section, the threshold can be set relatively high because a minute positional difference may be within the natural range of motion error, and in the precision assembly or precision machining section, the threshold can be set low because even a small error can have a significant impact on quality. Such a dynamic threshold structure prevents excessive correction while ensuring that sensitive correction is performed in the necessary sections.

[0066] The correction processing module (44) compares the composite error with a dynamic threshold value to determine whether and to what extent actual correction is performed, and includes a correction judgment unit (441) and a partial state correction unit (442). The correction judgment unit (441) evaluates whether the composite error exceeds the dynamic threshold value, and it is preferable to initiate correction only when it exceeds it. When the composite error is below the threshold value, the simulation prediction state is maintained, thereby suppressing unnecessary state overwriting and screen jumping. The partial state correction unit (442) does not update the entire facility state collectively when correction is required, but selectively updates only the state where the error occurred or the parameters directly related thereto. For example, if only the position error is large, only the coordinate information is corrected, and the state logic itself can be maintained as is. As another example, if time delays have accumulated, the state can be gradually aligned by applying the interpolation value of the corresponding section. In this way, the partial state correction unit (442) is a key component that enhances consistency with the actual process while maintaining the continuity and stability of the entire system.

[0067] When the above configuration is combined, the authoring layer (10) structures real-world process information into data, models, and logic, and generates an execution package (20) through an execution package generation unit (141). The execution package (20) is loaded into an event-based simulation engine (30) and computes a virtual state through an event scheduler (311), a state transition manager (321), a state dependency analysis unit (341), a spatial filtering unit (342), and an optional physical executor (331). Subsequently, the computed predicted state is transmitted to a real-time synchronization and correction engine (40), and the difference from the actual state is managed through a timestamp alignment unit (411), a composite error calculation unit (423), a dynamic threshold generation unit (431), a correction judgment unit (441), and a partial state correction unit (442). As a result, the web-based digital twin framework (1) can realize a structure in which process definition, optional simulation, and real-time correction are performed continuously.

[0068] A web-based digital twin framework that supports process status, physical operation, and real-time synchronization according to an embodiment of the present invention is described in detail as undergoing computational methods and algorithmic flows to operate in a real environment.

[0069] FIG. 2 is a core layer structure diagram of a web-based digital twin framework (1) according to the present invention.

[0070] As shown in Fig. 2, the framework operates by going beyond simply visualizing data to packaging and delivering physical / logical rules of the process defined in the Twin Authoring Layer into an independent execution format called the Twin Execution Package. The converted package is loaded into the engine to form a bidirectional data exchange (state / prediction and actual data) loop between the event-based simulation engine and the real-time synchronization and correction engine. Through this, it performs a mechanism to continuously correct errors by cross-validating the results of self-prediction calculations in the virtual space with actual sensor data flowing in from the field in real time.

[0071] FIG. 3 is a block diagram of the authoring layer and simulation engine detailed module constituting the framework of the present invention.

[0072] As shown in Figure 3, the authoring layer receives heterogeneous external process equipment data (Raw Data) and operates by parsing it through the Integrated Authoring Studio by dividing roles. The Asset Composer, Physical / Behavioral Constraint Editor, and State / Event Logic Definer perform geometry merging, collider and motion range definition, state transitions, and event mapping to extract the data into an independent standardized data schema (state, event, physical constraint, and synchronization schema) in an engine-compatible format. The extracted fragmented information is merged into a single execution unit by the Twin Package Builder, a packaging engine, to create a final Twin Execution Package.

[0073] Figure 4 is a core data flow diagram of the framework.

[0074] As shown in Fig. 4, the core data flow within the framework follows a structure in which the visualization and physical models representing the shape, and the event graph, state machine, and synchronization metadata areas controlling the operation are strictly separated. Through this packaging structure, the rendering pipeline and the state operation pipeline operate independently, and data circulates in such a way that the state machine governs the logical flow of objects and the state node definitions and transition rules of the event graph direct the selective control structure of physical operations based on the occurrence of events.

[0075] FIG. 5 is an event graph-based optional physical operation flowchart constituting the framework of the present invention.

[0076] As shown in Fig. 5, event graph-based selective physical operations are performed through a filtering algorithm that blocks unnecessary operations to ensure web real-time performance. When the simulation loop starts, the engine pre-scans for the occurrence of event triggers. When a trigger is detected, the state transition manager operates to evaluate the current state based on the event graph. Subsequently, it undergoes conditional branching logic to determine whether the corresponding node is in an active state; at this stage, instead of a simple 1:1 mapping, the state dependency matrix and spatial collision bounding box are calculated to extract dependent nodes within the physical interference radius as a group. Selective physical operations are performed restrictively only on the group of nodes activated by logical / spatial criteria (Yes), while inactive nodes outside the interference radius (No) are completely excluded from the operation schedule and remain in a state of operation exclusion and waiting.

[0077] The State Dependency Matrix (SDM) is an N×N matrix composed of rows and columns representing all equipment nodes within the process. It predefines the value of a cell as 1 if a preceding-following process relationship or a physical linkage relationship exists between two nodes, and as 0 if they are independent. For the source node where an event trigger occurred, the corresponding row of the SDM is queried to initially extract a set of candidate nodes with dependency relationships. Subsequently, a Spatial Bounding Box (SBB) cross-check is performed on the extracted candidate nodes. The SBB of each equipment node predefines the maximum operating range of the equipment in the form of an Axis-Aligned Bounding Box (AABB). By calculating whether there is an intersection between the source node's SBB and the candidate nodes' SBBs, only the nodes where actual physical interference is likely to occur are selected as the final target node cluster. In this way, through a two-stage filtering structure that applies Logical Dependency Management (SDM) and Spatial Proximity (SBB) in a dual manner, physical operations of facilities unrelated to events are completely excluded from the computation schedule. The condition for an arbitrary neighbor node (ni) to be included in the dynamic active group (Gactive) is defined by the mathematical formula Gactive = {ni | Mdep(nsrc, ni) = 1 ∧ Intersect(Bsrc, Bi) = True}, where Mdep is a state dependency matrix representing the kinematic connectivity between facilities, and the Intersect function determines whether two bounding boxes intersect in three dimensions. Inactive nodes that do not meet the condition are immediately culled from the physical computation pipeline, thereby preventing the waste of computing resources at the source.

[0078] FIG. 6 is a flowchart of a prediction-correction loop-based real-time synchronization method that constitutes the framework of the present invention.

[0079] As shown in Fig. 6, real-time synchronization based on a prediction-correction loop is performed using a feedback control method rather than simple data overwriting. When the prediction analyzer within the synchronization engine preemptively calculates the future predicted state through simulation, it simultaneously receives actual process sensor data and forcibly performs timestamp alignment of the two data based on the timestamp. Subsequently, when calculating the error between the two states, instead of simply measuring the distance in 3D coordinates (spatial error), the delay time between timestamps (temporal error) is summed to calculate a spatiotemporal composite error vector. The process then passes through a branching logic that evaluates whether the calculated error exceeds a threshold value; at this point, the threshold value is not a fixed constant but a dynamic threshold that varies in real-time according to the equipment's current operating speed or processing precision. Selective correction is performed only when an exceedance of the dynamic threshold is detected (Yes), targeting and updating only the state where the error occurred; if it is below the threshold value (No), the existing predicted state is maintained.

[0080] In addition, all computation processes of the event graph-based computation (Fig. 4) and the prediction-correction loop (Fig. 5) described above are executed in a distributed manner to prevent UI freezing in web browsers. Specifically, asynchronous communication using a shared memory (Shared Array Buffer) takes place between the main rendering thread, where the visualization engine draws the 3D canvas, and the background worker thread (Web Worker) where the physics / synchronization engine runs. In particular, to eliminate serialization and deserialization overhead that occurs during bidirectional data communication between threads, the rendering buffer of the visualization model and the physics computation result buffer are allocated based on shared memory (Shared Array Buffer). Coordinate data completed in the worker thread is processed in a zero-copy manner, where it is immediately reflected in the shared memory without copying data to the main thread, and a lock-free synchronization technique based on atomic computation is applied to prevent data races. This completes a differentiated web framework execution structure that prevents garbage collection (GC) spikes and maintains smooth real-time simulation of large-scale processes without screen frame drops.

[0081] Modifiable embodiments other than the above embodiments are described.

[0082] Since state dependency information is defined to include at least one of the precede-succeed relationship, kinematic coupling relationship, and control dependency relationship between process equipment, defining state dependency information not as a simple coupling relationship but as a complex relationship including process sequence, kinematic coupling, and control flow enables more accurate identification of the equipment affected in the actual process when an event occurs. Consequently, not only is the number of unnecessary computation targets reduced, but simulations that simultaneously reflect the logical consistency and physical connectivity of the process flow become possible, thereby improving the accuracy of the digital twin in reflecting reality.

[0083] State dependency information is organized into a matrix-type data structure that allows for the rapid retrieval of other states associated with a specific state. By structuring state dependency information in a matrix format, the set of states associated with a specific event can be quickly derived within a set timeframe. Consequently, the processing time for selecting computation targets in event-based simulations is reduced, and real-time simulations can be performed without delay, even in large-scale facility environments.

[0084] Since spatial proximity or collision probability is determined based on spatial boundary regions that are dynamically updated according to the equipment's current state, determining the computation target based on spatial boundary regions that change according to the current state and operation, rather than the equipment's static shape, enables the selection of computation targets more precisely by reflecting the operating conditions in the actual process. Consequently, it is possible to accurately reflect interference situations that may occur during the movement or operation of the equipment, and it is effective in maintaining physical accuracy while reducing unnecessary calculations.

[0085] Selective physics operations are performed by setting computational priorities based on the frequency of events or state changes. By establishing priorities based on these factors, computational resources can be concentrated on critical equipment that changes frequently. Consequently, the overall responsiveness of the system can be improved, and critical process states can be prioritized even in environments with limited computational resources.

[0086] Since composite errors are configured to apply different weights depending on the type of equipment, setting different weights for spatial and temporal errors according to equipment characteristics allows for the application of error evaluation criteria suitable for each piece of equipment. For example, positional errors can be primarily reflected in precision machining equipment, while temporal errors can be primarily reflected in high-speed transfer equipment, thereby enabling precise synchronization that simultaneously satisfies various process characteristics.

[0087] Partial state correction is performed using an interpolation method that gradually reduces the difference between the current state and the measured data. By performing the correction gradually through interpolation rather than overwriting it all at once, it is possible to prevent instability caused by abrupt changes in the virtual state. Consequently, this has the effect of ensuring consistency with the actual process while maintaining the continuity of the simulation and natural state changes.

[0088] Since dynamic thresholds are additionally adjusted based not only on the equipment's operating speed but also on process stages or task importance, setting these thresholds to reflect process stages or task importance rather than simple speed criteria allows for the application of different correction policies depending on the situation, even for the same equipment. Accordingly, stricter corrections are performed during critical process stages, while operations operate to reduce the computational burden during non-critical stages, thereby simultaneously improving the efficiency and stability of the entire system.

[0089] Since the execution package is configured to be divided into multiple parts that can be loaded and executed independently, it is possible to selectively simulate only a portion of the entire process or gradually expand the system by dividing the execution package into multiple units for loading and execution. Accordingly, this has the effect of reducing initial loading time, optimizing memory usage, and facilitating the establishment of a partial process test environment.

[0090] Since the event-based simulation engine is configured to process multiple events in parallel, the response speed to multiple process events occurring simultaneously can be improved. Accordingly, this enables simulation to be performed without delay even in environments where various events occur simultaneously, such as complex manufacturing lines.

[0091] The real-time synchronization and correction engine continuously updates the state through a prediction-correction loop that is executed at regular intervals. By repeatedly running this loop, it is possible to continuously correct errors accumulated over time. Consequently, discrepancies between the virtual and actual states can be minimized even in processes operating for extended periods, thereby ensuring the long-term reliability of the digital twin.

[0092] An event-based simulation engine is configured to selectively perform physical operations on a set of states that includes not only the operation target directly corresponding to the state where the event occurred, but also surrounding states linked to the state where the event occurred through predefined relationship information. By expanding the scope of selective physical operations beyond the single state where the event directly occurred to a set of states that includes surrounding states logically or structurally linked through predefined relationship information, it is possible to encompass cases where a third party adopts a substantially identical operation target limitation structure, even without using a specific data structure known as a state dependency matrix. From a technical perspective, when an event occurs at a specific piece of equipment within a process, it is possible to consider surrounding or subsequent equipment that is actually affected, thereby enabling operation predictions that are more consistent with the actual process than narrow simulations centered on a single object. Furthermore, compared to methods that indiscriminately compute all equipment, the operation target remains limited to a set of relationship-based states, which has the effect of sufficiently reflecting process interactions while suppressing an increase in computational load. In particular, even when indirect influence exists between equipment within a process line, it can reduce prediction errors caused by omitted physical operations and enhance the reliability of the digital twin.

[0093] The event-based simulation engine is configured to utilize at least one of a bounding box, a bounding sphere, a polygonal region, or a combination thereof to determine the spatial relationship between equipment when selecting targets for physical operations. By extending the means for determining spatial proximity or collision probability beyond specific formats like Axis-Aligned Bounding Boxes (AABB) to include bounding spheres, polygonal regions, or complex spatial influence regions, it is possible to respond within the scope of rights even if a third party attempts avoidance design by modifying the spatial determination method. Technically, the most suitable spatial representation method can be selected based on the shape or motion characteristics of the equipment, allowing for more precise modeling of rotating equipment, curved moving equipment, and irregular work areas that are difficult to accurately reflect with simple box-based representations. Consequently, errors in spatial determination can be reduced, enabling the more precise extraction of only equipment with actual potential for interference, thereby improving both computational efficiency and physical accuracy. Furthermore, since different spatial representation methods can be mixed depending on process characteristics, there is also the effect of increasing system scalability and application flexibility.

[0094] The real-time synchronization and correction engine determines whether correction is necessary based on the difference between the predicted state and the actual data. Since the difference is configured to be calculated based on state comparison information including at least one of position information, time information, attitude information, velocity information, and state value, the scope of error calculation is not limited to position and time information but is expanded to include various state comparison information such as attitude information, velocity information, and state value. This allows for the inclusion of substantially identical configurations even if a third party attempts to circumvent the existing claims by partially modifying the error definition items. From a technical perspective, it is possible to reflect equipment rotation attitude errors, motion velocity discrepancies, and logical state discrepancies, which are difficult to sufficiently capture with simple coordinate differences alone, thereby enabling more multidimensional synchronization judgment. For example, in the case of a robot arm, the actual process execution result may differ if the joint attitudes differ even if the end-point positions are similar; according to this modified claim, such differences can be reflected in the correction judgment. Consequently, discrepancies between the virtual state and the actual state can be detected more precisely, and there is an effect of implementing advanced correction logic tailored to process characteristics.

[0095] The real-time synchronization and correction engine is configured to selectively perform correction on data corresponding to the state where an error occurred or on data of adjacent states linked to said state where the error occurred, rather than updating the entire state collectively. Therefore, the target of partial state correction is not limited to the state where the error was directly detected, but also includes data of adjacent states linked to that state. This allows for a response within the substantive scope of the present invention even if a third party implements a slightly expanded scope of partial correction. Technically, since a single state error often propagates to adjacent equipment or continuous states in actual processes, a method that considers adjacent states can maintain overall process consistency more stably than a method that corrects only the point where the direct error was measured. For example, if a position error in a transfer facility affects the readiness or alignment state of a subsequent loading facility, the possibility of subsequent process malfunction can be reduced by linking correction to adjacent states. This enables the maintenance of a stable digital twin state that considers process continuity beyond simple local correction, and has the effect of reducing the cumulative propagation of state inconsistencies.

[0096] The execution package is not limited to a single file or data structure; instead, it is configured to enable integrated interpretation and execution even when distributed and stored or loaded across multiple interconnected data units. By incorporating structures where multiple data units are interconnected for integrated interpretation and execution, rather than restricting the form of the execution package to a single file or data structure, it is possible to respond even when a third party attempts a design avoidance by altering the package's physical storage or loading method. Technically, in environments consisting of large-scale processes or multiple equipment groups, visualization assets, physical models, state logic, and synchronization metadata can be managed as separate units while being integrated for interpretation at runtime, thereby improving package management efficiency and system scalability. Furthermore, since it is possible to selectively load only partial processes or prioritize loading only necessary data based on the network environment, initial loading time and memory usage can be reduced. By maintaining the logical integrity of the execution package while allowing for flexible physical configuration methods, it is possible to implement a digital twin execution structure capable of responding to various deployment and operational environments.

[0097] Due to the above-described web-based digital twin framework (1), an execution package containing state information, event information, physical constraint information, and synchronization information defined in the authoring layer is directly transmitted to the simulation engine and operates in a state-based execution structure without separate additional coding. By adopting a structure in which physical calculations are selectively performed only on the state activated by the occurrence of an event, and simultaneously selectively correcting only partial states based on the error between the predicted state and actual process data, the amount of computation is substantially reduced compared to the conventional method of performing batch calculations on the entire facility, while continuously maintaining consistency between the virtual state and the actual state. In particular, dynamic simulations at the process unit or facility unit level can be performed in real time even in environments where computational resources are limited, such as web-based environments, thereby enabling stable digital twin operation even in processes involving large-scale facilities.

[0098] Furthermore, by including state machines and event graphs together in the execution package, state changes and event flows of process equipment are defined as executable structures rather than simple data. Transition conditions between states and relationships between events are explicitly expressed, structurally embedding process logic. Consequently, state transition rules and event flows defined by the process designer can be executed directly within the simulation engine, minimizing the need for separate program logic implementation or interface mapping. Moreover, changes to the process can be reflected immediately by modifying only the state machines and event graphs, thereby improving system maintenance efficiency.

[0099] Furthermore, by including a state transition manager that determines the active state based on whether state transition conditions are met upon the occurrence of an event, state transitions become possible that simultaneously consider the logical relationship between the current state and the event, rather than relying solely on whether an event has occurred. Consequently, incorrect state transitions or abnormal process flows can be prevented, and interdependent state changes among multiple facilities can be consistently maintained, thereby improving the stability and reliability of the entire process.

[0100] In addition, by including an optional physics executor that performs only physics operations corresponding to active states and does not perform operations for inactive states, the number of computations is drastically reduced compared to the conventional method of calculating the physics of all equipment in every frame, as the scope of computation is limited. In particular, even when the process scale increases and the number of equipment increases, computations are concentrated only on some equipment that affect actual operation, allowing for effective control of system load and enabling stable simulation without frame drops even in a web browser environment.

[0101] Furthermore, by selecting the states to be processed using state dependency information that indicates the dependency relationships between states based on the state in which an event occurred, it is possible to form a processing target group that includes not only the individual equipment where the event occurred but also equipment logically connected to it. This enables more realistic simulations that reflect process sequences and kinematic connections, while simultaneously improving processing efficiency by excluding equipment unrelated to the event from the processing targets.

[0102] In addition, by secondarily selecting computation targets based on spatial proximity or collision probability for states initially selected by state dependency, the inefficiency of performing computations on facilities that are logically related but do not actually cause physical interference can be eliminated. In particular, by determining whether there is an intersection based on spatial boundary regions or axis-aligned bounding boxes, computation targets can be limited only to facilities within the actual physical influence range, thereby ensuring both accuracy and efficiency in physical computation.

[0103] Furthermore, by calculating the difference between the predicted state and actual data as a composite error that considers both spatial and temporal errors, rather than limiting it to a simple positional difference, it becomes possible to perform a comprehensive state evaluation that reflects not only the positional accuracy of the equipment but also the alignment of operation timing. Consequently, it is possible to simultaneously consider processes where time delay is critical, such as high-speed transfer equipment, and processes where positional accuracy is critical, such as precision machining equipment, thereby improving the accuracy of the digital twin in reflecting reality.

[0104] In addition, by configuring the system to perform state correction only when the composite error exceeds a preset threshold value, unnecessary correction is not performed on minute errors within an acceptable range, thereby preventing discontinuities in state changes or visual jumps. Furthermore, by reducing unnecessary data updates, the system processing load can be reduced and the stability of the overall simulation can be improved.

[0105] In addition, by configuring the reference value as a dynamic threshold that is variably set according to the equipment's operating speed or process precision, the correction sensitivity can be flexibly adjusted according to process conditions. For example, it can be configured to allow relatively large errors in high-speed transfer sections and to react sensitively to small errors in precision work sections, thereby enabling the implementation of a synchronization policy optimized for process characteristics.

[0106] Furthermore, during state correction, instead of updating the entire state in bulk, selectively correcting only the data corresponding to the state where an error occurred allows for ensuring consistency with the actual process while maintaining simulation continuity. In particular, it prevents abrupt state changes or unstable operation that may occur when overwriting the entire state, and since minimal modifications are made only to necessary parts, both system efficiency and stability can be improved simultaneously.

[0107] In addition, by configuring the visualization model and the physical model separately and performing rendering and state calculation processing independently, mutual interference between visualization performance and calculation performance can be minimized. Accordingly, physical calculation performance is not degraded even when high-resolution visualization is required, and conversely, the responsiveness of the user interface can be maintained even when complex physical calculations are performed.

[0108] Furthermore, by asynchronously configuring a main execution flow that performs visualization processing in a web browser environment and a separate execution flow that performs physics calculations and synchronization processing, user interface rendering and background computation can be separated. As a result, smooth visualization is maintained without screen freezing even under high computational load, while continuous simulation and synchronization processing can be performed in the background simultaneously, thereby improving the overall system's real-time processing performance and user experience. Explanation of the symbols

[0110] 1: Web-based digital twin framework 10: Authoring Layer 11: Data Definition Module 111: State Information Definition Section 112: Event Information Definition Section 113: Physical Constraint Information Definition Section 114: Synchronization Information Definition Section 12: Model Creation Module 121: Visualization Model Creation Section 122: Physical Model Generation Section 13: Logic Generation Module 131: State Machine Generation Unit 132: Event Graph Generation Unit 14: Package creation module 141: Executable package creation section 20: Executable package 21: Executable package module 211: Visualization Model 212: Physical Model 213: Event Graph 214: State Machine 215: Synchronization Metadata 30: Event-based simulation engine 31: Event handling module 311: Event scheduler 32: State Transition Management Module 321: State Transition Manager 33: Optional physics computation module 331: Optional physics implementer 34: State Selection Module 341: State Dependency Analysis Unit 342: Spatial filtering section 343: Active node group 35: State dependency matrix 36: Spatial boundary region 37: Axis-aligned bounding box 38: Node cluster 40: Real-time synchronization and correction engine 41: Data alignment module 411: Timestamp alignment section 42: Error calculation module 421: Spatial error calculation unit 422: Time error calculation unit 423: Composite error calculation unit 43: Threshold setting module 431: Dynamic threshold generation unit 44: Correction processing module 441: Correction judgment unit 442: Partial state correction unit

Claims

Claim 1 A web-based digital twin framework for implementing a digital twin in a virtual space based on the state, events, and physical behavior of process equipment, comprising: an authoring layer that defines a data structure including state information, event information, physical constraint information, and synchronization information based on data collected from real process equipment, and generates an execution package including a visualization model, a physical model, an event graph, a state machine, and synchronization metadata; and an event-based simulation engine that receives the execution package, manages state transitions according to the occurrence of an event, and selectively performs only physical operations corresponding to the state activated by the event. A web-based digital twin framework comprising a real-time synchronization and correction engine that aligns a predicted state generated by the simulation engine with actual measurement data received from actual process equipment based on time, calculates an error between the two, and selectively corrects a part of the state based on the error; wherein the event-based simulation engine reduces the computational load by limiting the state to be computed according to the occurrence of an event and excluding physical computations for inactive states; wherein the real-time synchronization and correction engine corrects only the partial state corresponding to the state where an error occurred without updating the entire state; and wherein the event-based simulation engine includes a selective physical executor that performs only physical computations corresponding to the active state and excludes physical computations corresponding to the inactive state from the computation schedule. Claim 2 A web-based digital twin framework according to claim 1, characterized in that the execution package includes a state machine including transition conditions between states and an event graph defining connection relationships between state nodes according to events. Claim 3 A web-based digital twin framework according to paragraph 2, wherein the event-based simulation engine includes a state transition manager that determines an active state based on whether a state transition condition is satisfied when an event occurs. Claim 4 delete Claim 5 A web-based digital twin framework according to claim 1, wherein the selective physical executor primarily selects a state to be operated on based on state dependency information indicating dependency relationships between states based on the state in which an event occurred. Claim 6 A web-based digital twin framework according to claim 5, wherein the selective physics implementer secondarily selects a state to be subject to physics operation based on spatial proximity or collision possibility with respect to the firstly selected states. Claim 7 A web-based digital twin framework according to claim 6, wherein the real-time synchronization and correction engine calculates a composite error including a spatial error based on the spatial difference between the predicted state and the actual data and a temporal error based on the temporal difference. Claim 8 A web-based digital twin framework according to claim 7, wherein the real-time synchronization and correction engine performs state correction only when the composite error exceeds a preset reference value. Claim 9 A web-based digital twin framework characterized in that, in claim 8, the reference value is a dynamic threshold value that is variably set according to the operating speed or process precision of the equipment. Claim 10 A web-based digital twin framework according to claim 1, wherein the real-time synchronization and correction engine performs interpolation or partial updating only on data corresponding to the state where an error occurred, without updating the entire state during state correction. Claim 11 A web-based digital twin framework according to claim 10, wherein the execution package includes a visualization model and a physical model separately, and rendering processing and state operation processing are performed independently of each other. Claim 12 A web-based digital twin framework according to claim 11, characterized in that it operates in a web browser environment and a main execution flow that performs visualization processing and a separate execution flow that performs physical calculations and synchronization processing operate asynchronously.

Images