A sintering machine operation and maintenance strategy simulation test and optimization method and system based on digital twinning

Abstract

The application discloses a sintering machine operation and maintenance strategy simulation test and optimization method and system based on digital twinning, relates to the technical field of sintering intelligent operation and maintenance, and comprises the following steps: a simulation environment is constructed; real-time operation data of equipment is acquired, the real-time operation data of the equipment is mapped into a current equipment state snapshot through a digital twinning platform, and the current equipment state snapshot is converted into an initial current state vector; an operation and maintenance strategy to be tested is packaged into a decision function module and dynamically linked to the simulation environment; a simulation engine of the simulation environment is called, each single-strategy simulation deduction cycle in the operation and maintenance strategy is executed step by step with the initial current state vector as a starting point; a state trajectory, an action sequence and an event log are recorded; and a single-strategy comprehensive performance score corresponding to each single strategy is calculated. The method solves the technical problems that in the prior art, an operation and maintenance strategy lacks simulation verification, is directly put into production, results in high trial-and-error cost, and strategy reliability cannot be quantitatively evaluated.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for sintering, and in particular to a simulation testing and optimization method and system for operation and maintenance strategies of sintering machines based on digital twins. Background Technology

[0002] Sintering machines are key pieces of equipment in steel production. They are mobile clusters of equipment consisting of hundreds of sintering machines, and their stable, efficient, and long-term operation is of paramount importance.

[0003] Currently, the operation and maintenance decisions of sintering machines (such as equipment maintenance cycles and process parameter settings) mainly rely on the following two modes: one is preventive maintenance based on fixed rules, which strictly follows fixed time or production cycles for maintenance and process inspection. This method is simple and easy to implement, but it lacks response to the actual condition of the equipment, which can easily lead to over-maintenance and waste of resources, or under-maintenance and sudden failures. The other is decision optimization based on expert experience, where experienced engineers adjust decisions based on real-time data and historical conditions. This method is effective to a certain extent, but it relies heavily on personal experience, is difficult to quantify, solidify and promote, and cannot cope with complex working conditions and conflicts of multiple objectives.

[0004] Research reveals that the fundamental flaw in existing technologies lies in the lack of a safe, efficient, and quantifiable verification environment before any new and more refined operation and maintenance (O&M) strategies (such as predictive maintenance based on equipment health status and multi-objective process optimization algorithms) can be effectively implemented. Directly testing on the production line is costly and risky, resulting in lengthy iteration cycles and difficulties in implementing advanced O&M strategies. Furthermore, most existing simulation technologies exist as independent systems, disconnected from the real-time operating status and full lifecycle historical data of physical equipment. Their models lack continuous calibration and evolution, leading to insufficient fidelity in the simulation environment and low reliability of the extrapolated results. Digital twin technology provides an ideal foundation for achieving a comprehensive mapping of physical entities in the information space; however, how to deeply utilize digital twins to construct a simulation testing environment with shared data sources, symbiotic models, and a closed-loop virtual-physical system, and specifically use it to solve the verification challenges of O&M strategies, remains a critical area that has not been fully explored. Therefore, there is an urgent need in this field for a technical means to conduct forward-looking and quantitative verification and optimization of operation and maintenance strategies without affecting production, so as to ultimately obtain maintenance strategies that meet the objectives under the verification of simulation environment and realize intelligent operation and maintenance of sintering machines. Summary of the Invention

[0005] The main objective of this invention is to provide a simulation testing and optimization method and system for sintering machine operation and maintenance strategies based on digital twins. This aims to solve the technical problem in the prior art where the current operation and maintenance strategy is generated based solely on data decisions, but without verification testing based on simulation, resulting in insufficient intelligent operation and maintenance of the sintering machine.

[0006] To achieve the above objectives, this invention provides a simulation testing and optimization method for sintering machine operation and maintenance strategies based on digital twins, comprising the following steps: S10, based on the historical data lake of the digital twin platform, adopts a mechanism and data fusion modeling method to construct equipment performance degradation model, process model, disturbance model and actuator model to form a simulation environment; S20, acquire real-time operating data of the device, map the real-time operating data of the device to a snapshot of the current device state through the digital twin platform, and convert the snapshot of the current device state into the initial current state vector S(0); S30 encapsulates the operation and maintenance strategy to be tested into a decision function module and dynamically links it to the simulation environment. The input of the decision function module is the current state vector S, which includes the current equipment health status, current process parameter distribution, current mileage, and current raw material composition status of all trolleys. The output of the decision function module is the decision action vector A, which includes the damper opening setpoint, maintenance instructions, and process parameter adjustment instructions. The simulation engine of the simulation environment is invoked, starting from the initial current state vector S(0), and the simulation deduction loop of each single strategy in the operation and maintenance strategy is executed step by step. Within each deduction step of the simulation deduction loop, the following actions are performed: read the current state vector S(t); call the equipment performance degradation model, decision function module, actuator model, process model and disturbance model to generate the simulation state vector S(t+Δt) for the next time step; repeat the above steps until the preset simulation duration is reached, and record the state trajectory, action sequence and event log. S40 calculates the economic, reliability, quality, and energy efficiency indicators corresponding to each single strategy based on state trajectories, action sequences, and event logs, and obtains the comprehensive performance score of the single strategy through a weighted comprehensive evaluation function.

[0007] Furthermore, it also includes the following steps: S50, determine the maintenance strategy corresponding to the optimal single-strategy comprehensive performance score as the ideal strategy; S60, the target policy is obtained by optimizing the policy parameters in the ideal policy using the Bayesian optimization method; S70, after executing the target strategy, updates the digital twin platform and simulation environment.

[0008] Furthermore, in step S10, The equipment performance degradation model uses the Archard wear model as its basic framework. It uses the wheel and axle replacement cycle and corresponding running mileage from historical maintenance records in the historical data lake as observations. The wear coefficient is fitted by nonlinear least squares method. The input of the equipment performance degradation model is the current equipment health, current running mileage, current machine speed, load and simulation step size. The output is the simulated running mileage and simulated equipment health at the next time step. The process model is modeled using the heat balance equation of the sintering material layer based on energy conservation. The effective thermal conductivity and reaction heat parameters are obtained by back-calculating historical process data. The input of the process model is the current process parameter distribution, the current machine speed, the simulated physical response state and the reaction heat boundary conditions. The output is the simulated process parameter distribution for the next time step. The perturbation model fits a Gaussian mixture model through statistical analysis of historical abnormal operating conditions. The input of the perturbation model is the current process parameters and the statistical distribution parameters of historical abnormal operating conditions, and the output is the perturbation process parameters after injecting random perturbation values. The actuator model obtains function parameters by fitting historical control commands and feedback data. The input of the actuator model is the decision action vector A, and the output is the simulated physical response state.

[0009] Furthermore, in step S20, the current equipment status snapshot includes the current equipment health, current process parameter matrix, current operating mileage, and current raw material composition calculated by the digital twin platform based on real-time equipment operating data. The initial current state vector S(0) includes the current equipment state field and the simulation control field. The current equipment state field is determined by the current equipment health, current process parameter matrix, current running mileage and current raw material composition in the current equipment state snapshot. The simulation control field is generated by the initialization of the simulation environment and includes the simulation time stamp, simulation action record queue and simulation cost accumulation value.

[0010] Further, in step S30, the input of the decision function module is the current state vector S, which includes the current equipment health, current process parameter distribution, current running mileage, and current raw material composition status of all trolleys; the output of the decision function module is the decision action vector A, which includes the damper opening setting value, maintenance instructions, and process parameter adjustment instructions.

[0011] Furthermore, in step S30, actions are performed within each iteration of the simulation iteration loop: Read the current state vector S(t). When t=0, the current state vector S(t) is the initial current state vector S(0). When t is greater than 0, the current state vector S(t) is the simulation state vector generated by the previous simulation step. The device performance degradation model is invoked to output the simulation mileage and simulation device health status for the next time step; Call the decision function module, take the current state vector S(t) containing the health status of the simulation equipment as input, and output the decision action vector A(t); The actuator model is invoked to convert the decision action vector A(t) into the simulated physical response state; Call the process model, take the current process parameter distribution, current machine speed, simulated actual physical response state and reaction thermal boundary conditions as input, and output the simulated process parameter distribution for the next time step; The perturbation model is invoked to inject random perturbation values ​​into the simulated process parameter distribution to generate perturbed process parameters; The simulation mileage, simulation equipment health, simulation physical response state, and post-disturbance process parameters are structurally spliced ​​together and, after verification by physical constraints, the simulation state vector S(t+Δt) for the next time step is generated. The simulation state vector S(t+Δt) is used as the current state vector S(t) for the next simulation step.

[0012] Furthermore, economic indicators include energy consumption costs, maintenance material costs, and downtime losses; reliability indicators include average equipment health and number of unplanned downtimes; quality indicators include sinter qualification rate; and energy efficiency indicators include energy consumption per ton of ore.

[0013] Furthermore, when using Bayesian optimization to automatically optimize the key parameters of an ideal strategy, the optimal parameters are found through a Gaussian process surrogate model and intelligent sampling iteration, with the single-strategy comprehensive performance function as the objective.

[0014] This invention also provides a simulation testing and optimization system for sintering machine operation and maintenance strategies based on digital twins. This includes a data access layer, used to obtain modeling data from the historical data lake of the digital twin platform, and to obtain current device status snapshots from real-time data streams; The model building layer is used to construct equipment performance degradation models, process models, disturbance models, and actuator models using a mechanistic and data fusion modeling method, thereby forming a simulation environment. The simulation inference layer is used to configure the test scenario set. The initial simulation state is used as the initial current state vector S(0) of the inference loop. The inference loop is executed step by step to generate the simulation state vector S(t+Δt) of the next time step, and the state trajectory, action sequence and event log are recorded. The evaluation and optimization layer calculates the economic, reliability, quality, and energy efficiency indicators corresponding to each individual strategy based on state trajectories, action sequences, and event logs. The overall performance score of each individual strategy is then calculated using a weighted comprehensive evaluation function.

[0015] Compared with existing technologies, the simulation testing and optimization method for sintering machine operation and maintenance strategies based on digital twins provided by this invention has the following beneficial effects: This invention provides a simulation testing and optimization method for sintering machine operation and maintenance strategies based on digital twins. It constructs four types of simulation models—equipment performance degradation model, process model, disturbance model, and actuator model—based on the historical data lake of the digital twin platform, forming a simulation environment. Then, based on real-time equipment operation data, the real-time state of the physical equipment is mapped to an initial current state vector S(0), which serves as the starting point for simulation deduction. Next, the operation and maintenance strategy to be tested is encapsulated into a decision function module, driving the simulation engine to iteratively call the degradation model, strategy module, actuator model, process model, and disturbance model step-by-step, generating complete state trajectories, action sequences, and event logs. Finally, based on the above records, multi-dimensional performance indicators are calculated and weighted scores are applied, thus establishing a complete technical chain of physical state synchronization, strategy simulation deduction, and quantitative performance evaluation. This differs from existing technologies that only generate strategies offline based on historical data and then directly implement them. This invention addresses the black-box decision-making model for production deployment by utilizing a digital twin platform to achieve real-time mapping from physical equipment status to the simulation environment. It ensures the extrapolation credibility and physical interpretability of simulation deductions through a mechanistic model, improves simulation accuracy through data-driven model calibration, and tests the robustness of strategies by injecting random uncertainties into the perturbation model. This allows strategy verification to be completed in a zero-risk, quantifiable, and reproducible virtual environment, forming a closed-loop verification principle of data homogeneity, model symbiosis, and virtual-real interaction. This solves the technical problems in existing technologies where the lack of simulation verification for operation and maintenance strategies before production leads to high trial-and-error costs, inability to quantify strategy reliability, and difficulty in predicting multi-objective conflicts. It achieves safe sandbox verification and quantitative optimization of operation and maintenance strategies before practical application, reducing decision-making risks in intelligent operation and maintenance of sintering machines, improving the efficiency and credibility of strategy iteration, and promoting operation and maintenance decision-making from experience-based to data-driven scientific decision-making. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a simulation test and optimization method for sintering machine operation and maintenance strategy based on digital twins, according to one embodiment of the present invention.

[0018] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0022] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0023] Please refer to the appendix. Figure 1 This invention provides a simulation testing and optimization method for sintering machine operation and maintenance strategy based on digital twins. S10, based on the historical data lake of the digital twin platform, a mechanism and data fusion modeling method is adopted to construct equipment performance degradation model, process model, disturbance model and actuator model to form a simulation environment. S20, acquire real-time operating data of the device, map the real-time operating data of the device to a snapshot of the current device state through the digital twin platform, and convert the snapshot of the current device state into the initial current state vector S(0); S30 encapsulates the operation and maintenance strategy to be tested into a decision function module and dynamically links it to the simulation environment. The input of the decision function module is the current state vector S, which includes the current equipment health status, current process parameter distribution, current mileage, and current raw material composition status of all trolleys. The output of the decision function module is the decision action vector A, which includes the damper opening setpoint, maintenance instructions, and process parameter adjustment instructions. The simulation engine of the simulation environment is invoked, starting from the initial current state vector S(0), and the simulation deduction loop of each single strategy in the operation and maintenance strategy is executed step by step. Within each deduction step of the simulation deduction loop, the following actions are performed: read the current state vector S(t); call the equipment performance degradation model, decision function module, actuator model, process model and disturbance model to generate the simulation state vector S(t+Δt) for the next time step; repeat the above steps until the preset simulation duration is reached, and record the state trajectory, action sequence and event log. S40 calculates the economic, reliability, quality, and energy efficiency indicators corresponding to each single strategy based on state trajectories, action sequences, and event logs, and obtains the comprehensive performance score of the single strategy through a weighted comprehensive evaluation function.

[0024] This invention provides a simulation testing and optimization method for sintering machine operation and maintenance strategies based on digital twins. It constructs four types of simulation models—equipment performance degradation model, process model, disturbance model, and actuator model—based on the historical data lake of the digital twin platform, forming a simulation environment. Then, based on real-time equipment operation data, the real-time state of the physical equipment is mapped to an initial current state vector S(0), which serves as the starting point for simulation deduction. Next, the operation and maintenance strategy to be tested is encapsulated into a decision function module, driving the simulation engine to iteratively call the degradation model, strategy module, actuator model, process model, and disturbance model step-by-step, generating complete state trajectories, action sequences, and event logs. Finally, multi-dimensional performance indicators are calculated and weighted scores are applied based on these records, thus establishing a complete technical chain of physical state synchronization, strategy simulation deduction, and quantitative performance evaluation. This differs from existing technologies that directly implement strategies based solely on offline generation of historical data. This invention, a black-box decision-making model for production, utilizes a digital twin platform to achieve real-time mapping from physical equipment status to the simulation environment. It ensures the extrapolation credibility and physical interpretability of simulation deductions through a mechanistic model, improves simulation accuracy through data-driven model calibration, and tests the robustness of strategies by injecting random uncertainties into the perturbation model. This allows strategy verification to be completed in a zero-risk, quantifiable, and reproducible virtual environment, forming a closed-loop verification principle of data homogeneity, model symbiosis, and virtual-real interaction. This solves the technical problems in existing technologies where operation and maintenance strategies lack simulation verification before being put into production, leading to high trial-and-error costs, inability to quantify strategy reliability, and difficulty in predicting multi-objective conflicts. It achieves safe sandbox verification and quantitative optimization of operation and maintenance strategies before practical application, reducing the decision-making risk of intelligent operation and maintenance of sintering machines, improving the efficiency and credibility of strategy iteration, and promoting operation and maintenance decision-making from experience-based to data-driven scientific decision-making.

[0025] Understandably, a digital twin platform is a comprehensive virtual platform with physical information neural networks and physical mechanism models as its core and geometric models as its form carrier. Based on the simulation and testing framework enabled by digital twins, the digital twin is upgraded from a static mapping model to a dynamic simulation engine. Through the deep integration of five levels—data, model, closed loop, interaction, and mode—it provides a high-fidelity, quantifiable, and zero-risk testing environment for operation and maintenance strategies.

[0026] Understandably, the simulation environment and the digital twin platform constitute a collaborative architecture of data homogeneity, model symbiosis, and virtual-real interaction. Based on the technical framework of this invention, data homogeneity drives the simulation environment to directly access the real-time data stream and historical data lake of the digital twin (digital twin platform), ensuring that the initial state of the simulation is synchronized with the physical entity and that historical patterns are homogeneous with the real world. Model symbiotic evolution occurs when the core simulation model (equipment, process) inherits and deepens the existing model in the digital twin, employing a mechanism and data fusion approach for construction and calibration, and co-optimizing with the twin model. Closed-loop feedback optimization occurs when the simulation test results are fed back to the digital twin to optimize its model parameters, forming a closed loop of simulation-twin co-evolution. A virtual-real interaction sandbox is used where operation and maintenance strategies are executed in a virtual environment tightly coupled with the digital twin, and their perception and actions are highly realistically interacted and extrapolated through the twin model. A parallel system mode is constructed, creating a digital operation and maintenance parallel system parallel to the physical production line, where strategies are accelerated, compared, optimized, and validated in extreme scenarios.

[0027] Understandably, the preset simulation duration can be the duration of one time step or the duration of multiple time steps, depending on the actual situation.

[0028] Furthermore, in order to determine the target strategy for achieving automated operation and maintenance, the following steps are also included: S50, determine the maintenance strategy corresponding to the optimal single-strategy comprehensive performance score as the ideal strategy; S60, the target policy is obtained by optimizing the policy parameters in the ideal policy using the Bayesian optimization method; S70, after executing the target strategy, updates the digital twin platform and simulation environment.

[0029] Understandably, in one specific embodiment of the present invention, after the key process parameters are determined by the ideal strategy, the single strategy comprehensive performance score corresponding to the key process parameters is optimized based on the Bayesian optimization method, and the key process parameters are fine-tuned to obtain the target strategy.

[0030] Understandably, in step S20, mapping the real-time operating data of the equipment to a snapshot of the current equipment status includes: performing data cleaning, time-series alignment, and dimensional standardization on the multi-source heterogeneous physical monitoring parameters; uniformly converting the digital quantities of temperature, rotational speed, pressure, and component analysis into floating-point numerical sequences; using the equipment health assessment model of the digital twin platform to fuse and calculate the multi-source temperature data into a normalized health index; integrating the rotational speed encoder pulse signal and fusing it with historical accumulated data to obtain the operating mileage; and directly mapping the output of the component analyzer to the raw material component field. The step of converting the snapshot of the current equipment status into an initial current state vector S(0) includes: sorting the snapshot according to the predefined field order. The physical monitoring parameters are reorganized into a structured array, and the simulation environment-generated derivation control fields are added to form an initial state data structure that the simulation engine can recognize. In step S30, when the decision function module reads the current state vector S, it parses the required sub-vectors according to the preset interface protocol. After the strategy module outputs the decision action vector A, the actuator model converts the dimensionless control command into the actual control quantity within the range of the physical actuator. In step S60, the digital twin platform is updated after executing the target strategy, including decoding the simulation-optimized strategy parameters into set values ​​that can be executed by the physical device after safety boundary verification, and sending them to the actuator through the DCS system. In step S70, the strategy rules and optimized strategy parameters of the target strategy are decoded into physical control parameters. The decoding includes converting dimensionless decision action vector elements into actual control quantities within the physical range of the corresponding actuator, converting simulation-optimized threshold parameters into comparison benchmarks for sensor monitoring values, and converting timing control logic into ladder diagrams or function block instructions recognizable by the PLC. The physical control parameters are then subjected to safety boundary verification. After the verification is passed, the parameters are sent to the sintering machine actuator via the DCS system. The simulation data of the target strategy is fed back to the digital twin platform to update its equipment health assessment model and process prediction model parameters, and the model parameters of the simulation environment are updated synchronously, forming a closed loop of co-evolution between simulation testing and the digital twin.

[0031] Furthermore, in step S10, The equipment performance degradation model uses the Archard wear model as its basic framework. It uses the wheel and axle replacement cycle and corresponding running mileage from historical maintenance records in the historical data lake as observations. The wear coefficient is fitted by nonlinear least squares method. The input of the equipment performance degradation model is the current equipment health, current running mileage, current machine speed, load and simulation step size. The output is the simulated running mileage and simulated equipment health at the next time step. The process model is modeled using the heat balance equation of the sintering material layer based on energy conservation. The effective thermal conductivity and reaction heat parameters are obtained by back-calculating historical process data. The input of the process model is the current process parameter distribution, the current machine speed, the simulated physical response state and the reaction heat boundary conditions. The output is the simulated process parameter distribution for the next time step. The perturbation model fits a Gaussian mixture model through statistical analysis of historical abnormal operating conditions. The input of the perturbation model is the current process parameters and the statistical distribution parameters of historical abnormal operating conditions, and the output is the perturbation process parameters after injecting random perturbation values. The actuator model obtains function parameters by fitting historical control commands and feedback data. The input of the actuator model is the decision action vector A, and the output is the simulated physical response state. The simulation environment and the digital twin platform form a collaborative architecture with data from the same source, model symbiosis, and virtual-real interaction.

[0032] Understandably, sub-equipment health is used to assess the health status of various sub-components, including the health of the guardrail, grate bars, and wheel axles. The current equipment health is calculated by integrating multiple indicators for key components. Specifically, the general formula for calculating guardrail health is: ,in To measure the actual deformation, This represents the maximum allowable deformation. If the calculated result is negative, it is set to 0. The health status of other components can be defined in a similar manner, such as the health status of the grate bars. wheel and axle health Etc. In one specific embodiment, the health of the guardrail is as follows: the measured deformation is 0.8 mm, and the maximum allowable deformation is 5 mm. =0.84, converted to 84 points; Grate health: visual recognition shows a blockage area ratio of 2%. =0.98, converted to 98 points; among which, wheel and axle health: measured temperature 65℃ (threshold ≤80℃), vibration 1.2mm / s, substituted into the composite function =0.92, which translates to a current device health score of 92.

[0033] Furthermore, in step S20, the current equipment status snapshot includes the current equipment health, current process parameter matrix, current operating mileage, and current raw material composition calculated by the digital twin platform based on real-time equipment operating data. The initial current state vector S(0) includes the current equipment state field and the simulation control field. The current equipment state field is determined by the current equipment health, current process parameter matrix, current running mileage and current raw material composition in the current equipment state snapshot. The simulation control field is generated by the initialization of the simulation environment and includes the simulation time stamp, simulation action record queue and simulation cost accumulation value.

[0034] In specific implementation, real-time equipment operation data is acquired; through the equipment health assessment model and process mapping model of the digital twin platform, the real-time equipment operation data is calculated and converted into a current equipment status snapshot. The current equipment status snapshot includes the trolley health, process parameter distribution, running mileage and raw material composition status normalized by the model; the current equipment status snapshot is copied into the physical state field of the initial current state vector S(0) according to the field correspondence relationship, and the simulation environment adds the simulation control field to form a complete simulation simulation starting point.

[0035] Further, in step S30, the input of the decision function module is the current state vector S, which includes the current equipment health, current process parameter distribution, current running mileage, and current raw material composition status of all trolleys; the output of the decision function module is the decision action vector A, which includes the damper opening setting value, maintenance instructions, and process parameter adjustment instructions.

[0036] Furthermore, in step S30, actions are performed within each iteration of the simulation iteration loop: Read the current state vector S(t). When t=0, the current state vector S(t) is the initial current state vector S(0). When t is greater than 0, the current state vector S(t) is the simulation state vector generated by the previous simulation step. The device performance degradation model is invoked to output the simulation mileage and simulation device health status for the next time step; Call the decision function module, take the current state vector S(t) containing the health status of the simulation equipment as input, and output the decision action vector A(t); The actuator model is invoked to convert the decision action vector A(t) into the simulated physical response state; (taking the decision action vector A(t) and the current actuator state as input, the simulated actual physical response state at the next time step is output;) Call the process model, take the current process parameter distribution, current machine speed, simulated actual physical response state and reaction thermal boundary conditions as input, and output the simulated process parameter distribution for the next time step; The perturbation model is invoked to inject random perturbation values ​​into the simulated process parameter distribution to generate perturbed process parameters; The simulation mileage, simulation equipment health, simulation physical response state, and post-disturbance process parameters are structurally spliced ​​together and, after verification by physical constraints, the simulation state vector S(t+Δt) for the next time step is generated. The simulation state vector S(t+Δt) is used as the current state vector S(t) for the next simulation step.

[0037] Understandably, in the solution of the present invention, the current state vector S(0) is obtained; the decision function module π(S) generates the current decision action A(t) based on the current state vector S(0); the actuator model response obtains the actual physical response; and the next state vector is determined based on the process, disturbance and degradation model. Furthermore, economic indicators include energy consumption costs, maintenance material costs, and downtime losses; reliability indicators include average equipment health and number of unplanned downtimes; quality indicators include sinter qualification rate; and energy efficiency indicators include energy consumption per ton of ore.

[0038] Furthermore, when using Bayesian optimization to automatically optimize the key parameters of an ideal strategy, the optimal parameters are found through a Gaussian process surrogate model and intelligent sampling iteration, with the single-strategy comprehensive performance function as the objective.

[0039] Furthermore, historical operation and maintenance records of the sintering machine and expert experience rules were collected to extract equipment maintenance cycle, health threshold and process parameter adjustment rules as basic strategy templates; Based on equipment failure modes and maintenance effect data in the historical data lake, an association rule mining algorithm is used to analyze the causal relationship between different maintenance actions and equipment health recovery, and a set of candidate strategies is generated. The feasibility of the candidate strategy set is screened, and strategies that exceed the physical constraints of the equipment and the process safety boundaries are eliminated to obtain the operation and maintenance strategy to be tested.

[0040] Understandably, in practical implementation, the construction of a high-fidelity dynamic simulation environment in step S10 specifically includes: To build the data foundation, the simulation environment's data comes from two levels: historical data lake and real-time data stream. The historical data lake is a complete production data of no less than 12 months extracted from the enterprise's data warehouse / historical database, including second-level / minute-level process parameters (air box temperature, negative pressure, machine speed, etc.) recorded by the DCS system, production events and quality data recorded by the MES system, and inspection and maintenance records. The real-time data stream includes the current equipment status snapshot (such as trolley health) provided by the digital twin platform in real time through the data bus during the simulation test, which serves as the initial state of the simulation (initial current state vector S(0)). Data preprocessing is performed, and the data used must meet the requirements of temporal integrity and field alignment. Data cleaning is required, including handling invalid and outlier values, and Z-score standardization of key process variables to eliminate the influence of dimensions, ultimately forming a structured time-series dataset for model training and calibration.

[0041] Mechanism and data fusion modeling; constructing an equipment performance degradation model. The inputs to the equipment performance degradation model are the current equipment health, current operating mileage, current machine speed, load, and preset extrapolation step size. The outputs are the estimated operating mileage and estimated equipment health for the next time step. Optionally, for trolley wheel axle wear, the Archard wear model is used as the basic framework, with the function expression V=(K W L) / H, where V is the wear amount, W is the load, L is the sliding distance (proportional to the running mileage), K is the wear coefficient, and H is the material hardness. The parameters are fitted using the component replacement cycle and running mileage recorded in the historical maintenance records, and used as the observed values ​​for the model output. When fitting the parameters, the unknown coefficient K in the Archard model is fitted using the nonlinear least squares method to make the wear life distribution predicted by the model consistent with the historical statistical distribution. For grate clogging, a multiple linear regression model can be established with factors such as clogging rate and running mileage, sinter basicity, etc., and the coefficients are fitted using historical data.

[0042] The process model takes the current process parameter distribution, current machine speed, actual actuator state, and reaction heat boundary conditions as inputs and outputs the estimated process parameter changes for the next time step. Specifically, a lumped parameter model is used to model the sintering process. For example, based on energy conservation, a one-dimensional heat balance equation for the sintering material layer is established using temperature, time, material layer height direction, machine speed, and reaction exothermics. When fitting parameters, temperature and negative pressure data from different times and different wind boxes in historical data are used, and a system identification method (such as parameter estimation based on gradient descent) is used to inversely calculate key parameters such as effective thermal conductivity and reaction heat in the equation, so that the model output curve matches the actual historical process curve.

[0043] The invention comprises two models: a disturbance model and an actuator model. The disturbance model is constructed using statistical analysis of historical abnormal operating conditions to create a Gaussian mixture model. During the prediction phase, the input to the disturbance model is a random seed, and the output is the estimated random disturbance value for the next time step. In practice, statistical analysis is performed on abnormal operating condition segments in historical data to fit a Gaussian mixture model of key process parameters (such as raw material moisture content). Random disturbances are injected into the simulation using Monte Carlo sampling. The actuator model obtains transfer function parameters by fitting historical control commands and feedback data. During the prediction phase, the input to the actuator model is a decision action vector, and the output is the simulated physical response state for the next time step. In practice, a first-order inertial plus pure time delay model is established to simulate the dynamic response of actuators (such as dampers) by analyzing the time-series data of control commands and valve opening feedback in the DCS. The actuator model's gain, time constant, and time delay are fitted using historical data. The key to this invention is that it does not rely on a pure black-box model but combines white-box mechanisms and data-driven methods to construct an interpretable and extrapolable simulation model (simulation environment).

[0044] Understandably, in practical implementation, the encapsulation of operation and maintenance strategies and the comprehensive performance score of single strategies specifically include: During strategy encapsulation, the operation and maintenance strategies to be tested are uniformly encapsulated into software modules that conform to preset interface specifications. The core of the software module is a decision function π(S)->A, where S is the current state vector S(t) read from the simulation environment (such as the health of all trolleys, the current process parameter matrix), and A is the output decision action vector (such as the damper opening setpoint, maintenance instructions). During dynamic loading and interaction, the simulation platform (simulation environment) provides a standard function call interface. During simulation initialization, the encapsulated strategy module is dynamically linked to the simulation core. At each step of the simulation run, the simulation engine passes the current state S to the π(S) function of the strategy module and receives the decision action vector A returned by it, thereby driving the simulation environment to evolve to the next state. When conducting accelerated simulations across multiple scenarios, multiple single-strategy combinations are defined to form a series of test scenarios, including baseline operating conditions, equipment performance degradation scenarios (such as pre-setting a batch of trolleys to enter the end of their lifespan in the simulation, causing their health to decline rapidly), and strong process disturbance scenarios (such as simulating large fluctuations in raw material composition), in order to test the adaptability and robustness of the strategies. During accelerated execution, the simulation engine executes the simulation at a speed much faster than real-time (for example, simulating one month of operation only takes a few minutes) and records all key process data throughout the process.

[0045] A quantitative evaluation method for comprehensive strategy effectiveness is used to calculate the comprehensive effectiveness score of a single strategy. After the simulation, multi-dimensional effectiveness indicators are calculated based on the recorded data. Among them, the economic indicator is used to evaluate the total operation and maintenance cost, which is calculated as: total operation and maintenance cost = energy consumption cost + maintenance material cost + downtime loss. The reliability indicator is used to evaluate the average health of the equipment and the number of unplanned downtimes. The quality indicator is obtained based on the sintered ore qualification rate, and the energy efficiency indicator is obtained based on the energy consumption per ton of ore. In actual operation, the comprehensive effectiveness score of a single strategy is obtained by weighted summation based on the indicator weights of each indicator. The indicator weights of each indicator can be set according to the enterprise's objectives.

[0046] Understandably, in specific implementation, step S60 involves determining the optimal single-strategy comprehensive performance score based on strategy comparison and selection, including the following steps: During comparative analysis, the comprehensive performance scores J of different strategies in the same scenario set are compared, and a radar chart visualization tool is used to display the advantages and disadvantages of each strategy in different sub-indicators; robustness ranking involves calculating the variance of the comprehensive score J of each strategy under different disturbance scenarios, and prioritizing strategies with smaller variances. When optimizing parameters, key parameters within the strategy (such as the health alarm threshold) are used as optimization variables, and a Bayesian optimization algorithm is used for automatic optimization in the simulation environment. This algorithm aims at the comprehensive performance function J(θ) and, through intelligent sampling, finds the optimal parameters that maximize J(θ).

[0047] In one optional embodiment of the present invention, the testing and optimization of the lubrication strategy for the trolley axles of a 450m² sintering machine are specifically described using an existing digital twin platform built on a 450m² sintering machine. The digital twin platform already possesses functions for trolley lifecycle tracking, equipment health assessment, and process data archiving. The steps are as follows: Step 1: Directly access the historical database of the digital twin platform to extract the DCS operation data and maintenance work orders (recording wheel and axle replacement information) of the sintering machine for the past 15 months; based on the existing trolley wear analysis module in the digital twin, establish a wheel and axle wear model based on the Archard wear model, use the replacement cycle in the maintenance work order, and fit the wear coefficient K using the nonlinear least squares method. The calibrated model can accurately reflect the wear life under different loads. Construct an equipment performance degradation model, assuming that the process model, disturbance model and actuator model have no coupling effect on the current environment.

[0048] Step two, strategy encapsulation and injection, to achieve virtual-real interaction; the operation and maintenance strategy is encapsulated as a decision function module. The simulation engine obtains the real-time state S from the digital twin, and the action A output by the strategy is executed in the virtual environment composed of the twin model. Among them, the operation and maintenance strategy includes strategy A, π(S) -> apply grease every 500 kilometers of running mileage, and strategy B (new regulation), π(S) -> apply grease if the wheel and axle health (calculated based on the wear model) < 88%.

[0049] Step 3: Conduct simulation and evaluation. Set up test scenarios that include normal load and periodic overload, accelerate the simulation for 180 days. The results show that, under the premise of maintaining the same equipment reliability, strategy B reduces grease consumption by 28% and has a higher overall performance score J than strategy A. Therefore, strategy B is determined as the target strategy.

[0050] Step four involves optimizing and feeding back the parameters in Strategy B to achieve closed-loop feedback. Bayesian optimization is used to find the optimal parameters for Strategy B. Data generated during optimization is fed back to the digital twin to calibrate its health assessment model, making its predictions of future states more accurate. Specifically, Bayesian optimization is performed on the health threshold parameter θ of Strategy B. It is found that the comprehensive performance score J reaches its peak at θ=85%, further reducing oil consumption by 3% compared to the threshold of 88% while maintaining reliability. The final recommended updated strategy with θ=85% is output, along with a detailed simulation benefit report.

[0051] This invention also provides a simulation testing and optimization system for sintering machine operation and maintenance strategies based on digital twins. This includes a data access layer, used to obtain modeling data from the historical data lake of the digital twin platform, and to obtain current device status snapshots from real-time data streams; The model building layer is used to construct equipment performance degradation models, process models, disturbance models, and actuator models using a mechanistic and data fusion modeling method, thereby forming a simulation environment. The simulation inference layer is used to configure the test scenario set. The initial simulation state is used as the initial current state vector S(0) of the inference loop. The inference loop is executed step by step to generate the simulation state vector S(t+Δt) of the next time step, and the state trajectory, action sequence and event log are recorded. The evaluation and optimization layer calculates the economic, reliability, quality, and energy efficiency indicators corresponding to each individual strategy based on state trajectories, action sequences, and event logs. The overall performance score of each individual strategy is then calculated using a weighted comprehensive evaluation function.

[0052] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A simulation testing and optimization method for sintering machine operation and maintenance strategies based on digital twins, characterized in that, Includes the following steps: S10, based on the historical data lake of the digital twin platform, adopts a mechanism and data fusion modeling method to construct equipment performance degradation model, process model, disturbance model and actuator model to form a simulation environment; S20, acquire real-time operating data of the device, map the real-time operating data of the device to a current device state snapshot through the digital twin platform, and convert the current device state snapshot into an initial current state vector S(0); S30, the operation and maintenance strategy to be tested is encapsulated into a decision function module and dynamically linked to the simulation environment; the input of the decision function module is the current state vector S, which includes the current equipment health, current process parameter distribution, current running mileage and current raw material composition status of all trolleys; the output of the decision function module is the decision action vector A, which includes the damper opening setting value, maintenance command and process parameter adjustment command. The simulation engine of the simulation environment is invoked, and starting from the initial current state vector S(0), the simulation deduction loop of each single strategy in the operation and maintenance strategy is executed step by step. Within each deduction step of the simulation deduction loop, the following actions are performed: read the current state vector S(t); invoke the equipment performance degradation model, the decision function module, the actuator model, the process model and the disturbance model to generate the simulation state vector S(t+Δt) for the next time step. Repeat the above steps until the preset simulation duration is reached, and record the state trajectory, action sequence, and event log. S40, based on the state trajectory, action sequence and event log, calculate the economic indicators, reliability indicators, quality indicators and energy efficiency indicators corresponding to each single strategy, and calculate the comprehensive performance score of the single strategy through the weighted comprehensive evaluation function.

2. The simulation testing and optimization method for sintering machine operation and maintenance strategy based on digital twin as described in claim 1, characterized in that, It also includes the following steps: S50, determine the maintenance strategy corresponding to the optimal single-strategy comprehensive performance score as the ideal strategy; S60, the target policy is obtained by optimizing the policy parameters in the ideal policy using the Bayesian optimization method; S70, after executing the target strategy, update the digital twin platform and the simulation environment.

3. The simulation testing and optimization method for sintering machine operation and maintenance strategy based on digital twin as described in claim 1, characterized in that, In step S10, The equipment performance degradation model uses the Archard wear model as its basic framework. It uses the wheel and axle replacement cycle and corresponding running mileage from the historical maintenance records in the historical data lake as observations. The wear coefficient is fitted by the nonlinear least squares method. The input of the equipment performance degradation model is the current equipment health, current running mileage, current machine speed, load and simulation step size. The output is the simulated running mileage and simulated equipment health at the next time step. The process model is modeled using the heat balance equation of the sintering material layer based on energy conservation. The effective thermal conductivity and reaction heat parameters are obtained by back-calculating historical process data. The input of the process model is the current process parameter distribution, the current machine speed, the simulated physical response state and the reaction heat boundary conditions. The output is the simulated process parameter distribution for the next time step. The disturbance model fits a Gaussian mixture model by statistical analysis of historical abnormal operating conditions. The input of the disturbance model is the current process parameters and the statistical distribution parameters of historical abnormal operating conditions, and the output is the process parameters after disturbance by injecting random disturbance values. The actuator model obtains function parameters by fitting historical control commands and feedback data. The input of the actuator model is the decision action vector A, and the output is the simulated physical response state.

4. The simulation testing and optimization method for sintering machine operation and maintenance strategy based on digital twin as described in claim 1, characterized in that, In step S20, the current equipment status snapshot includes the current equipment health, current process parameter matrix, current operating mileage, and current raw material composition calculated by the digital twin platform based on the real-time operating data of the equipment. The initial current state vector S(0) includes a current equipment state field and a simulation control field; the current equipment state field is determined by the current equipment health, current process parameter matrix, current running mileage and current raw material composition in the current equipment state snapshot; the simulation control field is generated by the initialization of the simulation environment, and the simulation control field includes a simulation time stamp, a simulation action record queue and a cumulative simulation cost value.

5. The simulation testing and optimization method for sintering machine operation and maintenance strategy based on digital twin as described in claim 1, characterized in that, In step S30, the input of the decision function module is the current state vector S, which includes the current equipment health, current process parameter distribution, current running mileage, and current raw material composition status of all trolleys; the output of the decision function module is the decision action vector A, which includes the damper opening setting value, maintenance instructions, and process parameter adjustment instructions.

6. The simulation testing and optimization method for sintering machine operation and maintenance strategy based on digital twin as described in claim 1, characterized in that, In step S30, the following actions are performed within each step of the simulation deduction cycle: Read the current state vector S(t). When t=0, the current state vector S(t) is the initial current state vector S(0). When t is greater than 0, the current state vector S(t) is the simulation state vector generated by the previous simulation step. The device performance degradation model is invoked to output the simulation mileage and simulation device health status for the next time step; The decision function module is invoked, taking the current state vector S(t) containing the health status of the simulation device as input, and outputting the decision action vector A(t); The actuator model is invoked to convert the decision action vector A(t) into a simulated physical response state; The process model is invoked, and the current process parameter distribution, current machine speed, simulated actual physical response state and reaction thermal boundary conditions are used as inputs to output the simulated process parameter distribution for the next time step. The perturbation model is invoked to inject random perturbation values ​​into the simulated process parameter distribution to generate perturbed process parameters; The simulation mileage, simulation equipment health, simulation physical response state, and disturbance-induced process parameters are structurally spliced ​​together and, after verification by physical constraints, the simulation state vector S(t+Δt) for the next time step is generated. The simulation state vector S(t+Δt) is used as the current state vector S(t) for the next simulation step.

7. The simulation testing and optimization method for sintering machine operation and maintenance strategy based on digital twin as described in claim 1, characterized in that, The economic indicators include energy consumption cost, maintenance material cost, and downtime losses; the reliability indicators include average equipment health and number of unplanned downtimes; the quality indicators include sintered ore qualification rate; and the energy efficiency indicators include energy consumption per ton of ore.

8. The simulation testing and optimization method for sintering machine operation and maintenance strategy based on digital twins according to any one of claims 2 to 7, characterized in that, When using Bayesian optimization to automatically optimize the key parameters of an ideal strategy, the optimal parameters are found by using a Gaussian process surrogate model and intelligent sampling iteration, with the single-strategy comprehensive performance function as the objective.

9. A simulation testing and optimization system for sintering machine operation and maintenance strategies based on digital twins, characterized in that, This includes a data access layer, used to obtain modeling data from the historical data lake of the digital twin platform, and to obtain current device status snapshots from real-time data streams; The model building layer is used to construct equipment performance degradation models, process models, disturbance models, and actuator models using a mechanistic and data fusion modeling method, thereby forming a simulation environment. The simulation inference layer is used to configure the test scenario set. The initial simulation state is used as the initial current state vector S(0) of the inference loop. The inference loop is executed step by step to generate the simulation state vector S(t+Δt) of the next time step, and the state trajectory, action sequence and event log are recorded. The evaluation and optimization layer calculates the economic, reliability, quality, and energy efficiency indicators corresponding to each single strategy based on the state trajectory, action sequence, and event log. The comprehensive performance score of the single strategy is then calculated using a weighted comprehensive evaluation function.

Images