Digital Twin Intelligent Monitoring System and Fault Prediction Method for Coal Preparation Plant

Abstract

This invention relates to the field of next-generation information technology, and in particular to a digital twin intelligent monitoring system and fault prediction method for a coal preparation plant. The system includes a process data acquisition module, a digital twin monitoring module, a consistency arbitration module, and a closed-loop verification module. The system collects data from the entire process through the process data acquisition module. The system also constructs a mechanistic twin monitoring model and a data-driven monitoring model through the digital twin monitoring module. The system further integrates fault judgments from the mechanistic twin monitoring model and the data-driven monitoring model through the consistency arbitration module. The system also performs on-site verification through the closed-loop verification module, thereby improving the accuracy, timeliness, and reliability of fault prediction.

Description

Technical Field

[0001] This invention relates to the field of next-generation information technology, and in particular to a digital twin intelligent monitoring system and fault prediction method for the entire process of coal preparation plants. Background Technology

[0002] In the intelligent transformation of coal preparation plants, traditional equipment monitoring and fault prediction methods mainly rely on two independent technical routes. One is modeling and simulation based on physical mechanisms, which has the advantage of providing a theoretical benchmark for equipment health, but it is often difficult to model complex systems and is not sensitive to new or early faults. The other is machine learning methods driven by historical data, which are good at identifying complex patterns from massive amounts of data, but heavily depend on the quality and quantity of data, and the results lack physical interpretability. These two technical routes are usually applied in isolation, forming a disconnect between "mechanism models that are not grounded in reality and data models that do not see the physical world". In addition, existing systems generally lack the ability to automatically and in a closed loop feed back the field verification results to the model and continuously optimize it, making the model performance easy to solidify and unable to adapt to equipment aging and changes in operating conditions, thus restricting the actual effect and reliability of predictive maintenance.

[0003] Chinese Patent Publication No. CN119356930A discloses an online diagnostic method, device, equipment, and storage medium for a coal preparation plant system. The method includes: invoking any preset data judgment script to determine faults in system data on a data platform, where the system data is obtained by performing digital twin calculations on the coal preparation plant system using a digital twin computing model; if the judgment result of the data judgment script indicates the existence of a fault, analyzing at least one fault cause and corresponding handling opinions for each fault cause based on the fault phenomenon and a pre-built fault cause knowledge graph. It is evident that this solution still suffers from problems such as a single fault prediction method, a lack of dual fault prediction and consistency arbitration through a mechanistic twin monitoring model and a data-driven monitoring model, and insufficient accuracy and reliability of fault prediction due to the lack of closed-loop verification. Summary of the Invention

[0004] To address these issues, this invention provides a digital twin intelligent monitoring system and fault prediction method for the entire process of a coal preparation plant. This system overcomes the problems of existing technologies, such as the single fault prediction method, the lack of dual fault prediction and consistency arbitration based on both mechanistic twin monitoring models and data-driven monitoring models, and the lack of closed-loop verification leading to insufficient accuracy and reliability in fault prediction.

[0005] To achieve the above objectives, on the one hand, the present invention provides a digital twin intelligent monitoring system for the entire process of a coal preparation plant, the system comprising: The process data acquisition module is used to collect data from the entire process. The digital twin monitoring module is used to construct the mechanism twin monitoring model and the data-driven monitoring model, and to acquire the mechanism residual value and fault prediction results. It is also used to calibrate the parameters of the mechanism twin monitoring model based on the mechanism residual value, and to perform convergence processing on the parameter calibration process. Furthermore, it is used to drive early warning based on the fault prediction results. The consistency arbitration module is used to perform consistency arbitration on the mechanism twin monitoring model and the data-driven monitoring model, obtain the consistency arbitration result, and output the final fault prediction result based on the consistency arbitration result. The closed-loop verification module is used to perform closed-loop verification on the final fault prediction results, obtain the closed-loop verification results, and optimize the mechanism twin monitoring model and supplement the data-driven monitoring model based on the closed-loop verification results. The digital twin monitoring module includes: The model building unit is used to build mechanism twin monitoring models and data-driven monitoring models; The model prediction unit is used to obtain the mechanism residual value based on the full process data and the mechanism twin monitoring model, and to obtain the fault prediction result based on the full process data and the data-driven monitoring model. The model calibration unit is used to calibrate the parameters of the mechanism twin monitoring model based on the mechanism residual value, obtain the total number of parameter calibrations, perform convergence processing on the parameter calibration process based on the total number of parameter calibrations, and drive early warning based on the fault prediction results.

[0006] Furthermore, the model building unit constructs the mechanism twin monitoring model by: performing geometric modeling based on the target mechanism data to obtain a preliminary mechanism model, and setting physical law constraints on the preliminary mechanism model to obtain the mechanism twin monitoring model.

[0007] Furthermore, the model building unit constructs a data-driven monitoring model according to a data-driven monitoring model construction method, the data-driven monitoring model construction method including: Step B01: Divide the historical dataset into a 70% training set, a 20% validation set, and a 10% test set. Step B02: Initialize the parameters of the recurrent neural network model to obtain the initial neural network model; Step B03: Input the training set into the initial neural network model for forward propagation to obtain the initial prediction result, and calculate the loss of the initial prediction result using the cross-entropy function to obtain the loss value. Perform backpropagation based on the loss value, and update the parameters through the optimizer to obtain the trained neural network model. Step B04: Input the trained neural network model into the validation set, and then perform forward propagation again on the test set for the trained neural network model with the lowest loss value on the validation set to obtain the final prediction result. The trained neural network model with a final prediction accuracy greater than 95% is output as the data-driven monitoring model.

[0008] Furthermore, the model prediction unit obtains the mechanism residual value based on the full-process data and the mechanism twin monitoring model. Specifically, it obtains the actual component vibration yx and inputs the actual working condition parameters into the mechanism twin monitoring model to obtain the component vibration zx simulated by the mechanism twin monitoring model. Based on the actual component vibration yx and the component vibration zx simulated by the mechanism twin monitoring model, the mechanism residual value mx is calculated, and mx is set to [|(yx-zx)| / yx]×100%. The model prediction unit acquires the fault prediction results based on the full-process data and the data-driven monitoring model. It inputs the target operation data from the full-process data into the data-driven monitoring model to obtain the fault prediction results output by the data-driven monitoring model. The fault prediction results include the fault mode and the predicted fault probability Pg.

[0009] Further, the model calibration unit performs parameter calibration on the mechanism twin monitoring model based on the mechanism residual value, compares the mechanism residual value mx with the preset mechanism residual value mx0, judges the compliance level of the mechanism residual value based on the comparison result, and performs parameter calibration on the mechanism twin monitoring model based on the judgment result, wherein: When zy≤zy0, the model calibration unit determines that the compliance level of the mechanism residual value is met and does not perform parameter calibration on the mechanism twin monitoring model; When zy > zy0, the model calibration unit determines that the compliance level of the mechanism residual value is not met, and performs parameter calibration on the mechanism twin monitoring model.

[0010] Further, the model calibration unit acquires the total number of parameter calibrations, vp, compares the total number of parameter calibrations, vp, with a preset total number of parameter calibrations, vp0, judges the status of the total number of parameter calibrations based on the comparison result, and performs convergence processing on the parameter calibration process based on the judgment result, wherein: When vp≤vp0, the model calibration unit determines that the total number of parameter calibrations is normal and does not perform convergence processing on the parameter calibration process. When vp > vp0, the model calibration unit determines that the total number of parameter calibrations is abnormal and performs convergence processing on the parameter calibration process: the parameter calibration process of the preliminary mechanism twin monitoring model is forcibly converged, and the equipment components corresponding to the mechanism residual values ​​are pushed.

[0011] Furthermore, the model calibration unit drives early warning based on the fault prediction results, specifically by comparing the predicted fault probability Pg in the fault prediction results with the preset predicted fault probability Pg0, judging the state of the predicted fault probability based on the comparison results, and driving early warning based on the judgment results, wherein: When Pg≤Pg0, the model calibration unit determines the predicted fault probability state as normal and does not perform a drive warning; When Pg > Pg0, the model calibration unit determines that the predicted fault probability is abnormal and drives an early warning: it alerts maintenance personnel by flashing indicator lights and pushes the fault mode in the fault prediction results.

[0012] Furthermore, the consistency arbitration module performs consistency arbitration on the mechanism twin monitoring model and the data-driven monitoring model to obtain a consistency arbitration result, which includes consistent results and inconsistent results. When the consensus arbitration result is consistent, the fault prediction result will be output as the final fault prediction result. When the consistency arbitration result is inconsistent, a digital twin simulation is initiated to obtain the simulation result, and the simulation result is output as the final fault prediction result. When the consensus arbitration result is a special case, an investigative warning will be issued.

[0013] Furthermore, the closed-loop verification module performs closed-loop verification on the final fault prediction result. Specifically, the operation and maintenance personnel conduct on-site verification based on the final fault prediction result to obtain the actual fault result. The actual fault result is then compared with the final fault prediction result to obtain the closed-loop verification result. The closed-loop verification result includes prediction verification qualified and prediction verification unqualified. When the closed-loop verification result is that the prediction verification is qualified, no model optimization is performed on the mechanism twin monitoring model and no data supplementation is performed on the data-driven monitoring model. When the closed-loop verification result is that the prediction verification is unqualified, the mechanism twin monitoring model is optimized: the parameters of the mechanism twin monitoring model are fine-tuned according to the actual fault results to obtain the fine-tuned mechanism twin monitoring model, and the fine-tuned mechanism twin monitoring model is used as the mechanism twin monitoring model. The closed-loop verification module supplements the data-driven monitoring model by adding the actual fault results and their corresponding target operating data to the historical operating dataset and re-propagating the initial neural network model forward.

[0014] On the other hand, the present invention also provides a method for intelligent monitoring and fault prediction of the entire process of a coal preparation plant using a digital twin, the method comprising: Step S1: Collect data for the entire process; Step S2 involves constructing a mechanism twin monitoring model and a data-driven monitoring model, obtaining mechanism residual values ​​based on the full-process data and the mechanism twin monitoring model, obtaining fault prediction results based on the full-process data and the data-driven monitoring model, calibrating the mechanism twin monitoring model based on the mechanism residual values, obtaining the total number of parameter calibrations, performing convergence processing on the parameter calibration process based on the total number of parameter calibrations, and driving early warning based on the fault prediction results. Step S3: Perform consistency arbitration on the mechanism twin monitoring model and the data-driven monitoring model to obtain the consistency arbitration result, and output the final fault prediction result based on the consistency arbitration result; Step S4: Perform closed-loop verification on the final fault prediction result to obtain the closed-loop verification result, and optimize the mechanism twin monitoring model and supplement the data-driven monitoring model based on the closed-loop verification result.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: The system collects full-process data through a process data acquisition module. Furthermore, the system constructs a mechanistic twin monitoring model and a data-driven monitoring model through a digital twin monitoring module, and acquires mechanistic residual values ​​and fault prediction results. This facilitates fault simulation and prediction from both physical twin and historical data experience levels, thereby improving the accuracy, timeliness, and reliability of fault prediction. The system also performs parameter calibration on the mechanistic twin monitoring model through the digital twin monitoring module, further calibrating the model. When the total number of parameter calibrations is abnormal, timely forced convergence is implemented to prevent untimely warnings of latent faults in the operating conditions, thus improving the robustness and deep perception capabilities of intelligent monitoring. When the predicted failure probability is abnormal, timely early warning is triggered to facilitate real-time monitoring of potential failure modes within a future timeframe. This allows maintenance personnel to perform early repairs, thereby improving the practicality and timeliness of intelligent monitoring and failure early warning. The system also integrates the failure judgments of the mechanistic twin monitoring model and the data-driven monitoring model through a consistency arbitration module, solving the problems of unreliability of single models and conflicting conclusions. This improves the accuracy and reliability of failure monitoring and early warning. Furthermore, the system uses a closed-loop verification module for on-site verification, comparing the final failure prediction results with the actual failure results. This allows for real-time monitoring of the accuracy of failure prediction and timely calibration of the mechanistic twin monitoring model and the data-driven monitoring model, further enhancing the accuracy of failure prediction. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of the digital twin intelligent monitoring system for the entire process of a coal preparation plant in this embodiment; Figure 2 This is a schematic diagram of the structure of the digital twin monitoring module in this embodiment; Figure 3 This is a flowchart illustrating the fault prediction method for the digital twin-based intelligent monitoring of the entire coal preparation plant process in this embodiment. Detailed Implementation

[0017] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0018] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0019] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0020] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0021] Please see Figure 1 As shown, this is a schematic diagram of the structure of the digital twin intelligent monitoring system for the entire process of a coal preparation plant in this embodiment. The system includes: The process data acquisition module is used to collect data from the entire process. The digital twin monitoring module is used to construct a mechanism twin monitoring model and a data-driven monitoring model, and to acquire mechanism residual values ​​and fault prediction results based on the full-process data, the mechanism twin monitoring model, and the data-driven monitoring model. It is also used to calibrate the parameters of the mechanism twin monitoring model based on the mechanism residual values, to perform convergence processing on the parameter calibration process, and to drive early warning based on the fault prediction results. The digital twin monitoring module is connected to the process data acquisition module. The consistency arbitration module is used to perform consistency arbitration on the mechanism twin monitoring model and the data-driven monitoring model, obtain the consistency arbitration result, and output the final fault prediction result based on the consistency arbitration result. The consistency arbitration module is connected to the digital twin monitoring module. The closed-loop verification module is used to perform closed-loop verification on the final fault prediction result, obtain the closed-loop verification result, and optimize the mechanism twin monitoring model and supplement the data-driven monitoring model based on the closed-loop verification result. The closed-loop verification module is connected to the consistency arbitration module.

[0022] Specifically, the digital twin intelligent monitoring system for the entire coal preparation plant is applied to monitoring equipment such as intelligent edge computing boxes. The system collects data from the entire process through a process data acquisition module. It also constructs a mechanistic twin monitoring model and a data-driven monitoring model through a digital twin monitoring module, acquiring mechanistic residual values ​​and fault prediction results. This allows for fault simulation and prediction from both physical twin and historical data experience levels, improving the accuracy, timeliness, and reliability of fault prediction. The system calibrates the parameters of the mechanistic twin monitoring model through the digital twin monitoring module, further refining the model. When the total number of parameter calibrations is abnormal, the system is forced to converge promptly, preventing untimely warnings of latent faults in the operating conditions, thereby improving intelligent monitoring. The system exhibits robustness and deep perception capabilities, and promptly issues early warnings when the predicted failure probability is abnormal. This facilitates real-time monitoring of potential failure modes within a future timeframe, enabling maintenance personnel to conduct early repairs and improving the practicality and timeliness of intelligent monitoring and fault early warning. Furthermore, the system integrates fault judgment from the mechanistic twin monitoring model and the data-driven monitoring model via a consistency arbitration module, addressing the unreliability of single models and conflicting conclusions. This enhances the accuracy and reliability of fault monitoring and early warning. The system also employs a closed-loop verification module for on-site verification, comparing the final fault prediction results with actual fault results. This allows for real-time monitoring of the accuracy of fault prediction and timely calibration of the mechanistic twin monitoring model and the data-driven monitoring model, further improving the accuracy of fault prediction.

[0023] Specifically, the process data acquisition module collects data from the entire process, including target mechanism data and target operation data.

[0024] Specifically, the target mechanism data includes equipment component dimensions, equipment structure, assembly relationships, component materials, component density, and component strength. The target operating data includes component vibration, temperature, pressure, current, and maintenance records. The equipment refers to key mechanical equipment in a coal preparation plant that directly participates in the core production process, and whose operating status directly affects product output and quality, such as crushers, centrifuges, and hydrocyclones. The equipment component dimensions refer to the geometric dimensions of the equipment components, such as the diameter and length of the crusher's toothed rollers and the diameter and cone angle of the hydrocyclone's cylinder. In this embodiment, the equipment component dimensions are collected from the equipment's factory drawings. The equipment structure refers to the mechanical construction and composition of the equipment, such as whether a centrifuge is vertical or horizontal. The assembly relationships refer to the connection and fit between components, such as the specific way a bearing is installed on a shaft. In this embodiment, the equipment structure and assembly relationships are collected from the equipment structure drawings and equipment assembly drawings. The component materials refer to the manufacturing materials of equipment components, such as toothed plates, screen baskets, and pipe liners, such as high manganese steel, polyurethane, and ceramics. In this embodiment, the component materials are obtained from the equipment material list. The component density refers to the component material... The mass density of the component refers to its mass density, and the component strength refers to the mechanical properties of the component material, such as yield strength and tensile strength. In this embodiment, the component density and component strength are collected by consulting mechanical design manuals. The component vibration refers to the acceleration of key component points of the equipment, such as bearing housings and housings, in three directions. In this embodiment, the vibration of the components is collected by setting vibration acceleration sensors at key component points of the equipment. The temperature refers to key temperature nodes of the equipment, such as the temperature values ​​of bearing surfaces, motor windings, lubricating oil, and hydraulic systems. In this embodiment, the temperature is measured by an infrared thermometer. The pressure refers to the pressure value in the fluid system of the equipment, such as the outlet pressure of the pump, the feed pressure of the filtration system, and the air jacket pressure. The current refers to the three-phase operating current of the drive motor, such as the pump and the fan. In this embodiment, the current is collected by a current transformer. The maintenance record refers to the record document of the equipment's previous repairs, maintenance, inspections, and fault handling. This embodiment does not limit the specific method of collecting maintenance records. Relevant personnel in the art can freely choose according to actual needs, such as collecting maintenance records through equipment management information systems and on-site records of maintenance personnel.

[0025] Specifically, the process data acquisition module collects data from the entire process to facilitate subsequent model building based on the data, thereby improving the comprehensiveness of intelligent monitoring of the entire digital twin process.

[0026] Specifically, the consistency arbitration module performs consistency arbitration on the mechanism twin monitoring model and the data-driven monitoring model to obtain a consistency arbitration result, which includes consistent results and inconsistent results. When the consensus arbitration result is consistent, the fault prediction result will be output as the final fault prediction result. When the consistency arbitration result is inconsistent, a digital twin simulation is initiated to obtain the simulation result, and the simulation result is output as the final fault prediction result. When the consensus arbitration result is a special case, an investigative warning will be issued.

[0027] Specifically, the consistency arbitration refers to the process of comparing the equipment fault conditions reflected by the mechanistic twin monitoring model and the data-driven monitoring model. For example, when the total number of parameter calibrations is abnormal, it indicates that the mechanistic twin monitoring model determines that the equipment has a fault; when the predicted fault probability is abnormal, it indicates that the data-driven monitoring model determines that the equipment has a fault. A consistent arbitration result means that both the mechanistic twin monitoring model and the data-driven monitoring model determine that the equipment has a fault. An inconsistent arbitration result means that the mechanistic twin monitoring model determines that the equipment has a fault, but the data-driven monitoring model determines that the equipment does not have a fault. The digital twin simulation refers to injecting the fault type determined by the mechanistic model into the mechanistic twin monitoring model. For example, assuming a 20% decrease in bearing stiffness, a modified mechanistic twin monitoring model is obtained, and the simulated target operating data is acquired through the modified mechanistic twin monitoring model. The simulated target operating data is then input into... The data-driven monitoring model outputs simulated fault prediction results. These results are then compared with preset predicted fault probabilities. If the predicted fault probability is deemed abnormal, the simulated fault prediction result is output as the simulation result. If the predicted fault probability is deemed normal, a new fault type is indicated as a notification message and pushed to maintenance personnel via a pop-up window in the system control interface. The consistency arbitration result is considered a special case when the mechanistic twin monitoring model determines that the equipment has no fault, but the data-driven monitoring model determines that the equipment has a fault. The investigative warning refers to the process of pushing the fault prediction results output by the data-driven monitoring model in special cases. For example, if the data-driven monitoring model detects early pitting corrosion characteristics in the bearing of centrifuge No. 3 and predicts a fault probability of 75%, but the mechanistic model does not find significant physical deviations, this is pushed to maintenance personnel via a pop-up window in the system control interface.

[0028] Specifically, the consistency arbitration module solves the problems of unreliability of a single model and conflicting conclusions by integrating the fault judgment of the mechanistic twin monitoring model and the fault judgment of the data-driven monitoring model, thereby improving the accuracy of fault monitoring and early warning and the reliability of decision-making.

[0029] Specifically, the closed-loop verification module performs closed-loop verification on the final fault prediction result. Specifically, the operation and maintenance personnel conduct on-site verification based on the final fault prediction result to obtain the actual fault result. The actual fault result is then compared with the final fault prediction result to obtain the closed-loop verification result. The closed-loop verification result includes prediction verification qualified and prediction verification unqualified. When the closed-loop verification result is that the prediction verification is qualified, no model optimization is performed on the mechanism twin monitoring model and no data supplementation is performed on the data-driven monitoring model. When the closed-loop verification result is that the prediction verification is unqualified, the mechanism twin monitoring model is optimized: the parameters of the mechanism twin monitoring model are fine-tuned according to the actual fault results to obtain the fine-tuned mechanism twin monitoring model, and the fine-tuned mechanism twin monitoring model is used as the mechanism twin monitoring model. The closed-loop verification module supplements the data-driven monitoring model by adding the actual fault results and their corresponding target operating data to the historical operating dataset and re-propagating the initial neural network model forward.

[0030] Specifically, the on-site verification refers to the process by which maintenance personnel, based on the final fault prediction results, conduct a series of inspections, tests, and confirmations at the equipment site. The actual fault result refers to the standardized record of the actual equipment status and fault details objectively confirmed through on-site verification. The comparison refers to the structured analysis process of comparing the final fault prediction results with the actual fault results in terms of time, space, type, and degree. The closed-loop verification result refers to the quantitative evaluation of the prediction performance generated through comparison. The prediction verification pass means that the final fault prediction result matches the actual fault result. The prediction verification fail means that the final fault prediction result does not match the actual fault result. The parameter fine-tuning refers to the process of optimizing and adjusting the uncertain physical parameters, such as stiffness, damping, and friction coefficient, in the preliminary mechanism twin monitoring model while keeping the dynamic principles unchanged, so that the output of the mechanism twin monitoring model converges with the actual fault result.

[0031] Specifically, the closed-loop verification module compares the final fault prediction result with the actual fault result by conducting on-site verification, so as to monitor the accuracy of fault prediction in real time and calibrate the mechanism twin monitoring model and the data-driven monitoring model in a timely manner, thereby improving the accuracy of fault prediction.

[0032] Please see Figure 2 As shown, this is a structural schematic diagram of the digital twin monitoring module in this embodiment. The digital twin monitoring module includes: The model building unit is used to build mechanism twin monitoring models and data-driven monitoring models; The model prediction unit is used to obtain the mechanism residual value based on the full process data and the mechanism twin monitoring model, and also to obtain the fault prediction result based on the full process data and the data-driven monitoring model. The model prediction unit is connected to the model construction unit. The model calibration unit is used to calibrate the parameters of the mechanism twin monitoring model according to the mechanism residual value, to obtain the total number of parameter calibrations, to perform convergence processing on the parameter calibration process according to the total number of parameter calibrations, and to drive early warning based on the fault prediction results. The model calibration unit is connected to the model prediction unit.

[0033] Specifically, the model building unit constructs the mechanism twin monitoring model by: performing geometric modeling based on the target mechanism data to obtain a preliminary mechanism model, and setting physical law constraints on the preliminary mechanism model to obtain the mechanism twin monitoring model.

[0034] Specifically, the geometric modeling refers to a geometric solid model reflecting the three-dimensional shape and assembly relationship of the equipment, established using dynamics software based on target mechanism data. This embodiment does not limit the dynamics software; those skilled in the art can freely choose according to actual needs, such as Simpack. Simpack refers to multibody dynamics simulation software, its full name being Simpack Multibody. In Simulation, the physical law constraint setting refers to the process of setting the physical relationships and constraints between components of the preliminary mechanism model based on dynamic principles using dynamic software. The dynamic principles refer to the systematic establishment and solution of a set of mathematical equations describing the motion and forces of the preliminary mechanism model, such as the Newton-Euler equations describing the translational and rotational motion of equipment components under the action of forces and torques, and the constraint equations describing the connection relationships between equipment components, such as the relationship between hinges and sliders, which restrict the motion. This embodiment does not limit the specific implementation method of the physical law constraint setting. Those skilled in the art can freely choose according to actual needs. For example, in Simpack, physical law constraints can be set for the preliminary mechanism model by dragging and dropping predefined physical component icons and connecting them, such as dragging in a mass block and a spring damper and connecting them. The software background will automatically generate the corresponding differential algebraic equations.

[0035] Specifically, the model building unit constructs a data-driven monitoring model according to a data-driven monitoring model construction method, which includes: Step B01: Divide the historical dataset into a 70% training set, a 20% validation set, and a 10% test set. Step B02: Initialize the parameters of the recurrent neural network model to obtain the initial neural network model; Step B03: Input the training set into the initial neural network model for forward propagation to obtain the initial prediction result, and calculate the loss of the initial prediction result using the cross-entropy function to obtain the loss value. Perform backpropagation based on the loss value, and update the parameters through the optimizer to obtain the trained neural network model. Step B04: Input the trained neural network model into the validation set, and then perform forward propagation again on the test set for the trained neural network model with the lowest loss value on the validation set to obtain the final prediction result. The trained neural network model with a final prediction accuracy greater than 95% is output as the data-driven monitoring model.

[0036] Specifically, the historical operation dataset refers to the target operation data that occurred within a preset historical period and its corresponding fault prediction results. The fault prediction results include fault modes and predicted fault probabilities. The fault mode refers to a specific type of fault that the equipment may experience, such as uneven rotor mass distribution or misalignment of the center lines of the two shafts connected by the coupling. The predicted fault probability refers to the quantified value of the likelihood of a fault mode occurring within a specific future time period, such as the next 24 hours. The parameter initialization refers to the process of assigning initial values ​​to the weights and biases in the recurrent neural network model. This initialization can be randomized. This embodiment does not limit the selection of the optimizer. Those skilled in the art can freely choose according to actual needs, such as Adam. The final prediction result refers to the fault prediction result obtained after the trained neural network model is re-propagated forward through the test set.

[0037] Specifically, the model building unit constructs a mechanism twin monitoring model and a data-driven monitoring model to provide an interpretable physical baseline while identifying complex failure modes from historical data, thereby improving the accuracy, timeliness and reliability of failure prediction.

[0038] Specifically, the model prediction unit obtains the mechanism residual value based on the full-process data and the mechanism twin monitoring model. Specifically, it obtains the actual component vibration yx and inputs the actual working condition parameters into the mechanism twin monitoring model to obtain the component vibration zx simulated by the mechanism twin monitoring model. Based on the actual component vibration yx and the component vibration zx simulated by the mechanism twin monitoring model, the mechanism residual value mx is calculated, and mx is set to [|(yx-zx)| / yx]×100%. The model prediction unit acquires the fault prediction results based on the full-process data and the data-driven monitoring model. It inputs the target operation data from the full-process data into the data-driven monitoring model to obtain the fault prediction results output by the data-driven monitoring model. The fault prediction results include the fault mode and the predicted fault probability Pg.

[0039] Specifically, the actual component vibration refers to the component vibration during operation under known stable operating conditions in reality. The actual operating condition parameters refer to the pressure and flow rate under known stable operating conditions in reality. The known stable operating conditions refer to the working conditions under which the equipment continuously operates in a preset, repeatable state, and the fluctuation of the actual operating condition parameters remains within the allowable range of the project.

[0040] Specifically, the model prediction unit acquires the mechanism residual value and fault prediction results to simulate and predict faults from the perspectives of physical twins and historical data experience judgment, thereby further improving the accuracy of fault prediction.

[0041] Specifically, the model calibration unit calibrates the parameters of the mechanism twin monitoring model based on the mechanism residual value, compares the mechanism residual value mx with a preset mechanism residual value mx0, judges the compliance level of the mechanism residual value based on the comparison result, and calibrates the parameters of the mechanism twin monitoring model based on the judgment result, wherein: When zy≤zy0, the model calibration unit determines that the compliance level of the mechanism residual value is met and does not perform parameter calibration on the mechanism twin monitoring model; When zy > zy0, the model calibration unit determines that the compliance level of the mechanism residual value is not met, and performs parameter calibration on the mechanism twin monitoring model.

[0042] Specifically, the preset mechanism residual value refers to a preset value used to judge the degree of compliance of the mechanism residual value. This embodiment does not limit the specific value setting of the preset mechanism residual value mx0. Those skilled in the art can freely choose according to actual needs. For example, the preset mechanism residual value mx0 can be set according to the actual needs of the equipment components. For example, for centrifuges and steam turbines, according to engineering experience, when mx0 > 5%, it is easy to lead to insufficient accuracy of the mechanism twin monitoring model, and when mx0 < 5%, it is easy to lead to low model training efficiency. Therefore, mx0 = 5% is set. The degree of compliance of the mechanism residual value includes compliance and non-compliance. The parameter calibration refers to the process of optimizing and adjusting the uncertain physical parameters in the preliminary mechanism twin monitoring model, such as stiffness, damping and friction coefficient, under the premise of keeping the dynamic principle unchanged, so that the output of the mechanism twin monitoring model achieves the best match with the measured data of the real system in a healthy state.

[0043] Specifically, the model calibration unit judges the degree of compliance of the mechanism residual value in order to further calibrate the mechanism twin monitoring model, thereby improving the reliability of fault prediction.

[0044] Specifically, the model calibration unit acquires the total number of parameter calibrations, vp, compares vp with a preset total number of parameter calibrations, vp0, judges the status of the total number of parameter calibrations based on the comparison result, and performs convergence processing on the parameter calibration process based on the judgment result, wherein: When vp≤vp0, the model calibration unit determines that the total number of parameter calibrations is normal and does not perform convergence processing on the parameter calibration process. When vp > vp0, the model calibration unit determines that the total number of parameter calibrations is abnormal and performs convergence processing on the parameter calibration process: the parameter calibration process of the preliminary mechanism twin monitoring model is forcibly converged, and the equipment components corresponding to the mechanism residual values ​​are pushed.

[0045] Specifically, the total number of parameter calibrations refers to the total number of times the digital twin monitoring module calibrates the mechanism twin monitoring model within a preset time period. The preset time period refers to a pre-set historical backtracking time length, such as 24 hours. In this embodiment, the total number of parameter calibrations is recorded and obtained through local system logs. The preset total number of parameter calibrations refers to a preset value for judging the status of the total number of parameter calibrations. This embodiment does not limit the specific value of the preset total number of parameter calibrations vp0. Those skilled in the art can freely choose according to actual needs. For example, based on historical experience, when vp0 > 3 times, the device... The actual probability of failure is 89%. When vp0 < 3 times, the actual probability of equipment failure is 31%. Therefore, vp0 is set to 3 times. The status of the total number of parameter calibrations refers to the normality of the total number of parameter calibrations, including normal and abnormal. The forced convergence refers to interrupting the parameter calibration process. The equipment component corresponding to the mechanism residual value refers to the equipment component used to calculate the mechanism residual value, such as centrifuges and steam turbines. This embodiment does not limit the specific method of pushing the equipment component corresponding to the mechanism residual value. Those skilled in the art can freely choose according to actual needs, such as pushing through the pop-up window of the system control interface.

[0046] Specifically, the model calibration unit judges the status of the total number of parameter calibrations. When the status of the total number of parameter calibrations is abnormal, it forces convergence in a timely manner to avoid the system failing to provide timely warnings of hidden faults in the operating conditions, thereby improving the robustness and deep perception capability of intelligent monitoring.

[0047] Specifically, the model calibration unit drives early warning based on the fault prediction results, which involves comparing the predicted fault probability Pg in the fault prediction results with the preset predicted fault probability Pg0, judging the state of the predicted fault probability based on the comparison results, and driving early warning based on the judgment results. When Pg≤Pg0, the model calibration unit determines the predicted fault probability state as normal and does not perform a drive warning; When Pg > Pg0, the model calibration unit determines that the predicted fault probability is abnormal and drives an early warning: it alerts maintenance personnel by flashing indicator lights and pushes the fault mode in the fault prediction results.

[0048] Specifically, the preset predicted fault probability refers to a preset value for judging the state of the predicted fault probability. This embodiment does not limit the specific value of the preset predicted fault probability Pg0. Those skilled in the art can freely choose according to actual needs. For example, the value of the preset predicted fault probability Pg0 can be set according to the fault mode. When the fault mode is centrifuge bearing failure, according to engineering experience, when Pg0 > 75%, it is easy to cause untimely warning. Therefore, Pg0 = 75% is set. The state of the predicted fault probability refers to the degree of normality of the predicted fault probability judged by the predicted fault probability and the preset predicted fault probability, including normal and abnormal. This embodiment does not limit the specific method of pushing fault modes. Those skilled in the art can freely choose according to actual needs. For example, the fault mode can be pushed through a pop-up window in the system control interface.

[0049] Specifically, the model calibration unit judges the state of the predicted fault probability. When the state of the predicted fault probability is abnormal, it promptly drives an early warning to facilitate real-time monitoring of possible fault modes in the future time period. This allows maintenance personnel to perform maintenance in advance, thereby improving the practicality and timeliness of intelligent monitoring and fault early warning.

[0050] Please see Figure 3 As shown, this is a flowchart illustrating the intelligent monitoring and fault prediction method for the entire digital twin process of a coal preparation plant in this embodiment. The method includes: Step S1: Collect data for the entire process; Step S2 involves constructing a mechanism twin monitoring model and a data-driven monitoring model, obtaining mechanism residual values ​​based on the full-process data and the mechanism twin monitoring model, obtaining fault prediction results based on the full-process data and the data-driven monitoring model, calibrating the mechanism twin monitoring model based on the mechanism residual values, obtaining the total number of parameter calibrations, performing convergence processing on the parameter calibration process based on the total number of parameter calibrations, and driving early warning based on the fault prediction results. Step S3: Perform consistency arbitration on the mechanism twin monitoring model and the data-driven monitoring model to obtain the consistency arbitration result, and output the final fault prediction result based on the consistency arbitration result; Step S4: Perform closed-loop verification on the final fault prediction result to obtain the closed-loop verification result, and optimize the mechanism twin monitoring model and supplement the data-driven monitoring model based on the closed-loop verification result.

[0051] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A digital twin intelligent monitoring system for the entire process of a coal preparation plant, characterized in that: The system includes: The process data acquisition module is used to collect data from the entire process. The digital twin monitoring module is used to construct the mechanism twin monitoring model and the data-driven monitoring model, and to acquire the mechanism residual value and fault prediction results. It is also used to calibrate the parameters of the mechanism twin monitoring model based on the mechanism residual value, and to perform convergence processing on the parameter calibration process. Furthermore, it is used to drive early warning based on the fault prediction results. The consistency arbitration module is used to perform consistency arbitration on the mechanism twin monitoring model and the data-driven monitoring model, obtain the consistency arbitration result, and output the final fault prediction result based on the consistency arbitration result. The closed-loop verification module is used to perform closed-loop verification on the final fault prediction results, obtain the closed-loop verification results, and optimize the mechanism twin monitoring model and supplement the data-driven monitoring model based on the closed-loop verification results. The digital twin monitoring module includes: The model building unit is used to build mechanism twin monitoring models and data-driven monitoring models; The model prediction unit is used to obtain the mechanism residual value based on the full process data and the mechanism twin monitoring model, and to obtain the fault prediction result based on the full process data and the data-driven monitoring model. The model calibration unit is used to calibrate the parameters of the mechanism twin monitoring model based on the mechanism residual value, obtain the total number of parameter calibrations, perform convergence processing on the parameter calibration process based on the total number of parameter calibrations, and drive early warning based on the fault prediction results.

2. The intelligent monitoring system for the entire process of a coal preparation plant using a digital twin as described in claim 1, characterized in that, The model building unit constructs the mechanism twin monitoring model by: performing geometric modeling based on the target mechanism data to obtain a preliminary mechanism model, and setting physical law constraints on the preliminary mechanism model to obtain the mechanism twin monitoring model.

3. The intelligent monitoring system for the entire process of a coal preparation plant using a digital twin as described in claim 2, characterized in that, The model building unit constructs a data-driven monitoring model according to a data-driven monitoring model construction method, which includes: Step B01: Divide the historical dataset into a 70% training set, a 20% validation set, and a 10% test set. Step B02: Initialize the parameters of the recurrent neural network model to obtain the initial neural network model; Step B03: Input the training set into the initial neural network model for forward propagation to obtain the initial prediction result, and calculate the loss of the initial prediction result using the cross-entropy function to obtain the loss value. Perform backpropagation based on the loss value, and update the parameters through the optimizer to obtain the trained neural network model. Step B04: Input the trained neural network model into the validation set, and then perform forward propagation again on the test set for the trained neural network model with the lowest loss value on the validation set to obtain the final prediction result. The trained neural network model with a final prediction accuracy greater than 95% is output as the data-driven monitoring model.

4. The intelligent monitoring system for the entire process of a coal preparation plant using a digital twin as described in claim 3, characterized in that, The model prediction unit obtains the mechanism residual value based on the full-process data and the mechanism twin monitoring model. Specifically, it obtains the actual component vibration yx and inputs the actual working condition parameters into the mechanism twin monitoring model to obtain the component vibration zx simulated by the mechanism twin monitoring model. Based on the actual component vibration yx and the component vibration zx simulated by the mechanism twin monitoring model, the mechanism residual value mx is calculated, and mx is set to [|(yx-zx)| / yx]×100%. The model prediction unit acquires the fault prediction results based on the full-process data and the data-driven monitoring model. It inputs the target operation data from the full-process data into the data-driven monitoring model to obtain the fault prediction results output by the data-driven monitoring model. The fault prediction results include the fault mode and the predicted fault probability Pg.

5. The intelligent monitoring system for the entire process of a coal preparation plant using a digital twin as described in claim 4, characterized in that, The model calibration unit calibrates the parameters of the mechanism twin monitoring model based on the mechanism residual value, compares the mechanism residual value mx with the preset mechanism residual value mx0, judges the degree of compliance of the mechanism residual value based on the comparison result, and calibrates the parameters of the mechanism twin monitoring model based on the judgment result, wherein: When zy≤zy0, the model calibration unit determines that the compliance level of the mechanism residual value is met and does not perform parameter calibration on the mechanism twin monitoring model; When zy > zy0, the model calibration unit determines that the compliance level of the mechanism residual value is not met, and performs parameter calibration on the mechanism twin monitoring model.

6. The intelligent monitoring system for the entire process of a coal preparation plant using a digital twin as described in claim 5, characterized in that, The model calibration unit acquires the total number of parameter calibrations, vp, compares vp with a preset total number of parameter calibrations, vp0, judges the status of the total number of parameter calibrations based on the comparison result, and performs convergence processing on the parameter calibration process based on the judgment result, wherein: When vp≤vp0, the model calibration unit determines that the total number of parameter calibrations is normal and does not perform convergence processing on the parameter calibration process. When vp > vp0, the model calibration unit determines that the total number of parameter calibrations is abnormal and performs convergence processing on the parameter calibration process: the parameter calibration process of the preliminary mechanism twin monitoring model is forcibly converged, and the equipment components corresponding to the mechanism residual values ​​are pushed.

7. The intelligent monitoring system for the entire process of a coal preparation plant using a digital twin as described in claim 6, characterized in that, The model calibration unit drives early warning based on the fault prediction results, specifically by comparing the predicted fault probability Pg in the fault prediction results with the preset predicted fault probability Pg0, judging the state of the predicted fault probability based on the comparison results, and driving early warning based on the judgment results, wherein: When Pg≤Pg0, the model calibration unit determines the predicted fault probability state as normal and does not perform a drive warning; When Pg > Pg0, the model calibration unit determines that the predicted fault probability is abnormal and drives an early warning: it alerts maintenance personnel by flashing indicator lights and pushes the fault mode in the fault prediction results.

8. The intelligent monitoring system for the entire process of a coal preparation plant using a digital twin as described in claim 7, characterized in that, The consistency arbitration module performs consistency arbitration on the mechanism twin monitoring model and the data-driven monitoring model to obtain a consistency arbitration result, which includes consistent results and inconsistent results. When the consensus arbitration result is consistent, the fault prediction result will be output as the final fault prediction result. When the consistency arbitration result is inconsistent, a digital twin simulation is initiated to obtain the simulation result, and the simulation result is output as the final fault prediction result. When the consensus arbitration result is a special case, an investigative warning will be issued.

9. The intelligent monitoring system for the entire process of a coal preparation plant using a digital twin as described in claim 8, characterized in that, The closed-loop verification module performs closed-loop verification on the final fault prediction result. Specifically, the operation and maintenance personnel conduct on-site verification based on the final fault prediction result to obtain the actual fault result. The actual fault result is then compared with the final fault prediction result to obtain the closed-loop verification result. The closed-loop verification result includes prediction verification qualified and prediction verification unqualified. When the closed-loop verification result is that the prediction verification is qualified, no model optimization is performed on the mechanism twin monitoring model and no data supplementation is performed on the data-driven monitoring model. When the closed-loop verification result is that the prediction verification is unqualified, the mechanism twin monitoring model is optimized: the parameters of the mechanism twin monitoring model are fine-tuned according to the actual fault results to obtain the fine-tuned mechanism twin monitoring model, and the fine-tuned mechanism twin monitoring model is used as the mechanism twin monitoring model. The closed-loop verification module supplements the data-driven monitoring model by adding the actual fault results and their corresponding target operating data to the historical operating dataset and re-propagating the initial neural network model forward.

10. A fault prediction method for intelligent monitoring of the entire process of a coal preparation plant using a digital twin as described in any one of claims 1-9, the method comprising: Step S1: Collect data for the entire process; Step S2 involves constructing a mechanism twin monitoring model and a data-driven monitoring model, obtaining mechanism residual values ​​based on the full-process data and the mechanism twin monitoring model, obtaining fault prediction results based on the full-process data and the data-driven monitoring model, calibrating the mechanism twin monitoring model based on the mechanism residual values, obtaining the total number of parameter calibrations, performing convergence processing on the parameter calibration process based on the total number of parameter calibrations, and driving early warning based on the fault prediction results. Step S3: Perform consistency arbitration on the mechanism twin monitoring model and the data-driven monitoring model to obtain the consistency arbitration result, and output the final fault prediction result based on the consistency arbitration result; Step S4: Perform closed-loop verification on the final fault prediction result to obtain the closed-loop verification result, and optimize the mechanism twin monitoring model and supplement the data-driven monitoring model based on the closed-loop verification result.

Images