A closed-loop digital twin simulation method and system based on multi-physical signal fusion

Abstract

The application belongs to the field of industrial digital twin simulation control, and discloses a closed-loop digital twin simulation method and system based on multi-physical signal fusion. A collaborative architecture with data homology, task parallelism and effective instruction feedback is constructed, equipment monitoring signals are received and multi-physical field coupling simulation is carried out, and convergent homologous simulation data are generated; equipment fault and simulation computing power prediction are executed in parallel within a unified decision cycle; only when both tasks output effective results, the standardized instruction feedback is synthesized. The application solves the problems of data island, response delay and low computing power utilization rate existing in the traditional chimney type architecture, and can be widely applied to digital twin simulation and operation and maintenance control scenes of various industrial equipment.

Description

Technical Field

[0001] This invention relates to the field of industrial digital twin simulation control, and in particular to a systematic architecture method and system for improving simulation accuracy and real-time response capability and optimizing computing power utilization. Background Technology

[0002] Existing industrial digital twin systems generally adopt a siloed, discrete architecture, with various types of monitoring data stored and processed independently, easily forming data silos. Functional modules operate serially, resulting in significant asynchronous latency between modules, making it difficult to meet the real-time control requirements of industrial sites. Furthermore, traditional solutions employ inefficient computing power allocation, leading to severe resource waste. Simulation, diagnostic, and scheduling modules are fragmented, resulting in poor coordination. Inconsistent simulation convergence rules and instruction formats across different devices and platforms lead to inconsistent simulation quality and high difficulty in cross-platform integration. These issues collectively hinder the large-scale application of digital twin technology in high-end industrial equipment. Summary of the Invention

[0003] The technical problem to be solved by the present invention To address the shortcomings of existing digital twin systems, such as inconsistent data, high system latency, low computing power utilization, poor module coordination, chaotic simulation and instruction standards, and difficulties in device integration, this invention proposes a closed-loop collaborative scheme that is redesigned at the architectural level to systematically solve the above problems. 1. The technical solution of the present invention This invention establishes three core constraints: unique input of data from the same source, parallel operation of dual tasks with a unified cycle, and data can only be transmitted after both tasks are valid. It deeply integrates multiphysics simulation, fault diagnosis, and computing power scheduling into a logically inseparable closed-loop whole. The specific solution is as follows: 1.1 Method and Flow S1, Receive monitoring signals from industrial equipment. S2. Update the digital twin model of the equipment based on the monitoring signals, perform multiphysics coupling simulation and dynamically adjust boundary conditions until the simulation model converges, and output converged homogeneous simulation data. This converged homogeneous simulation data serves as the unique and exclusive input for downstream fault diagnosis and computing power prediction tasks. This multiphysics coupling simulation can employ various coupling modes such as electrothermal, acoustic vibration, and heat flow as needed, with specific correlations defined in claim 3. The simulation uses the global relative average error of key physical quantities across the entire field as the convergence criterion, and the convergence threshold is preferably set to no more than 2%, which can be adjusted according to the physical characteristics and accuracy requirements of different devices. S3. Define a unified decision-making cycle. Based on the converged, homogeneous simulation data, execute the following two types of tasks in parallel and synchronously within the unified decision-making cycle: First, perform equipment fault diagnosis, the diagnostic results of which can be used to further guide the optimization of equipment operating parameters. Second, perform simulation computing power prediction. The fault diagnosis employs the two fault classification and identification mechanisms defined in claim 6 in parallel; the computing power prediction follows the model-based scheduling strategy generation process described in claim 4, preferably using the neural network model of claim 5. See Figure 3 for the specific operation timing. S4. Only when both the fault diagnosis task and the computing power prediction task generate valid output content will the two types of output be integrated into a standardized control instruction set and synchronously transmitted back to the device-side control system and computing power scheduling platform. This step sets up null value interception logic: if a field in the control instruction set is empty or parsing fails, the instruction set assembly and distribution will not be executed; at the same time, the system records error codes and operation logs, and triggers a multi-level exception handling process including retry logic, hierarchical alarms, and manual intervention; all exception information is synchronously retained for backtracking and optimization of fault diagnosis and computing power prediction algorithms. The standardized control instruction set is transmitted using encrypted communication based on the TLS protocol. This instruction set includes three types of fixed fields using standardized encoding rules: operating condition labels, parameter instructions, and pre-scheduling; if the encrypted transmission link is abnormal, or any field is judged to be encoded incorrectly or missing during parsing, the entire control instruction set will be judged as invalid and will not be distributed, and the system will generate a power safety policy switching signal to trigger the device to switch to low power or safe standby mode until the next valid instruction is received. 1.2 System Architecture The present invention also provides a closed-loop digital twin simulation system for implementing the above method, which is composed of a data access module, a bidirectional coupling simulation module, a parallel processing module, and a synchronous backhaul module cascaded in sequence. The data access module receives equipment monitoring signals; the bidirectional coupling simulation module performs multi-physics coupling simulation and data convergence; the parallel processing module includes a fault diagnosis submodule and a computing power prediction submodule, simultaneously executing two types of core tasks; and the synchronous feedback module implements instruction encapsulation, anomaly interception, and instruction issuance. All modules are directly connected in one direction, forming an end-to-end closed-loop operating architecture. 1.3 Application Constraints This solution takes into account industrial functional safety requirements. In areas where relevant functional safety specifications must be followed (e.g., when applied to vehicle-mounted electronic control equipment), the output should be strictly limited to non-core safety functions such as thermal management optimization and perception redundancy verification, and should not generate direct control commands involving personal safety, such as vehicle braking and steering. 2. Beneficial effects Eliminating data silos and improving simulation accuracy. By using data from the same source as unified input and combining standardized simulation convergence rules, the relative average error across the entire simulation domain is controlled within 2%, reducing the error by more than 28% compared to traditional methods (see appendix). Figure 5 The stability and reliability of the simulation results have been greatly improved. Reduce response latency and adapt to real-time scenarios. The unified cycle parallel architecture abandons the serial mode, reducing the system end-to-end closed-loop latency from 420ms to 160ms (see appendix). Figure 5 The real-time performance is significantly enhanced. Dynamically allocate computing power to reduce resource waste. Intelligent dynamic scheduling is achieved based on task characteristics and historical data, increasing the average utilization rate of computing power from a baseline of 45% to 75%–79% (see appendix). Figure 5 This effectively reduces the loss of computing power during idle operation. Unified command standards reduce integration difficulty. Standardized fields and encrypted transmission rules reduce cross-platform integration workload by 80% (see appendix). Figure 5 This significantly improves the efficiency of large-scale equipment deployment. Improve the exception handling mechanism to enhance system robustness. Configure a full-link mechanism including null value interception, link anomaly detection, security mode switching, logging, automatic retry, and algorithm backtracking, further enhancing the system's fault tolerance and operational stability. Supplementary explanation: Based on the controlled experiments (refer to Comparative Example 1, Comparative Example 2, Comparative Example 3 and see appendix),... Figure 5 As can be seen, the three core constraints of this invention are interconnected and indispensable. If any one constraint is missing, the overall performance of the system will decline significantly. The combination of the three can achieve synergistic technical effects that cannot be achieved by a single technical means. Attached Figure Description Figure 1 shows the overall architecture of the three-in-one closed-loop collaborative system (converging and homogeneous simulation → unified decision-making cycle parallel tasks → atomic instruction issuance). Figure 2 is a flowchart of the dynamic boundary of the thermoelectric field + global error iterative convergence. Figure 3 is a schematic diagram of the timing of parallel tasks in the unified decision-making cycle; Figure 4 shows the complete logic diagram of atomic instructions, null value interception, hierarchical alarms, log alarms and retries; Figure 5 is a summary table comparing the system performance of the present invention with that of the traditional solution. Detailed Implementation Example 1: Power Battery Pack / Energy Storage Station Cell voltage and temperature monitoring signals are collected, multiphysics coupled simulation is performed, and boundary conditions are dynamically adjusted until the simulation model converges, generating converged homogeneous simulation data. Within a unified decision-making cycle, fault diagnosis and computing power prediction tasks are completed in parallel. In the fault stage, a convolutional long short-term memory network is used to identify aging faults, and a finite element impact response simulation model is used to analyze instantaneous thermal faults. In the computing power stage, feature vectors are constructed by extracting task and historical operation data, and a scheduling strategy is generated based on a radial basis function neural network. After both tasks output valid results, standardized instructions are synthesized and synchronously transmitted back to the device-side control system and computing power scheduling platform. If the encrypted transmission link is abnormal, or if instruction fields are parsed incorrectly or missing, the entire instruction is deemed invalid and not issued. The system automatically generates a power safety switching signal, controlling the device to enter a low-power or safe standby mode. Simultaneously, logs are recorded and operations such as retry, tiered alarms, and manual intervention are performed. Abnormal data is used for algorithm backtracking and optimization. This solution was tested on a simulation testing platform, and configuration and verification were strictly carried out in accordance with the three core constraints defined in this invention: "data homogeneity, parallel tasks, and valid instruction transmission." The above embodiments verify the operational effectiveness of the specific implementation methods of the present invention. To further verify the substantial features and significant technical advancements of the present invention, the following comparative experiments were conducted. Through rigorously designed comparative experiments (see Comparative Examples 1, 2, and 3 and Appendix Table 1), tests were conducted under the condition of breaking any one of the core constraints one by one, all of which resulted in a significant degradation of the overall system performance. As shown in Comparative Examples 1 and 2 in Appendix Table 1, the performance indicators of the corresponding dimension immediately deteriorate when any core constraint is missing; while the traditional scheme (Comparative Example 3) which simultaneously lacks all three constraints lags behind the performance indicators of the present invention in all aspects. This comparative experiment not only proves the necessity of the three core constraints, but also verifies the "1+1+1>3" nonlinear synergistic effect produced by their combination. Example 2: High and Low Voltage Power Equipment The system collects bus voltage and equipment housing temperature signals, runs electrothermal coupling simulation to generate convergent homogeneous data, and completes fault diagnosis and computing power prediction tasks in parallel within a unified decision-making cycle. The remaining operating logic and parameters are consistent with those in Example 1. Example 3: Rail Transit Equipment Switch to the sound field-vibration field coupling mode, generate convergent homogeneous data based on the linkage between sound pressure and vibration acceleration, complete the fault diagnosis task, and reuse the aforementioned scheme throughout the entire process logic. Example 4 Industrial Fluid / Heat Exchange Equipment Enable thermal-flow coupling, use the flow field calculation results as the dynamic convection boundary of the heat conduction equation to complete the simulation, and the subsequent fault diagnosis, computing power prediction and command issuance process remains unchanged. Examples 5-11 This solution is applied sequentially to semiconductor process equipment, high-end medical imaging equipment, aerospace power plants, offshore oil and gas platforms, computing server clusters, vehicle-mounted electronic control systems, and megawatt-level green hydrogen electrolysis plants. For vehicle-mounted electronic control systems, which involve driving safety, the output of this solution is only used for non-core safety purposes such as thermal management optimization and perception redundancy verification. It does not output direct control commands related to personal safety, such as braking and steering. The remaining technical implementation processes are the same as in Example 1. Comparative Example Comparative Example 1 Based on the traditional architecture, only the multiphysics simulation module was optimized, and the unique input constraint of data from the same source was removed. Fault diagnosis and computing power prediction tasks adopted multi-source data input. Test results showed that data inconsistency became prominent, fault diagnosis accuracy decreased, and the overall system performance significantly declined. Comparative Example 2 By retaining the constraint of shared data sources, but removing the unified decision-making cycle and parallel operation mechanism, fault diagnosis and computing power prediction tasks were changed to serial execution. Test results show that the waiting latency between modules increased, and the system's real-time performance could not meet industrial application requirements. Comparative Example 3 It adopts a traditional siloed architecture and lacks three core constraints. As the current mainstream benchmark solution in the industrial field, it suffers from a series of problems such as data silos, high latency, low computing power utilization, and high integration difficulty.

Claims

1. A closed-loop digital twin simulation method based on multi-physics signal fusion, characterized in that, Includes the following steps: S1, Receive monitoring signals from industrial equipment; S2. Update the digital twin model of the equipment according to the monitoring signal, perform multi-physics field coupling simulation and dynamically adjust the boundary conditions until the simulation model converges, and output the converged homogeneous simulation data. The converged homogeneous simulation data is the only exclusive input source for the downstream fault diagnosis task and computing power prediction task. S3. Define a unified decision-making cycle, and execute fault diagnosis and computing power prediction tasks in parallel and synchronously within the unified decision-making cycle based on the converged homogeneous simulation data. S4. Only when both the fault diagnosis task and the computing power prediction task generate valid output content will the two types of output be integrated into a standardized control instruction set and synchronously transmitted back to the device-side control system and computing power scheduling platform. Therefore, steps S2→S3 construct a data constraint chain with converged, homogeneous simulation data as the sole exclusive input to downstream tasks; steps S3→S4 construct a synchronization constraint chain with both tasks generating valid outputs as the instruction triggering condition. Through the synergistic effect of these two constraints, multiphysics simulation, fault diagnosis, and computing power scheduling are deeply bound together in terms of data flow and time window, forming a logically inseparable closed-loop collaborative process.

2. The method according to claim 1, characterized in that, In step S2, the multiphysics coupling simulation uses the global relative average error of the key physical quantities in the entire field as the convergence criterion, and the convergence threshold is set to no more than 2%.

3. The method according to claim 1, characterized in that, The multiphysics coupling includes any one of electrothermal coupling, acoustic-vibrational field coupling, and thermal-flow field coupling. Among them, acoustic-vibrational field coupling uses the sound pressure field distribution as an excitation applied to the vibrational field structure, or applies the structural vibration displacement and acceleration as boundary conditions to the sound pressure field, to achieve bidirectional data transmission and physical quantity linkage. Thermal-flow field coupling uses the pressure and velocity distribution calculated by flow field simulation as the dynamic convection boundary of the heat conduction equation.

4. The method according to claim 1, characterized in that, The computing power prediction task includes: extracting the attribute parameters of the current simulation task, historical simulation resource usage and running time data, and constructing a standardized multi-dimensional feature vector; based on this vector, generating a computing power scheduling strategy that includes containerized deployment specifications and computing power pre-occupancy time windows through machine learning or statistical regression models.

5. The method according to claim 4, characterized in that, The machine learning model is a radial basis function neural network model.

6. The method according to claim 1, characterized in that, The fault diagnosis task includes two fault classification and identification mechanisms: using convolutional long short-term memory networks to diagnose progressive aging faults in equipment, and using finite element impact response simulation models to analyze instantaneous sudden faults in equipment.

7. The method according to claim 1, step S4 includes null value interception logic: if a field in the control instruction set is empty or parsing fails, the instruction set assembly and distribution are not executed; at the same time, the system records error codes and running logs, and triggers a multi-level exception handling process including retry logic, hierarchical alarms, and manual intervention; all exception information is stored synchronously for fault diagnosis and backtracking optimization of computing power prediction algorithms.

8. The method according to claim 7, characterized in that, The standardized control command set is transmitted using encrypted communication based on the TLS protocol. This command set includes three fixed fields using standardized encoding rules: operating condition tags, parameter commands, and pre-scheduling. If the encrypted transmission link malfunctions, or if any field is determined to be encoded incorrectly or missing during parsing, the entire control command set is deemed invalid and not issued. The system then generates a power safety policy switching signal, triggering the device to switch to low-power or safe standby mode until the next valid command is received.

9. A closed-loop digital twin simulation system based on multi-physics signal fusion, characterized in that, include: A data access module is used to perform step S1 of the method according to any one of claims 1 to 8; A bidirectional coupled simulation module for performing step S2 of the method according to any one of claims 1 to 8; A parallel processing module for performing step S3 of the method according to any one of claims 1 to 8; A synchronous feedback module is used to perform step S4 of the method according to any one of claims 1 to 8. The output of the bidirectional coupling simulation module is directly connected to the input of the parallel processing module, and the output of the parallel processing module is connected to the input of the synchronous feedback module.

Images