A coal mill fault modeling and diagnosis method based on deep learning and digital twin technology coupling
The coal mill fault modeling method, which couples deep learning and digital twin technologies, solves the problems of low accuracy and poor versatility of coal mill fault diagnosis systems, realizes real-time fault identification and monitoring, and improves the stability and energy consumption management of power plant systems.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUAZHONG UNIV OF SCI & TECH
- Filing Date
- 2026-02-26
- Publication Date
- 2026-06-09
AI Technical Summary
Existing coal mill fault diagnosis systems suffer from low accuracy and poor versatility. In particular, they are inaccurate in simulating the internal flow and fault mechanism of the coal mill, have limited fault data, and cannot adapt to different working conditions, resulting in poor fault identification performance.
A fault modeling method for coal mills is established by adopting a method based on the coupling of deep learning and digital twin technology. Fault data is generated through digital twin simulation model, and fault type identification and prediction are performed by combining deep learning model. Real-time simulation and visualization monitoring are carried out using digital twin technology.
It improves the accuracy and versatility of coal mill fault diagnosis, enabling real-time fault identification and abnormal status monitoring, enhancing the stability and energy consumption management of power plant systems, and providing a scientific fault diagnosis solution.
Smart Images
Figure CN122174632A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the technical field of intelligent operation of thermal power units, and more specifically, relates to a method for fault modeling and diagnosis of coal mills based on the coupling of deep learning and digital twin technology. Background Technology
[0002] Fault diagnosis involves identifying and determining the type of fault based on early fault characteristics before the fault occurs, and issuing an alarm to allow sufficient time for corresponding control actions to prevent the fault from worsening.
[0003] Traditional diagnostic methods, relying on measured and fixed-value warnings, are often limited by detection technology and human factors, resulting in biases and delays. This not only easily leads to misjudgments of fault types but also often results in damage or further deterioration by the time a warning occurs, posing a significant threat to the normal and safe operation of coal mill systems and power plant units. With the development of measurement and sensing technology and the rise of artificial intelligence algorithms, intelligent diagnostic methods, including fault feature analysis, probabilistic statistics, state classification, and state prediction, have become mainstream in coal mill fault identification engineering practices. While these intelligent methods improve the accuracy and speed of fault identification compared to measured and fixed-value warning methods, they generally lack detailed characterization of the mechanistic flow of raw coal in the coal mill under normal and fault conditions. Furthermore, the limited amount of fault data also restricts the performance of fault identification systems.
[0004] Currently, due to the lack of accurate simulation of the internal flow and fault mechanisms of coal mills, the development of coal mill fault diagnosis systems faces the following challenges: the internal flow and fault mechanisms of coal mills are not explicitly present in the fault identification model, reducing the versatility and interpretability of the fault identification model; the fault data of coal mills is limited, and if only the recorded data is used to train the fault identification model, the amount of training data and the types of annotations are insufficient, which will lead to low accuracy of the fault identification model; the working environment of coal mills is relatively complex, and the working state of coal mills will also change over time. If only existing operating data is used to build the fault identification model, there is a risk that the fault data will deviate from the characteristics of the existing fault data, resulting in identification failure.
[0005] Furthermore, the operating conditions of coal mills vary across different power plant systems, necessitating customized fault identification methods. Existing mature industrial fault identification methods often lack the ability to be readily adapted to local conditions, leading to inconsistent accuracy and deployment effectiveness of coal mill fault identification systems across different power plant systems. Therefore, a fault diagnosis method capable of addressing these issues is needed. Summary of the Invention
[0006] In view of the above-mentioned defects or improvement needs of existing technologies, this invention provides a coal mill fault modeling and diagnosis method based on the coupling of deep learning and digital twin technology, which solves the problems of low accuracy and poor versatility of various coal mill diagnostic systems.
[0007] To achieve the above objectives, according to one aspect of the present invention, a method for fault modeling of a coal mill based on the coupling of deep learning and digital twin technology is provided, the method comprising the following steps: A digital twin simulation model of a coal mill was built, and the historical parameters at the inlet and outlet of the coal mill were used as inputs and outputs, respectively, to identify the model parameters. Set the fault types of the coal mill, adjust the parameters of the coal mill during operation according to the set fault types, and input the adjusted parameters into the digital twin simulation model to generate fault data corresponding to various fault types. A deep learning model of fault types is built by using fault data as input and fault type as output. The fault type deep learning model is trained using data from the database, thereby realizing the modeling of the fault type deep learning model.
[0008] More preferably, the fault types of the coal mill include coal shortage fault, spontaneous combustion fault, and coal blockage fault; the deep learning model for training the fault types adopts a parameter transfer strategy.
[0009] More preferably, the digital twin simulation model includes an inlet boundary condition module, an outlet boundary condition module, a mass conservation module, and an energy conservation module. The inlet boundary condition module sets the input boundary conditions of the digital twin simulation model, the outlet boundary condition module sets the output boundary conditions, the mass conservation module sets the relationship between various physical quantities while maintaining mass conservation during the operation of the coal mill, and the energy conservation module sets the relationship between various physical quantities while maintaining mass conservation during the operation of the coal mill.
[0010] More preferably, the fault data under the coal shortage fault is generated by the digital twin simulation model of the coal mill, which uses the amount of coal entering the coal mill that exceeds the preset normal range as the fault data.
[0011] More preferably, the fault data acquisition method under the spontaneous combustion fault is as follows: adjust the calculation formula of the raw coal mass change in the grinding zone in the mass conservation module and the calculation formula of the temperature change at the coal mill outlet in the energy conservation module in the digital twin simulation model, thereby adjusting the digital twin simulation model, and using the adjusted digital twin simulation model to generate fault data under the spontaneous combustion fault.
[0012] More preferably, the calculation formula for the temperature change at the coal mill outlet in the energy conservation module is adjusted as follows:
[0013] in, For the raw coal on the adjusted grinding area platform, It is the mass flow rate of pulverized coal. The mass flow rate of the raw coal being ground. This indicates the proportion of spontaneously combusted pulverized coal in the coal mill. It represents the mass of raw coal currently present on the grinding zone platform, and t represents the time of temperature change.
[0014] More preferably, the calculation formula for the temperature change at the coal mill outlet in the energy conservation module is adjusted as follows:
[0015]
[0016]
[0017] in, This is the adjusted temperature at the coal mill outlet. The static pressure coefficient of the coal mill. The specific heat capacity of the primary air at constant pressure at the inlet. The inlet primary air temperature, The specific heat capacity of the raw coal at constant pressure. The temperature of the incoming raw coal. The specific heat capacity of the primary air at constant pressure at the outlet. For the primary air mass flow rate at the outlet, It is the mass of evaporated water. This represents the enthalpy change per unit mass of water evaporated. The specific heat capacity of pulverized coal at constant pressure for export. This represents the mass flow rate of pulverized coal after drying. The density enhancement factor of the primary wind pressure difference. This refers to the rated power of the coal mill. To ensure the quality of the raw coal entering the coal mill, It's about the quality of the coal powder on the platform. It was the quality of coal dust blown up by the wind. It is the heat capacity of raw coal. It refers to the carbon content in coal. It refers to the hydrogen content in coal. It refers to the oxygen content in coal. It refers to the sulfur content in coal. It refers to the moisture content in coal.
[0018] More preferably, the fault data under the coal blockage fault is obtained as follows: the calculation formula of the mass flow rate of the ground raw coal in the digital twin simulation model is adjusted, thereby changing the digital twin simulation model, and the fault data under the coal blockage fault is generated using the changed digital twin simulation model.
[0019] More preferably, the calculation formula for the mass flow rate of the ground raw coal in the adjusted digital twin simulation model is as follows:
[0020] in, This is the adjusted mass flow rate of the ground raw coal. It is the percentage of coal dust blown into the air by the wind once per second. It represents the percentage of raw coal being ground per second on the platform. It represents the change in the conversion efficiency of raw coal to pulverized coal.
[0021] According to another aspect of the present invention, a coal mill fault diagnosis method based on the coupling of deep learning and digital twin technology is provided. This method uses a fault type deep learning model established by the above-described coal mill fault modeling method based on the coupling of deep learning and digital twin technology to predict the faults of the coal mill.
[0022] In summary, the technical solutions conceived by this invention have the following beneficial effects compared with the prior art: 1. The coal mill fault modeling and diagnosis method proposed in this invention, which is based on the coupling of deep learning and digital twin technology, can determine faults in real time through deep learning and monitor abnormal states in the coal mill using a visual interface. This effectively solves the problem of low accuracy in coal mill diagnostic systems and improves the versatility of coal mill fault prediction, providing a reliable solution for the stable operation of power plant systems.
[0023] 2. The digital twin model established in this invention can simulate the dynamic changes in the operation of a coal mill and calculate the characteristics of unknown parameter changes inside the coal mill in real time. It can achieve real-time simulation based on the operating mechanism of the coal mill and sensor data, improving the stability of the power plant system. This model not only supports simulation but can also be used for daily operating condition monitoring and fault diagnosis, providing a scientific basis for the stable operation of the power plant system and overall energy consumption management.
[0024] 3. This embodiment discloses a coal mill fault modeling and diagnosis method based on the coupling of deep learning and digital twin technology. It enables real-time simulation based on the coal mill's operating mechanism and sensor data. The training set of the deep learning model is expanded by simulating operational data under different fault characteristics to improve model recognition accuracy. The deep learning model, trained on a fault database, can determine faults in real time and monitor possible abnormal states in the coal mill using a visual interface, providing a scientific basis and reliable solutions for the stable operation and energy consumption management of power plant systems. Attached Figure Description
[0025] Figure 1 A spatial schematic diagram of a coal mill for a coal mill fault modeling and diagnosis method based on the coupling of deep learning and digital twin technology provided by the present invention.
[0026] Figure 2 A schematic diagram of a digital twin model of a coal mill, which is a method for fault modeling and diagnosis of coal mills based on the coupling of deep learning and digital twin technology provided by the present invention.
[0027] Figure 3 The flowchart of the genetic algorithm parameter identification program for a coal mill fault modeling and diagnosis method based on the coupling of deep learning and digital twin technology provided by the present invention is shown.
[0028] Figure 4 This invention provides a schematic diagram of the changes in coal shortage fault parameters in a coal mill fault modeling and diagnosis method based on the coupling of deep learning and digital twin technology.
[0029] Figure 5 This invention provides a schematic diagram of the spontaneous combustion fault parameter changes in a coal mill fault modeling and diagnosis method based on the coupling of deep learning and digital twin technology.
[0030] Figure 6 This invention provides a schematic diagram of coal blockage fault parameter changes in a coal mill fault modeling and diagnosis method based on the coupling of deep learning and digital twin technologies.
[0031] Figure 7 The diagram shows the structure of a deep learning fault diagnosis model for a coal mill fault modeling and diagnosis method based on the coupling of deep learning and digital twin technology, which is provided by this invention.
[0032] Figure 8 The present invention provides a transfer learning fault diagnosis flowchart for a coal mill fault modeling and diagnosis method based on the coupling of deep learning and digital twin technology. Detailed Implementation
[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.
[0034] A method for fault modeling and diagnosis of coal mills based on the coupling of deep learning and digital twin technologies; please refer to [link to relevant documentation]. Figures 1 to 8 The method mainly includes the following steps: S1. Obtain characteristic information such as the type and model of the physical coal mill. Read historical data on inlet coal quantity, inlet primary air quantity, inlet primary air temperature, outlet temperature, and coal mill current.
[0035] S2. Establish an equation-based digital twin simulation model based on the acquired coal mill model, operating data, load, and historical operating environment data. The digital twin simulation model is developed using an open physical modeling computer language. Inputs to the digital twin simulation model include: inlet coal flow rate, inlet air flow rate, and inlet air temperature. Outputs include: outlet temperature and coal mill current.
[0036] S201. Perform preprocessing of the coal mill data to make it smooth data that the model can accept and is easy to calculate.
[0037] S202. Set up a soft measurement module for coal quality based on the combustion product formation mechanism. The coal quality soft measurement module takes flue gas composition information as input and outputs the elemental composition of pulverized coal at the pulverizer outlet. This module is designed based on the chemical combustion reaction mechanism of coal, taking into account the laws of conservation of mass and energy, as well as the reaction process of intermediate products. The quantitative relationships between the various material components in the flue gas and the elemental components in the coal are controlled by the following formula.
[0038] When considering in-furnace desulfurization, two reactions occur in the furnace: combustion and desulfurization. The chemical reaction formula is: (1) (2) (3) Limestone is used as a desulfurizing agent, and its main component is The reaction formula is: (4) (5) The aforementioned reaction equations can determine the mathematical relationship between coal composition and flue gas composition, thus enabling the development of a soft measurement program for coal quality based on an open physical modeling computer language. The input is flue gas composition information obtained from an on-site flue gas analyzer. After calculating the proportion of each gas in the dry flue gas using the measured values, the elemental composition of the pulverized coal at the mill outlet is obtained through iterative solutions using the open physical modeling computer language. The results calculated by this module can be compared with the elemental composition of the designed coal type to understand the mass exchange between pulverized coal and primary air, aiding subsequent simulation work.
[0039] S203, Setting the coal mill inlet boundary conditions module This module sets the working state of the coal mill at the input end of the model, including parameters such as inlet coal quantity, inlet air volume, and inlet air temperature. It uses these data as boundary conditions to drive the digital twin simulation model.
[0040] S204, Setting the coal mill outlet boundary conditions module This module allows you to obtain the model's operating status at the model's output, including the outlet temperature and the coal mill current. These parameters can then be used to calibrate the unknown parameters in the model.
[0041] S205, Set up the mass conservation module This module considers the law of conservation of mass, the spatial distribution inside the coal mill, and the conversion relationship between raw coal and pulverized coal inside the coal mill, and is mainly controlled by the following formulas.
[0042] Changes in raw coal quality on the platform: (6) In the formula: The weight of raw coal on the platform is (kg). The mass flow rate (kg / s) of raw coal entering the platform. The mass flow rate (kg / s) of pulverized coal returned to the platform. The mass flow rate (kg / s) of the raw coal being ground.
[0043] Changes in pulverized coal on the platform: (7) In the formula: The mass of pulverized coal on the platform (kg). The mass flow rate of pulverized coal is (kg / s).
[0044] Changes in pulverized coal in the air: (8) In the formula: The mass (kg) of coal dust blown up by a single airflow. The mass flow rate of pulverized coal leaving the grinding zone (kg / s).
[0045] Pulverized coal mass flow rate: (9) In the formula: The percentage of coal dust blown into the air once per second by the wind. The mass flow rate (kg / s) of the primary air entering the mill.
[0046] Mass flow rate of pulverized coal leaving the grinding zone: (10) In the formula: The percentage of coal powder leaving the grinding zone. The separator rotation speed (rad / s) This represents the maximum rotational speed of the separator (rad / s).
[0047] Mass flow rate of pulverized coal returned to the platform (11) In the formula: This refers to the proportion of pulverized coal that falls back onto the platform.
[0048] Mass flow rate of the ground raw coal (12) In the formula: This represents the percentage of raw coal that is ground per second on the platform.
[0049] Mass flow rate of raw coal entering the platform (13) In the formula: The mass flow rate (kg / s) of raw coal entering the mill.
[0050] The above formula describes the process where raw coal enters the coal mill, is ground into coal powder by the grinding rollers, is blown up by the primary air, some of the coal powder falls back to the platform due to its large particle size, and the remaining coal powder enters the separator. After being screened by the separator, some of the coal powder is screened back to the platform.
[0051] Coal mill current: (14) In the formula: The current of the coal mill is (A). The conversion coefficient between pulverized coal quantity and current is given. This is the conversion coefficient between raw coal quantity and electric current. This is the no-load current of the coal mill.
[0052] The moisture content leaving the pulverized coal is calculated as follows: (15) In the formula: The elemental composition of the pulverized coal entering the drying zone. The elemental composition of the raw coal entering the mill.
[0053] Maximum evaporation rate of air moisture: (16) In the formula: This represents the maximum evaporation rate of air moisture (kg / s). The moisture content of the air under standard conditions. The air is diverted at the entrance.
[0054] Maximum evaporation rate of pulverized coal moisture: (17) In the formula: This represents the maximum evaporation rate of moisture from pulverized coal (kg / s). The moisture content of the raw coal entering the platform, This refers to the moisture content of the dried coal powder.
[0055] Evaporated water mass flow rate: (18) In the formula: The mass flow rate of evaporated water is (kg / s). The moisture content of the raw coal entering the platform, This refers to the moisture content of the dried coal powder.
[0056] Mass flow rate of pulverized coal after drying: (19) In the formula: This represents the mass flow rate of pulverized coal after drying.
[0057] Moisture content of dried coal powder: (20) In the formula: Moisture content of dried coal powder This represents the percentage of coarse coal powder with a particle size greater than 90 μm in the total coal powder. The outlet temperature is (K).
[0058] Primary air mass flow rate at outlet: (twenty one) In the formula: The mass flow rate of the primary air at the outlet.
[0059] S206, Set up the energy conservation module The energy input to the coal mill mainly includes: the internal energy of the inlet raw coal, the internal energy of the inlet primary air, and the electrical power of the coal mill. The energy output or consumption mainly includes: the internal energy of the outlet pulverized coal, the internal energy of the outlet primary air, the energy consumed by moisture evaporation, the energy consumption due to the coal mill's efficiency, and the enthalpy change caused by temperature variations. Considering the energy conservation relationship between the above variables, this module is mainly controlled by the following formula.
[0060] Formula for calculating temperature change at the coal mill outlet using energy conservation: (twenty two) In the formula: The specific heat capacity of the coal mill. The specific heat capacity of the primary air at constant pressure at the inlet [J / (kg·℃)], The inlet primary air temperature (K) The specific heat capacity of the inlet raw coal at constant pressure [J / (kg·℃)], The inlet temperature of the raw coal (K) is generally taken as room temperature, i.e., 298.15K. The specific heat capacity of the primary air at constant pressure at the outlet [J / (kg·℃)], This represents the enthalpy change per unit mass of water evaporated (J / kg). The specific heat capacity of pulverized coal at constant pressure [J / (kg·℃)] is given. This represents the percentage of power loss in the coal mill. This refers to the rated power (W) of the coal mill.
[0061] Coal mill power: (twenty three) In the formula: This represents the percentage of power consumed per unit mass of pulverized coal. The percentage of power consumed per unit mass of raw coal. This represents the no-load power (W) of the coal mill.
[0062] S207, Setting the pressure loss module The pressure changes inside the coal mill are mainly due to changes in primary air density and pressure loss caused by the primary air blowing up coal powder, which is controlled by the following formula.
[0063] Coal mill pressure difference: (twenty four) In the formula: The pressure difference (Pa) is the pressure difference of the coal mill. The density enhancement factor of the primary wind pressure difference. The primary wind pressure loss is measured in Pa. This is the static pressure coefficient of the coal mill.
[0064] Primary air pressure loss: (25) In the formula: This is the coefficient of friction loss under no-load conditions.
[0065] S208, Encapsulated coal mill model The dynamic input data collected by the sensors is incorporated into the boundary conditions at the inlet of the coal mill, and the measured values of the output data are saved in a data table file. This encapsulates the coal mill model into a one-dimensional dynamic model from inlet to outlet, which more closely approximates the actual structure of the coal mill.
[0066] S209. Setting internal parameters of the coal mill model The internal parameters of the coal mill model were identified using a genetic algorithm. The input parameters of the coal mill—inlet coal flow rate, inlet air flow rate, and inlet air temperature—were input into the model. The genetic algorithm parameter identification method was then used to ensure that the coal mill output equals the measured values of the outlet temperature and the coal mill current, thus obtaining the unknown intermediate parameters of the model. After setting the identification results into the model, static and dynamic data under different operating conditions were used for verification.
[0067] like Figure 1 As shown, Figure 1 This is a schematic diagram of the internal space of a coal mill. Its approximate spatial location can be divided into the following areas: the grinding zone, also known as the platform zone, which involves the process of raw coal entering the coal mill and being ground into coal powder; the transportation zone, which involves the screening process for the fineness of the coal; and the drying zone, which involves the heat and mass transfer process between the coal powder and the primary air.
[0068] like Figure 2 As shown, Figure 2 It is a schematic diagram of the established digital twin model of the coal mill, including key information in the working process of the coal mill such as the inlet and outlet materials, coal powder fineness, and separator speed.
[0069] like Figure 3 As shown, Figure 3 This is a flowchart of parameter identification for the established digital twin model of the coal mill. A genetic algorithm is used to identify various intermediate parameters in the digital twin model of the coal mill.
[0070] S3. Establish a coal mill fault model based on the parameter changes of faults such as coal mill failure, coal mill spontaneous combustion, and coal mill blockage.
[0071] S301. Set up a coal shortage fault model (the model remains unchanged compared to the digital twin simulation model). Under the coal shortage fault, the coal feed rate drops sharply while the inlet and outlet air parameters remain unchanged. Based on this characteristic, the inlet coal rate in the boundary conditions of the digital twin simulation model established by S2 is modified to fall outside the normal range. This abnormal inlet coal rate is input into the simulation model to obtain the data output by the simulation model; thus, the coal mill coal shortage fault model is obtained.
[0072] like Figure 4 As shown, Figure 4 This demonstrates the parameter changes of the coal mill under a coal shortage fault. When a coal shortage fault occurs, the coal inlet flow rate of the coal mill decreases, the coal mill current decreases, and the outlet temperature increases.
[0073] S302, Setting up a spontaneous combustion fault model Under spontaneous combustion faults, some pulverized coal in the coal mill spontaneously combusts. Based on this characteristic, the mass and energy equations in the digital twin simulation model established in S2 are modified to obtain the spontaneous combustion fault model of the coal mill. Specifically, the rate of change of pulverized coal in the mass equation in S206 needs to be increased by a term representing the mass reduction due to spontaneous combustion of pulverized coal, and the energy equation in S207 needs to be increased by a term representing the heat generated by spontaneous combustion of pulverized coal in the energy change of the coal mill system.
[0074] like Figure 5 As shown, Figure 5 This demonstrates the parameter changes of the coal mill under spontaneous combustion failure. When spontaneous combustion failure occurs, some of the coal powder in the coal mill spontaneously combusts, and the outlet temperature rises.
[0075] The revised mass equation is as follows: (26) Among them, coefficient This indicates the proportion of spontaneously combusted pulverized coal in the coal mill. The revised energy equation is as follows: (27) (28) in, It is the heat capacity of raw coal, calculated using Mendeleev's empirical formula.
[0076] S303, Set up a coal blockage fault model Under coal blockage faults, the grinding rollers and pulverized coal outlet become clogged, resulting in a low conversion rate of raw coal to pulverized coal. Based on this characteristic, the mass equation in the digital twin simulation model established in S2 is modified to obtain the coal blockage fault model for the coal mill. Specifically, the raw coal conversion rate in the mass equation in S206 needs to be multiplied by an additional variation coefficient compared to the normal operating condition.
[0077] like Figure 6 As shown, Figure 6 This demonstrates the parameter changes of the coal mill under coal blockage. When coal blockage occurs, the amount of coal inside the coal mill increases, the coal mill current increases, and the outlet temperature decreases.
[0078] The revised mass equation is as follows: (29) Among them, coefficient This indicates the change in the conversion efficiency of raw coal to pulverized coal. S4. Combining fault models, deep learning, and transfer learning methods, a coal mill fault diagnosis system is constructed.
[0079] S401. Generate sample dataset Based on the fault model established in S3, sample data are generated for coal shortage fault, spontaneous combustion fault, and coal blockage fault.
[0080] Based on the digital twin simulation model established in S2, sample data under normal conditions is generated.
[0081] The above data, combined with actual normal operation and fault data, yielded a sample dataset.
[0082] S402, Training Deep Learning Models Deep learning models are trained using sample datasets, and parameter transfer is performed for different situations based on transfer learning theory to achieve fault diagnosis of coal mills under various working conditions.
[0083] like Figure 7 As shown, Figure 7 This is the neural network structure of a deep learning fault diagnosis model. In order to improve the model's global interactive modeling ability, local transient feature sensitivity, and long sequence parallel processing ability, the encoding layers of three models, CNN, Transformer, and BiLSTM, are combined to enhance the model's performance.
[0084] like Figure 8 As shown, Figure 8 This describes the parameter transfer process for the deep learning model established in this case. During model training, the deep learning model is first pre-trained based on simulation data and existing historical labeled data to obtain initial model parameters with general feature extraction capabilities. Then, using parameter transfer methods from transfer learning, the parameters of the CNN, BiLSTM, and Transformer encoding layers in the pre-trained model are transferred as initialization parameters to the target operating condition model. The model is then fine-tuned using a small amount of sample data from the target device or operating condition. This parameter transfer strategy effectively improves the model's adaptability across scenarios, devices, and operating conditions, significantly enhancing its robustness under different circumstances.
[0085] The inputs to the deep learning model are: inlet coal quantity, inlet air quantity, inlet air temperature, mill current, outlet temperature, outlet pulverized coal quantity, coal quantity ratio (the ratio of outlet pulverized coal quantity to inlet coal quantity), air-coal ratio (the ratio of inlet air quantity to inlet coal quantity), temperature ratio (the ratio of outlet air temperature to inlet air temperature), and current ratio (the ratio of mill current to inlet coal quantity). The output is one of the following four states: coal shortage fault, spontaneous combustion fault, coal blockage fault, or normal operation.
[0086] S5. The outputs of the deep learning model and digital twin simulation model built in S4 are summarized into a visualization interface and refreshed in real time according to the system's operation. The visualization module extracts the outputs of the sensors in the system and the digital twin simulation model established in S2, and displays the real-time air and coal flow, pressure distribution and calculation results of the coal mill through 3D graphics. Users can view the operating status of the coal mill and possible faults in real time through the interactive interface to achieve the best monitoring effect.
[0087] Preferably, in this invention, S1 collects data from various measuring points for more than one year and uses data processing methods to remove outliers from the data.
[0088] Preferably, in this invention, S2 chooses Modelica language for modeling to consider system scalability.
[0089] Preferably, in this invention, the data preprocessing in S201 uses an adaptive Kalman filter to process the data.
[0090] Preferably, in this invention, the dynamic verification experiment of the coal mill in S209 adopts a step response test of primary air volume, primary air temperature, and inlet coal volume.
[0091] Preferably, in this invention, the deep learning model in S402 adopts a deep learning model that combines three encoding layers: CNN, BiLSTM, and Transformer.
[0092] Preferably, in this invention, the refresh time of the fault diagnosis system constructed in S5 can be appropriately adjusted according to the sampling frequency of the sensors equipped with different coal mills and their load change characteristics.
[0093] Preferably, in this invention, the visualization module considered in S5 can be implemented using the Vue.js framework and other frameworks that can achieve similar functions, allowing users to view data updates in real time.
[0094] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A coal mill fault modeling method based on the coupling of deep learning and digital twin technology, characterized in that, The method includes the following steps: A digital twin simulation model of a coal mill was built, and the historical parameters at the inlet and outlet of the coal mill were used as inputs and outputs, respectively, to identify the model parameters. Set the fault types of the coal mill, adjust the parameters of the coal mill during operation according to the set fault types, and input the adjusted parameters into the digital twin simulation model to generate fault data corresponding to various fault types. A deep learning model of fault types is built by using fault data as input and fault type as output. The fault type deep learning model is trained using data from the database, thereby realizing the modeling of the fault type deep learning model.
2. The coal mill fault modeling method based on the coupling of deep learning and digital twin technology as described in claim 1, characterized in that, The defined fault types of the coal mill include coal shortage fault, spontaneous combustion fault, and coal blockage fault; the deep learning model for training the fault types adopts a parameter transfer strategy.
3. The coal mill fault modeling method based on the coupling of deep learning and digital twin technology as described in claim 2, characterized in that, The digital twin simulation model includes an inlet boundary condition module, an outlet boundary condition module, a mass conservation module, and an energy conservation module. The inlet boundary condition module sets the input boundary conditions of the digital twin simulation model, the outlet boundary condition module sets the output boundary conditions, the mass conservation module sets the relationship between various physical quantities while maintaining mass conservation during the operation of the coal mill, and the energy conservation module sets the relationship between various physical quantities while maintaining mass conservation during the operation of the coal mill.
4. The coal mill fault modeling method based on the coupling of deep learning and digital twin technology as described in claim 3, characterized in that, The fault data under the coal shortage fault is generated by the digital twin simulation model of the coal mill, which uses the amount of coal entering the coal mill that exceeds the preset normal range as the fault data.
5. The coal mill fault modeling method based on the coupling of deep learning and digital twin technology as described in claim 3, characterized in that, The fault data acquisition method under the spontaneous combustion fault is as follows: Adjust the calculation formula of the raw coal mass change in the grinding zone in the mass conservation module and the calculation formula of the temperature change at the coal mill outlet in the energy conservation module in the digital twin simulation model, thereby adjusting the digital twin simulation model, and using the adjusted digital twin simulation model to generate fault data under the spontaneous combustion fault.
6. The coal mill fault modeling method based on the coupling of deep learning and digital twin technology as described in claim 5, characterized in that, The adjusted calculation formula for the temperature change at the coal mill outlet in the energy conservation module is as follows: in, For the raw coal on the adjusted grinding area platform, It is the mass flow rate of pulverized coal. The mass flow rate of the raw coal being ground. This indicates the proportion of spontaneously combusted pulverized coal in the coal mill. It represents the mass of raw coal currently present on the grinding zone platform, and t represents the time of temperature change.
7. A coal mill fault modeling method based on the coupling of deep learning and digital twin technology as described in claim 5 or 6, characterized in that, The adjusted calculation formula for the temperature change at the coal mill outlet in the energy conservation module is as follows: in, This is the adjusted temperature at the coal mill outlet. The static pressure coefficient of the coal mill. The specific heat capacity of the primary air at constant pressure at the inlet. The inlet primary air temperature, The specific heat capacity of the raw coal at constant pressure. The temperature of the incoming raw coal. The specific heat capacity of the primary air at constant pressure at the outlet. For the primary air mass flow rate at the outlet, It is the mass of evaporated water. This represents the enthalpy change per unit mass of water evaporated. The specific heat capacity of pulverized coal at constant pressure for export. This represents the mass flow rate of pulverized coal after drying. The density enhancement factor of the primary wind pressure difference. This refers to the rated power of the coal mill. To ensure the quality of the raw coal entering the coal mill, It's about the quality of the coal powder on the platform. It was the quality of coal dust blown up by the wind. It is the heat capacity of raw coal. It refers to the carbon content in coal. It refers to the hydrogen content in coal. It refers to the oxygen content in coal. It refers to the sulfur content in coal. It refers to the moisture content in coal.
8. A coal mill fault modeling method based on the coupling of deep learning and digital twin technology as described in claim 5 or 6, characterized in that, The method for obtaining fault data under the coal blockage fault is as follows: adjust the calculation formula of the mass flow rate of the ground raw coal in the digital twin simulation model, thereby changing the digital twin simulation model, and use the changed digital twin simulation model to generate fault data under the coal blockage fault.
9. The coal mill fault modeling method based on the coupling of deep learning and digital twin technology as described in claim 8, characterized in that, The formula for calculating the mass flow rate of the raw coal being ground in the digital twin simulation model has been adjusted, and the adjusted formula is as follows: in, This is the adjusted mass flow rate of the ground raw coal. It is the percentage of coal dust blown into the air by the wind once per second. It represents the percentage of raw coal being ground per second on the platform. It represents the change in the conversion efficiency of raw coal to pulverized coal.
10. A coal mill fault diagnosis method based on the coupling of deep learning and digital twin technology, characterized in that, This method uses a deep learning model of fault types established by a coal mill fault modeling method based on the coupling of deep learning and digital twin technology as described in any one of claims 1-9 to predict the faults of the coal mill.