Digital twin workshop equipment cloud running high-availability collaborative computing method

Abstract

This invention discloses a highly available collaborative computing method for cloud-based operation of digital twin workshop equipment. Specifically, it involves: collecting physical workshop equipment and environmental information to perform multi-dimensional system information modeling, forming a high-fidelity digital model; then associating the digital model with the computational model; separating logical operations from visualization operations, with the former processed by the CPU and the latter transferred to the GPU via a programmable rendering pipeline; ensuring synchronization of the computation process by recording the GPU runtime between each frame as a simulation clock; transmitting the information connected to the computational model from the digital model to the twin model via a computational stream, obtaining the manufacturing equipment's topology nodes and control points, and controlling them through a PLC program to achieve bidirectional mapping between the physical equipment and the twin model; and providing real-time interaction and control to users through the twin model as a web service. This invention achieves higher operating efficiency and availability when the digital twin workshop equipment is running in the cloud.

Description

Technical Field

[0001] This invention belongs to the field of discrete manufacturing system modeling and simulation technology, specifically relating to a highly available collaborative computing method for cloud-based operation of digital twin workshop equipment. Background Technology

[0002] Discrete manufacturing systems are the most widely used form of manufacturing, used in the production of everything from high-speed trains and aerospace equipment to electronic products and daily necessities. However, due to the unpredictable changes in both external factors (market demand) and internal factors (products or production systems), discrete manufacturing systems are highly dynamic and uncertain. This poses significant challenges to their operation and management, leading to the need to integrate information technology with the workshop to build smart workshops.

[0003] Digital twins (DT), as a crucial enabling technology for promoting intelligent manufacturing upgrades and achieving smart manufacturing, emphasize various data machine tools, dynamic evolution throughout the entire lifecycle, and model-based optimization iterations. These aspects are highly suitable for building information technology-enabled smart workshops, and production workshops based on digital twin technology are called digital twin workshops. Currently, research on digital twin workshops has moved from theory to practical application. Digital twin workshops implemented using a service-oriented architecture allow users to easily access them from any device via a browser. Building a service-oriented digital twin workshop first requires constructing a virtual model of the workshop in the cloud to map the physical workshop and support a series of workshop applications. However, digital twin workshops have numerous equipment models, complex production logic, and the fusion of virtual and physical information across multiple dimensions consumes significant resources. This cloud-based service approach is limited by constraints in computing and storage resources, creating bottlenecks that hinder data synchronization and reduce real-time performance. Therefore, an efficient method is needed to process the data in the cloud-based digital twin workshop.

[0004] Currently, there are few methods for solving the problem of efficient computing in digital twin workshops. Many methods only provide a preliminary solution for efficient computing architecture without fully implementing the key technologies. In particular, there is a lack of research on how to perform high-fidelity mapping of the system environment and how to maintain a balance between mapping accuracy and operating efficiency. Summary of the Invention

[0005] To address the above problems, this invention provides a highly available collaborative computing method for cloud-based operation of digital twin workshop equipment.

[0006] The present invention provides a highly available collaborative computing method for cloud-based operation of digital twin workshop equipment, specifically comprising:

[0007] (1) Collect information on equipment and environment in the physical workshop, and use three-dimensional modeling technology based on computer graphics and discrete event system modeling technology based on multi-body systems to model the manufacturing equipment in terms of geometry, material, topology and behavior to form a high-fidelity digital model. Then, associate the digital model with the computational model.

[0008] Physical information acquisition: Acquisition methods include internal acquisition relying on the equipment's internal control system and external data communication interface, and external acquisition relying on various sensors; physical information includes manufacturing equipment information and workshop environment information. Manufacturing equipment information includes equipment coordinates, dimensions, materials, structure, and orientation, while workshop environment information includes lighting environment information.

[0009] Based on the collected information, the manufacturing equipment is geometrically modeled in 3D modeling software. By linking node materials and texture maps, rendering attributes are added to the geometric model to establish a 3D model. At the same time, according to the spatial degrees of freedom and assembly relationships of each component of the equipment, the moving parts are connected in sequence to form a topological chain and a parent-child structure is set to establish a local motion coordinate system. Based on the physical workshop and equipment operation process, its behavioral logic is determined and a logical model is established.

[0010] (2) Representation and solution of logic model based on simulation and motion topology: Separate logic operation and visualization operation. The former is processed in the CPU, while the latter is transferred to the GPU through programmable rendering pipeline. The operation process is synchronized by recording the GPU running time between each frame as the simulation clock.

[0011] (3) The information connected by the computational model is transmitted from the digital model to the twin model through the computational flow to obtain the topology nodes and control points of the manufacturing equipment. The PLC program controls the equipment to achieve bidirectional mapping between the physical equipment and the twin model. The twin model is used as a web service to provide users with real-time interaction and control.

[0012] Furthermore, a three-level mapping based on shadow maps is adopted for lighting environment information. The hard shadow rendering algorithm is set as a level I service; the PCF algorithm performs random sampling of fixed areas of the shadow map, which is a level II service; and the improved PCSS algorithm is a level III service.

[0013] The improved PCSS algorithm is as follows:

[0014] First, Poisson disk sampling is performed near the shading point to compare depths. Pixels in shadow are called occluders, and the average depth of the occluders is recorded.

[0015] Then, the shadow filtering range is determined based on the average depth of the occluder.

[0016] Finally, based on the desired filtering range, Poisson disk sampling is performed on the shadow map and the proportion of unoccluded sampling points is calculated. This proportion is then multiplied by the model lighting color obtained using the Blinn-Phong model to output the final pixel color.

[0017] Furthermore, the logical model is lightweighted: the quadratic error measure (QEM) algorithm is used to reduce the number of model patches, the EXT_meshopt_compression algorithm is used to compress the model mesh data, and finally the model is output as a .glb file.

[0018] The solution to the logical model is as follows:

[0019] The logic model of the digital twin workshop manufacturing equipment is described based on a multibody system, and its motion logic is solved based on discrete event simulation theory.

[0020] Multibody systems accurately describe the motion relationships of complex systems by abstracting and refining their motion characteristics. Topological models are abstract mappings of multibody systems, used to express the hierarchical relationship of system motion, control the motion of digital model components, and support the logical behavior of twin models. A topological model is a chain-like structure composed of multiple nodes, each node being associated with one or more three-dimensional model components. It describes the hierarchical relationship of device component motion and has at most one spatial degree of freedom. The model's motion logic is described by combining topological nodes in a chain.

[0021] To ensure synchronization between logic and rendering, the calculation process is handled in the CPU, and the output model transformation matrix is ​​passed to the GPU cache for visualization updates. At the same time, the simulation clock is updated and advanced when reading GPU rendering frames.

[0022] The beneficial technical effects of this invention are as follows:

[0023] (1) This method designs a system environment mapping service. Compared with direct mapping, it can achieve a balance between efficiency and accuracy in system operation through configuration and self-optimization, and ensure high availability of operation.

[0024] (2) This method designs the PCSS algorithm. Compared with the traditional PCSS algorithm based on random sampling, the proposed PCSS algorithm based on Poisson disk sampling has less noise and higher simulation fidelity.

[0025] (3) This method designs a collaborative computation of digital twin workshop operation visualization and logic. By driving the clock of logic simulation computation through rendering frames, it can bring advantages in terms of real-time performance and synchronization of model operation. Attached Figure Description

[0026] Figure 1A cloud-based, highly available, real-time collaborative computing framework for manufacturing equipment in digital twin workshops.

[0027] Figure 2 A schematic diagram for modeling multi-dimensional information.

[0028] Figure 3 The process of providing three-level illumination mapping services.

[0029] Figure 4 This is a schematic diagram of the topological nodes and control points of a six-degree-of-freedom robot.

[0030] Figure 5 Comparison of effects under different shadow algorithms (ad represents hard shadow, PCF, uniform sampling PCSS, and Poisson disk sampling PCSS, respectively).

[0031] Figure 6 The calculation process for a six-DOF robot is as follows (ae represents the start of movement, gripping the workpiece, placing the workpiece, robot resetting, and gripping failure, respectively).

[0032] Figure 7 This refers to the visual recognition process of a robotic arm.

[0033] Figure 8 This is the system's user interface. Detailed Implementation

[0034] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0035] This invention discloses a highly available collaborative computing method for cloud-based operation of digital twin workshop equipment. It places logical computation and visualization computation on the CPU and GPU respectively for collaborative processing to improve the real-time performance of the digital twin workshop system. The framework constructs high-fidelity digital models of the physical workshop manufacturing equipment from both 3D and logical model dimensions. Taking lighting and shadows, which enhance the realism of virtual scenes and assist visual recognition and noise reduction, as examples, it discusses high-fidelity mapping and service matching optimization schemes for the system's operating environment. Based on shadow mapping, it proposes a configurable and optimizable shadow computation service for visualization model computation, and adaptively improves the PCSS algorithm. It proposes a simulation-based equipment motion calculation algorithm for logical model computation, using the frame interval of visualization computation as the logical simulation clock. The motion calculation results obtained from the simulation are written to a GPU constant buffer to drive visualization rendering in real time, thus solving the synchronization problem between logical computation and visualization and ensuring accurate and continuous model motion.

[0036] A block diagram of a highly available collaborative computing method for cloud-based operation of digital twin workshop equipment according to the present invention is shown below. Figure 1As shown, the process begins by collecting physical workshop equipment and environmental information using devices such as RFID and sensors. Then, using computer graphics-based 3D modeling and multi-body system discrete event modeling techniques, the manufacturing equipment is modeled using multi-dimensional information including geometry, material composition, topology, and behavior, forming a high-fidelity digital model. This digital model is then linked to the computational model. The computational model can be divided into logical computation and visualization computation based on different functions. Since the overall operation of the digital twin workshop requires server-side computation and storage, to alleviate server resource pressure, visualization computation and manufacturing equipment logical operations are placed on the client side. This computation process consumes significant resources, but CPU computing resources are limited. Furthermore, there are issues such as high coupling, low cohesion, and poor scalability between logical and visualization computations. To achieve real-time performance, additional GPU computing resources are needed. Considering that logical computation involves smaller data volumes and higher computational complexity for individual models, a CPU-based architecture is more suitable; while visualization computation involves larger data volumes, lower computational complexity for model vertices, and higher similarity, making GPU computation more appropriate. Therefore, by separating logical operations from visualization operations, the former is processed in the CPU, while the latter is transferred to the GPU through a programmable rendering pipeline. The synchronization of the operation process is ensured by recording the GPU runtime between each frame as the simulation clock. This process results in low coupling and high cohesion between logical and visualization operations, which can reduce CPU pressure and improve GPU utilization and system computing efficiency.

[0037] Finally, the information connected by the computational model is transmitted from the digital model to the twin model through a computational stream to obtain the topology nodes and control points of the manufacturing equipment. The PLC program then controls the equipment to achieve bidirectional mapping between the physical equipment and the twin model. The twin model is then used as a web service to provide users with real-time interaction and control.

[0038] System multi-dimensional modeling:

[0039] A digital twin workshop system integrates data, logic, and control signals from the physical workshop, digital workshop, and twin workshop. The physical workshop and the twin workshop can interact and map bidirectionally in real time. To achieve this, it is first necessary to accurately model the multi-dimensional information of the equipment in the physical workshop within a virtual environment, establishing a high-fidelity digital model. Simultaneously, the operating environment of the physical equipment plays a crucial role in determining its efficiency and lifespan, requiring mapping of this environment to accurately calculate the overall system state and performance. A schematic diagram of the multi-dimensional information modeling is shown below. Figure 2 As shown.

[0040] To accurately map physical objects to digital models, their physical information needs to be collected. Collection methods include internal acquisition relying on the equipment's internal control system and external data communication interfaces, and external acquisition relying on various sensors. Physical information includes manufacturing equipment information and workshop environment information. The former includes equipment coordinates, dimensions, materials, structure, and pose, while the latter, in this invention, mainly uses ambient lighting information, including light source area, light intensity, and light direction. Based on the collected information, the manufacturing equipment is geometrically modeled in 3D modeling software (such as Blender). Rendering attributes are added to the geometric model by linking node materials and texture maps, establishing a 3D model. Simultaneously, based on the spatial degrees of freedom and assembly relationships of each component, the moving parts are sequentially connected to form a topological chain, and a parent-child structure is set to establish a local motion coordinate system. Based on the physical workshop and equipment operation flow, its behavioral logic is determined, and a logical model is established. To meet the requirements for lightweight model design and reduce system resource consumption, the Quadric Error Metrics (QEM) algorithm is used to reduce the number of model patches. The EXT_meshopt_compression algorithm is then used to compress the model mesh data, finally outputting it as a .glb file. Through these key steps, a multi-dimensional digital model is essentially established, accurately reflecting the physical equipment and workshop environment information, thus providing a foundation for subsequent collaborative data computation.

[0041] Lighting environment mapping based on shadow mapping:

[0042] Shadows are the result of light being blocked. Shadow effects not only enhance the realism of virtual scenes but also aid in workshop analysis and optimization. For example, in computer vision-based product inspection, it's necessary to remove shadow noise from images to avoid misjudgments. In this case, simulating a realistic lighting and shadow environment provides a reference for the process. Therefore, this invention uses lighting and shadow effects as an example to simulate the real environment of the system. With the development of computer graphics, many shadow generation algorithms have emerged. Currently, the most commonly used algorithms are those based on shadow mapping, such as PCF (Percentage CloserFilter), PCSS (Percentage Closer Soft Shadows), and VSSM (Variance Soft Shadow Mapping). These algorithms have partially solved the shadow rendering problem in the field of computer graphics. However, different algorithms have varying complexity and accuracy. To maintain a balance between simulation accuracy and operational efficiency, based on service matching and optimization theory, a three-level mapping based on shadow mapping is proposed. This integrates three representative algorithms (hard shadows, PCF, and PCSS) and selects the most suitable mapping service based on user needs and actual operating conditions.

[0043] The complexity differences among various algorithms mainly lie in the shadow shading calculation. The hard shadow rendering algorithm compares depth information only once, making it the most efficient, but it produces severe aliasing; therefore, it is designated as Level I service. The PCF algorithm randomly samples a fixed area of ​​the shadow map, reducing aliasing to some extent; this is designated as Level II service. The PCSS algorithm dynamically determines the PCF sampling area based on random sampling, simulating the phenomenon in digital twin workshops where shadows are softer closer to the light source due to varying sizes of manufacturing equipment (such as overhead cranes or AGVs); however, it has the highest resource consumption and is designated as Level III service. During system operation, users can select the appropriate service level based on their environmental information mapping accuracy requirements. Simultaneously, the system collects frame rate information in real time and degrades the service when the frame rate does not meet the requirements, ensuring real-time performance. The process is as follows: Figure 3 As shown.

[0044] Both PCF and PCSS algorithms require random sampling of shadow maps. Traditional algorithms use uniform sampling, which may cause sampling point overlap and requires a large number of samples. Furthermore, the diverse shapes and complex structures of manufacturing equipment in digital twin workshops necessitate reducing the sampling frequency in areas far from the sampling center point, thus concentrating sample points towards the center point and reducing noise. Therefore, this invention improves the PCSS algorithm by applying Poisson disk sampling instead of uniform sampling. The sampling algorithm is shown in Table 1.

[0045] Table 1 Poisson disk sampling algorithm

[0046]

[0047]

[0048] The improved PCSS algorithm mainly consists of the following steps: First, Poisson disk sampling is performed near the shading point, and depth comparison is conducted. Pixels in shadow are called occluders, and the average depth of the occluders is recorded. Then, the shadow filtering range is determined based on the average depth of the occluders. Finally, based on the determined filtering range, Poisson disk sampling is performed in the shadow map, and the proportion of unoccluded sampling points is calculated. This proportion is then multiplied by the model lighting color obtained using the Blinn-Phong model to output the final pixel color. The algorithm is shown in Table 2.

[0049] Table 2 Improved PCSS Algorithm

[0050]

[0051] Simulation-based motion logic solution:

[0052] To ensure smooth and accurate model motion in a digital twin workshop, this invention describes the logical model of manufacturing equipment in the digital twin workshop based on a multibody system and solves its motion logic based on discrete event simulation theory. A multibody system accurately describes the motion relationships of a complex system by abstracting and refining its motion characteristics. A topological model is an abstract mapping of the multibody system, used to express the hierarchical relationship of system motion, control the motion of digital model components, and support the logical behavior of the twin model. The topological model is a chain-like structure composed of multiple nodes, each node associated with one or more 3D model components, describing the hierarchical relationship of equipment component motion, and having at most one spatial degree of freedom (translation along the X, Y, Z directions or rotation around the X, Y, Z axes). The model's motion logic is described by combining topological nodes in a chain. Since the motion logic of a robotic arm is relatively complex, this section uses a six-DOF robotic arm as an example, representing and solving its logical model based on simulation and motion topology. To ensure synchronization between logic and rendering, the calculation process is handled in the CPU, and the output model transformation matrix is ​​passed to the GPU cache for visualization updates. Simultaneously, when reading GPU rendering frames, the simulation clock is updated.

[0053] Since the motion behavior of components needs to be determined based on their local coordinate system, control points are set for each component model bound to a topology node. These control points represent the center of the component's coordinate transformation and are used to control the motion behavior of the logical model. The topology model and control points of a six-DOF robot are as follows: Figure 4 As shown, when the node pointed to by the end of the arrow moves, the node pointed to by the start of the arrow moves accordingly.

[0054] The six-DOF robotic arm operates by grasping a workpiece at an arbitrary position, placing it in a designated storage location, and then resetting the arm to its initial posture. Since each component of the robotic arm moves differently and has different motion targets, each component needs to be calculated individually. Simulation methods are used to calculate the component motion targets and solve for the coordinate transformation matrix of each component in each frame. Because the grasping and placement logics of the robotic arm are essentially the same, the workpiece position and storage location are abstracted as target points.

[0055] The S-node controls the rotation of the robot arm's rotary axis. The forward vector of the S-node is directed towards the target point from S. Its motion target can be expressed as follows:

[0056]

[0057] Where θ s Let t be the angle that node S needs to rotate. s Let f be the vector projected onto the XOZ plane from the direction vector from S to the target point. sLet c be the forward vector of node S, corresponding to the vector SW projected from S to node W onto the XOZ plane. Since the rotation angle in equation (1) cannot be determined as positive or negative, the direction of rotation needs to be determined by the vector product, as shown in the following formula. s =1, the rotation angle is positive, and the direction is counterclockwise. If c s =-1, the rotation angle is positive, and the direction is clockwise.

[0058]

[0059] Node L controls the rotation of the lower arm's rotation axis. Node W is mapped to the target point, and the target angle of rotation for node L is θ. l Its formula is:

[0060] θ l =arccos(LW·LW′) (3)

[0061] Where W′ represents the coordinates of the target point, and the direction of rotation is determined by the vector product.

[0062] The U-node controls the rotation of the robotic arm's upper arm axis, adjusting the robotic arm's range of motion to the target point. The target angle of the U-node's rotation is θ. u Its formula is:

[0063]

[0064] The R node controls the rotation of the robot arm's forearm axis, and the right vector r of the R node... r With the right vector r of the target point w′ Alignment, the formula is:

[0065]

[0066] Node B controls the rotation of the robotic arm's swing axis, and the upper vector u of node B... b With the vector u at the target point w′ Alignment, the formula is:

[0067]

[0068] The T-node controls the rotation of the robotic wrist's pivot axis, and the right vector r of the T-node... t With the right vector r of the target point w′ Alignment, the formula is:

[0069]

[0070] Before each frame rendering call, the CPU advances each motion node once. Traditional calculation methods advance a fixed step size based on the motion speed, making it difficult to synchronize rendering and logic calculation, easily causing screen stuttering and flickering, and failing to guarantee consistent component motion speeds on different hosts. To solve the synchronization problem between visualization rendering and logic calculation, the GPU rendering time between each frame is used as the clock for the robot arm simulation calculation, dynamically calculating the motion step size of each component per frame. If the position of control point W is the same as the target point position and all nodes meet the motion target, the workpiece position is updated, and the goods are grasped or placed. If a component exceeds a certain limit constraint (e.g., the angle between node L and node U must not exceed 180° or be lower than 0°), the component cannot continue to move, and its coordinate transformation matrix is ​​not updated in this calculation. If all nodes have finished solving or have exceeded the limit, and the target point still cannot be reached, the target point is too far away, and the goods cannot be grasped or placed. The robot arm motion logic simulation calculation algorithm is shown in Table 3.

[0071] Table 3 Simulation algorithm for motion logic of a six-DOF manipulator

[0072]

[0073]

[0074] Example:

[0075] Based on the concept of cloud simulation, this invention constructs a web-based digital twin workshop modeling and simulation platform. Using a six-DOF robotic arm and a machining workshop for a steering frame as application examples, the proposed collaborative computing framework and its key methods are verified and analyzed through case studies. The platform adopts a front-end / back-end separation framework. Workshop visualization rendering and equipment motion logic calculation are placed in the front-end browser, while the collected physical equipment data, environmental data, and digital models are stored on the back-end server. The model and physical system parameters are loaded into the virtual environment via HTTP requests. Finally, JavaScript and GLSL programming are combined to complete the collaborative calculation of logical and visual data.

[0076] Figure 5 The image shows shadow effects rendered by a six-DOF robotic arm under parallel light source illumination using hard shadows, PCF, uniform sampling PCSS, and Poisson disk sampling PCSS, with the latter three algorithms sampling at a frequency of 200 times per frame. It can be seen that simulation accuracy and fidelity increase with the mapping service level. Furthermore, compared to the uniform sampling PCSS algorithm, the Poisson disk sampling PCSS algorithm proposed in this invention has more concentrated samples, generates less noise in the shadows, and achieves higher fidelity.

[0077] Based on the six-degree-of-freedom manipulator operation logic of this invention, the operating speed of each node of the manipulator is set to 0.5π / s. Simultaneously, the current position and placement point of the material are set, and the manipulator motion simulation is performed. The calculation process is as follows: Figure 6 As shown. From Figure 6 As can be seen, the calculated data is synchronized with the rendering, the motion process is continuous and smooth, and the calculation results are accurate and reliable.

[0078] Figure 7 (a) Images captured during the robot's visual recognition process show that the workpiece's exact position is difficult to determine due to shadow occlusion, making it impossible to ascertain whether the robot has reached its designated position. Therefore, the shadow mapping method presented in this paper is used to simulate the lighting environment, yielding the following rendering results: Figure 7 As shown in (b), the shadow weight in the fragment shader is then reduced and the image is rendered again, with the following result: Figure 7 As shown in (c). From Figure 7 As can be seen from the above steps, the position of the part can be clearly determined, and the position of the robotic arm can be accurately identified.

[0079] Finally, taking a bogie frame processing workshop as an example, the equipment model was laid out using the actual physical poses of the equipment, and a workshop model was established. This workshop includes 118 manufacturing equipment, and its total model volume is 22.52M. The interface is as follows. Figure 8 As shown.

Claims

1. A highly available collaborative computing method for cloud-based operation of digital twin workshop equipment, characterized in that, Specifically: (1) Collect information on equipment and environment in the physical workshop, and use three-dimensional modeling technology based on computer graphics and discrete event system modeling technology based on multi-body systems to model the manufacturing equipment in terms of geometry, material, topology and behavior to form a high-fidelity digital model. Then, associate the digital model with the computational model. Physical information acquisition: Acquisition methods include internal acquisition relying on the internal control system of the equipment and external data communication interface, and external acquisition relying on various sensors; physical information includes manufacturing equipment information and workshop environment information. Manufacturing equipment information includes equipment coordinates, dimensions, materials, structure, and orientation, while workshop environment information includes lighting environment information. The lighting environment information adopts a three-level mapping based on shadow maps. The hard shadow rendering algorithm is set as a level I service; the PCF algorithm performs random sampling of fixed areas of the shadow map, which is a level II service; and the improved PCSS algorithm is set as a level III service. The improved PCSS algorithm is specifically as follows: First, Poisson disk sampling is performed near the shading point to compare depths. Pixels in shadow are called occluders, and the average depth of the occluders is recorded. Then, the shadow filtering range is determined based on the average depth of the obstruction; Finally, based on the desired filtering range, Poisson disk sampling is performed in the shadow map and the proportion of unoccluded sampling points is calculated. This proportion is then multiplied by the model lighting color obtained using the Blinn-Phong model to output the final pixel color. Based on the collected information, the manufacturing equipment is geometrically modeled in 3D modeling software. By linking node materials and texture maps, rendering attributes are added to the geometric model to establish a 3D model. At the same time, according to the spatial degrees of freedom and assembly relationship of each component of the equipment, the moving parts are connected in sequence to form a topological chain and a parent-child structure is set to establish a local motion coordinate system. Based on the physical workshop and equipment operation process, its behavioral logic is determined and a logical model is established. The logical model is lightweighted by using the quadratic error measure (QEM) algorithm to reduce the number of model patches, compressing the model mesh data using the EXT_meshopt_compression algorithm, and finally outputting it as a .glb file. (2) Representation and solution of logical model based on simulation and motion topology: Separate logical operation and visualization operation. The former is processed in the CPU, while the latter is transferred to the GPU through programmable rendering pipeline. The operation process is synchronized by recording the GPU running time between each frame as the simulation clock. (3) The information connected by the computational model is transmitted from the digital model to the twin model through the computational flow, the topology nodes and control points of the manufacturing equipment are obtained, and the PLC program is used to control them to realize the bidirectional mapping between the physical equipment and the twin model. The twin model is used as a web service to provide users with real-time interaction and control.

2. The highly available collaborative computing method for cloud-based operation of digital twin workshop equipment according to claim 1, characterized in that, The solution to the logical model is as follows: The logic model of the digital twin workshop manufacturing equipment is described based on a multi-body system, and its motion logic is solved based on discrete event simulation theory. Multibody systems accurately describe the motion relationships of complex systems by abstracting and refining their motion characteristics. Topological models are abstract mappings of multibody systems, used to express the hierarchical relationship of system motion, control the motion of digital model components, and support the logical behavior of twin models. A topological model is a chain-like structure composed of multiple nodes, each node being associated with one or more three-dimensional model components. It describes the hierarchical relationship of device component motion and has at most one spatial degree of freedom. The model's motion logic is described by combining topological nodes in a chain. To ensure synchronization between logic and rendering, the calculation process is handled in the CPU, and the output model transformation matrix is ​​passed to the GPU cache for visualization updates. At the same time, the simulation clock is updated and advanced when reading GPU rendering frames.

Images