A digital twin interaction method for grain depot operation and maintenance

Abstract

The application belongs to the technical field of digital twinning and intelligent operation and maintenance, and particularly relates to a digital twinning interaction method for grain depot operation and maintenance, which collects grain depot environment, grain state and equipment operation parameters through a sensor network, completes data loss through a grain pile physical graph sparse completion algorithm to obtain high-quality interaction data, constructs a geometry, heat and humidity coupling transmission, gas diffusion and behavior model based on the data, maps the model into a multi-dimensional digital twinning model of the grain depot through a cooperative evolution algorithm to output a full evaluation state, locates an abnormal root cause by using a Fourier neural operator to rollingly predict a grain pile and equipment state field at the next moment and combining a causal inference algorithm, and finally generates an optimal maintenance strategy through a water droplet optimization algorithm based on a fault location and root cause. The method realizes intelligent operation and maintenance of the grain depot in the whole process, improves fault handling efficiency and reduces grain storage risks.

Description

Technical Field

[0001] This invention pertains to the field of digital twin and intelligent operation and maintenance technology, specifically a digital twin interaction method for grain warehouse operation and maintenance. Background Technology

[0002] Grain storage security is a core link in ensuring the stability of the grain supply chain, and the quality of grain warehouse operation and maintenance directly affects the quality and loss rate of stored grain. However, current grain warehouse operation and maintenance relies on manual inspections and experience-based judgment, which suffers from problems such as low efficiency and incomplete coverage, making it difficult to cope with large-scale and sophisticated storage needs.

[0003] Existing models mostly focus on the temperature field and fail to integrate multi-dimensional factors such as heat and humidity coupling, gas diffusion, and pest growth, making it difficult to reflect the complex dynamic changes in grain silos. In terms of fault handling, traditional methods can only passively respond to faults that have occurred, rarely have the ability to predict the state field of grain piles and equipment in advance, and cannot accurately locate the root cause of anomalies, resulting in strong blindness in the maintenance work of grain silos.

[0004] Therefore, a digital twin interaction method for grain warehouse operation and maintenance is needed to solve the above problems. Summary of the Invention

[0005] To address the technical problems mentioned in the background, this invention provides a digital twin interaction method for grain warehouse operation and maintenance.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] This invention provides a digital twin interaction method for grain warehouse operation and maintenance, comprising the following steps:

[0008] Step 1: Grain warehouse data acquisition: Through a sensor network deployed in the physical grain warehouse, the grain warehouse environmental parameters, grain status parameters and equipment operation parameters are collected. Based on the sparse completion algorithm of the grain pile physical map, the data missing caused by sparse or offline sensors in the edge area of ​​the grain warehouse is completed to obtain high-quality interactive data.

[0009] Step 2: Digital Twin Model Construction: Based on high-quality interactive data, geometric models, thermal-humidity coupling and transfer models, gas diffusion models, and behavioral models are established respectively. All models are mapped into a multi-dimensional digital twin model of the grain warehouse through a co-evolution algorithm, and the full assessment status of the grain warehouse is output.

[0010] Step 3, Fault Prediction: The state of the grain silo is predicted in a rolling manner by using the Fourier neural operator to obtain the state field of the grain pile and the state field of the equipment at the next moment. Then, the root cause of the abnormality of the grain silo is located by using the causal inference algorithm.

[0011] Step 4: Strategy Generation: Based on the fault location and root cause results, the controller obtains the optimal maintenance strategy for the current grain warehouse by modifying the Waterdrop Optimization Algorithm.

[0012] In this application, based on step one, a sensor network deployed within the physical grain warehouse is used to collect environmental parameters, grain status parameters, and equipment operating parameters. A sparse completion algorithm based on the physical map of the grain pile is then used to complete the data missing due to sparse or offline sensors in the grain warehouse edge area, resulting in high-quality interactive data. The specific steps are as follows:

[0013] Environmental data is collected in real time by environmental sensors, including temperature, humidity, and gas concentration data at multiple points inside the silo. Grain condition data, including internal temperature, humidity, biomass concentration, and moisture content, is collected in real time by temperature and humidity sensors, fluorescence biosensors, and moisture content sensors. Images of the grain pile surface are captured in real time by high-definition industrial cameras. Strain sensors and demodulators monitor strain changes at various points in the silo in real time to obtain the lateral pressure distribution and structural deformation of the silo, thus obtaining silo structural data. Data information from various silo equipment, including fans, refrigeration units, level gauges, and conveyors, is read by the equipment controller. Data information includes rotational speed, operating power, voltage, vibration frequency, and measurement accuracy deviation. All data are integrated into comprehensive silo data.

[0014] The grain pile is uniformly divided into grids of equal size, which are then set as graph nodes in the grain pile physical map. Physical sensors are then set as observation nodes, each containing spatial coordinates and real-time readings. Virtual nodes are created at arbitrary locations within the grain pile where no sensors are present. The weight Ee of the physical association edges between nodes is obtained, and its calculation logic is as follows: , where ij represents the indices of any two nodes, and exp is the natural exponential function. Let be the average bulk density of the grain pile, and rc be the spatial Euclidean distance. A transmission coefficient is preset for the grain pile to make the weight result fit the actual transmission law of the grain pile. The value is 0.8 in this grain warehouse scenario; a physical graph of the grain pile is constructed using graph nodes and physical association edge weights.

[0015] The observed node data in the physical graph of the grain pile is propagated to the virtual node data through Chebyshev graph convolution to obtain the virtual node data. The calculation logic is as follows: Where m is the convolution order number and M is the total number of convolution orders. These are the learnable weight coefficients of the Chebyshev polynomial. For Chebyshev polynomials, To normalize the graph Laplace matrix, virtual node data is embedded into the comprehensive data of the grain warehouse to obtain high-quality interactive data.

[0016] In this application, based on step two, a three-dimensional geometric model, a heat and moisture coupling transfer model, a gas diffusion and convection model, and a behavioral model are established based on high-quality interactive data. All models are mapped to a multi-dimensional digital twin model of the grain warehouse through a co-evolution algorithm, and the overall assessment status of the grain warehouse is output. The specific steps are as follows:

[0017] Based on BIM or 3D laser scanning technology, data information of grain silos and their equipment is assigned globally unique identifiers and semantically linked with static information in a database. This static information includes historical fault records, equipment specifications, models, purchase dates, and maintenance manuals, resulting in a 3D model of the grain silo. A multi-view stereo matching algorithm is used to process the grain pile surface image, generating a high-density grain pile surface point cloud. Then, based on the flow data of grain pile entry and exit, the height value of the grain pile voxels is updated in real time using a formula. The calculation logic is as follows: Where t is the time point number, and These represent the grain flow rates into and out of the warehouse, respectively; PE represents the grain bulk density; and ST represents the bottom area of ​​the grain silo. The settling rate of the grain pile. To update the time step, the high-density grain pile surface point cloud and real-time height values ​​are embedded into the 3D model of the grain silo to obtain a 3D geometric model.

[0018] Extract the preset respiration heat coefficient from the database, and obtain the grain's own respiration heat based on the grain's bulk density and respiration heat coefficient. The calculation logic is as follows: ,in To preset the respiratory heat coefficient, To determine the carbon dioxide emission rate of grain, the maximum specific growth rate under preset optimal conditions was extracted from the database. The real-time specific growth rate of mold can be obtained through a formula. The calculation logic of its formula is as follows: ,in and The growth rate is a function of temperature and water activity, where Te and ar are temperature and water, respectively, calculated using real-time specific growth rate and mold biomass concentration. The calculation logic for obtaining heat generated by mold growth is as follows: Where YQ is the heat production and yield coefficient; integrating the heat from grain respiration and the heat from mold growth, we obtain the heat source term HF. Embedding the heat source term into the temperature conduction equation yields the heat conduction equation for the grain pile temperature field, the calculation logic of which is as follows: Where Pa and Ce are the effective bulk density and effective specific heat capacity of the grain pile, respectively. For effective thermal conductivity, The term represents heat conduction, describing heat transfer caused by the temperature gradient. Kriging interpolation is used to fuse data points on grain moisture content, generating a moisture content distribution field (DC) for the grain pile. This distribution field is then embedded into the moisture diffusion formula to obtain the effective moisture diffusion coefficient. The calculation logic of its formula is as follows: ,in The effective moisture diffusion coefficient is then embedded into the heat conduction equation of the grain pile temperature field to obtain a heat and moisture coupling transfer model.

[0019] The three-dimensional spatial extent of the grain pile is obtained based on a three-dimensional geometric model. This three-dimensional spatial extent is divided into a uniform three-dimensional grid, with each grid being a cube. The porosity of the three-dimensional grid is then obtained. Its calculation logic is as follows: Where YC is the total solid volume of grain particles within the grid, TY is the total volume of the 3D grid, and ce is the grid number. The porosity of all 3D grids is assembled into a 3D array to obtain a non-uniform porosity field. The gas diffusion coefficient of a grain pile is obtained through porosity, and the calculation logic is as follows: ,in Tortuosity describes the degree of curvature of the gas diffusion path within pores, derived from empirical formulas. , Here, DU is an empirical constant, representing the preset binary molecular diffusion coefficient of gas in air, extracted from a database. A gas diffusion convection model is constructed based on porosity and the gas diffusion coefficient, with the following construction logic: ,in The molar concentration of the gas in the pores. For diffusion term, For concentration gradient;

[0020] The corresponding pest density is obtained by acquiring the corresponding 3D grid, and the average pest density of the grain pile is calculated by averaging the pest densities. Based on the average pest density JO and the real-time mold specific growth rate, the average pest density of the grain pile is calculated. A behavior model of the grain pile is established, and the logic for constructing the model is as follows: ,in and These are the response functions for temperature and moisture, respectively;

[0021] Based on a three-dimensional geometric model, several three-dimensional meshes are divided. Through a co-evolution algorithm, all models are unified into a state tensor in several three-dimensional meshes. The dimensions of the state tensor include the spatial coordinates, time dimension, and channel dimension corresponding to the three-dimensional mesh. The channel dimension is divided into several sub-channels, and the state information of different models is embedded into the corresponding sub-channels to obtain the equipment attribute channel, grain pile physical field channel, and behavior state channel. The state tensor is mapped to the digital twin space to obtain a multi-dimensional digital twin model of the grain warehouse. The full evaluation state of the grain warehouse is output through the multi-dimensional digital twin model of the grain warehouse.

[0022] In this application, based on step three, the grain pile state field and equipment state field at the next moment are obtained by rolling prediction of the overall state of the grain silo using the Fourier neural operator model. Then, the root cause of the anomaly is located through a causal inference algorithm. The calculation logic is as follows:

[0023] The full assessment state of the grain warehouse is input into the Fourier neural operator model, and the full assessment state of the grain warehouse is mapped to a high-dimensional feature space to obtain a three-dimensional spatial dimension. The three-dimensional spatial dimension is then transformed by a Fourier convolutional layer to obtain high-frequency modal features. The high-frequency modal features are then decoded by a decoding layer to output the prediction results of the grain pile state field and equipment state field at the next moment.

[0024] The difference between the data in the grain pile state field and the preset grain pile data in the corresponding database is used to obtain the grain pile data error. If the absolute value of the grain pile data error is greater than the preset grain pile error threshold, the corresponding three-dimensional grid coordinates and grain pile abnormal parameters are marked. The difference between the data in the equipment state field and the preset equipment data in the corresponding database is used to obtain the equipment data error. If the absolute value of the equipment data error is greater than the preset equipment error threshold, the corresponding equipment is marked as an abnormal candidate, and the equipment number and corresponding physical coordinates are recorded.

[0025] By using equipment operating parameters as intervention variables, abnormal grain pile parameters as outcome variables, and external temperature and humidity, mold, and pests as confounding variables, a causal graph is generated.

[0026] Extract the variable data associated with the abnormal regions and construct a structured dataset. Calculate the causal effect of each intervention variable H on the abnormal outcome S. The calculation logic is as follows: ,in Let do be the expectation operator, and do be the causal interference predictor. To force intervention variables to take outliers, The variables for mandatory intervention are set to normal values; the variables are sorted from largest to smallest according to their causal effects, and the first variable is set as the root cause of the abnormality. The fault location corresponding to the root cause of the abnormality is obtained based on the three-dimensional grid coordinates, and the fault location and the root cause of the abnormality are output and sent to the controller.

[0027] In this application, based on step four, the controller obtains the optimal maintenance strategy for the current grain silo by modifying the droplet optimization algorithm based on the fault location and root cause results. The specific steps are as follows:

[0028] Based on the root causes of failures and grain warehouse operation and maintenance specifications, an executable set of maintenance strategy solutions is constructed. The execution target, resource consumption, effect, and time consumption of each maintenance strategy are clearly marked. By modifying the droplet optimization algorithm, the set of maintenance strategy solutions is transformed into a droplet population. The objective function value of each droplet is obtained, and the flow rate is updated until a preset number of iterations is reached. The calculation logic for the objective function value is as follows: ,in The total maintenance cost of implementing the strategy, To maintain the total cost of fully executing the strategy, Total time consumed The total time taken to execute all actions sequentially. The effectiveness of fault elimination is represented by the percentage of grain temperature in area R5 that has returned to normal, with 1 indicating complete recovery. To maximize the fault elimination effect 1, where b is the scheme number, the calculation logic for its update flow rate is as follows: ,in Let v be the number of iterations and v be the flow rate of the water droplet. for -1 The objective function value of the droplet is used; the smaller the value, the better the strategy. Calculate the objective function for each updated droplet, select the optimal droplet with the smallest current objective function, record its corresponding strategy, and execute it.

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] Achieve full-dimensional state perception by integrating geometric, thermal-humid coupling and transfer, gas diffusion and behavior models, and construct a multi-dimensional digital twin model through a collaborative evolution algorithm, covering the physical field and equipment of the grain warehouse and the biological state, and outputting the full assessment state of the grain warehouse.

[0031] By modifying the waterdrop optimization algorithm to generate the optimal maintenance strategy, the maintenance cost, time consumption and fault elimination effect are balanced, and the blind operation and maintenance of grain silos based on the causes and locations of faults is reduced, thus minimizing the waste of manpower and resources. Attached Figure Description

[0032] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. The following drawings are not drawn to scale according to the actual size, but are intended to illustrate the main idea of ​​the present invention.

[0033] Figure 1 This is a diagram illustrating the method steps of the present invention. Detailed Implementation

[0034] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are also within the scope of protection of the present invention.

[0035] Please refer to Figure 1 As shown, this invention provides a digital twin interaction method for grain warehouse operation and maintenance, comprising the following steps:

[0036] Step 1: Grain warehouse data acquisition: Through a sensor network deployed in the physical grain warehouse, the grain warehouse environmental parameters, grain status parameters and equipment operation parameters are collected. Based on the sparse completion algorithm of the grain pile physical map, the data missing caused by sparse or offline sensors in the edge area of ​​the grain warehouse is completed to obtain high-quality interactive data.

[0037] Step 2: Digital Twin Model Construction: Based on high-quality interactive data, geometric models, thermal-humidity coupling and transfer models, gas diffusion models, and behavioral models are established respectively. All models are mapped into a multi-dimensional digital twin model of the grain warehouse through a co-evolution algorithm, and the full assessment status of the grain warehouse is output.

[0038] Step 3, Fault Prediction: The state of the grain silo is predicted in a rolling manner by using the Fourier neural operator to obtain the state field of the grain pile and the state field of the equipment at the next moment. Then, the root cause of the abnormality of the grain silo is located by using the causal inference algorithm.

[0039] Step 4: Strategy Generation: Based on the fault location and root cause results, the controller obtains the optimal maintenance strategy for the current grain warehouse by modifying the Waterdrop Optimization Algorithm.

[0040] In this application, based on step one, a sensor network deployed within the physical grain warehouse is used to collect environmental parameters, grain status parameters, and equipment operating parameters. A sparse completion algorithm based on the physical map of the grain pile is then used to complete the data missing due to sparse or offline sensors in the grain warehouse edge area, resulting in high-quality interactive data. The specific steps are as follows:

[0041] Environmental data is collected in real time by environmental sensors, including temperature, humidity, and gas concentration data at multiple points inside the silo. Grain condition data, including internal temperature, humidity, biomass concentration, and moisture content, is collected in real time by temperature and humidity sensors, fluorescence biosensors, and moisture content sensors. Images of the grain pile surface are captured in real time by high-definition industrial cameras. Strain sensors and demodulators monitor strain changes at various points in the silo in real time to obtain the lateral pressure distribution and structural deformation of the silo, thus obtaining silo structural data. Data information from various silo equipment, including fans, refrigeration units, level gauges, and conveyors, is read by the equipment controller. Data information includes rotational speed, operating power, voltage, vibration frequency, and measurement accuracy deviation. All data are integrated into comprehensive silo data.

[0042] The grain pile is uniformly divided into grids of equal size, which are then set as graph nodes in the grain pile physical map. Physical sensors are then set as observation nodes, each containing spatial coordinates and real-time readings. Virtual nodes are created at arbitrary locations within the grain pile where no sensors are present. The weight Ee of the physical association edges between nodes is obtained, and its calculation logic is as follows: , where ij represents the indices of any two nodes, and exp is the natural exponential function. Let be the average bulk density of the grain pile, and rc be the spatial Euclidean distance. A transmission coefficient is preset for the grain pile to make the weight result fit the actual transmission law of the grain pile. The value is 0.8 in this grain warehouse scenario; a physical graph of the grain pile is constructed using graph nodes and physical association edge weights.

[0043] The observed node data in the physical graph of the grain pile is propagated to the virtual node data through Chebyshev graph convolution to obtain the virtual node data. The calculation logic is as follows: Where m is the convolution order number and M is the total number of convolution orders. These are the learnable weight coefficients of the Chebyshev polynomial. For Chebyshev polynomials, To normalize the graph Laplacian matrix, virtual node data is embedded into the comprehensive data of the grain silo to obtain high-quality interactive data. It should be noted that Chebyshev graph convolution is an efficient approximation of spectral domain graph convolutional neural network. It achieves efficient completion of sparse data through polynomial approximation and is adapted to the graph structure data of grain piles.

[0044] In this application, based on step two, a three-dimensional geometric model, a heat and moisture coupling transfer model, a gas diffusion and convection model, and a behavioral model are established based on high-quality interactive data. All models are mapped to a multi-dimensional digital twin model of the grain warehouse through a co-evolution algorithm, and the overall assessment status of the grain warehouse is output. The specific steps are as follows:

[0045] Based on BIM or 3D laser scanning technology, data information of grain silos and their equipment is assigned globally unique identifiers and semantically linked to static information in a database. This static information includes historical fault records, equipment specifications, models, purchase dates, and maintenance manuals, resulting in a 3D model of the grain silo. It should be noted that during the grain loading stage, grain flow is monitored in real-time using belt scales and level gauges. The grain pile height increases linearly with the loading volume, and decreases uniformly during the unloading stage. Locally, funnel-shaped depressions form due to grain flow, necessitating the creation of a dynamic grain silo. A multi-view stereo matching algorithm is used to process the grain pile surface image, generating a high-density grain pile surface point cloud. Then, based on the grain pile loading and unloading flow data, the height value of the grain pile voxels is updated in real-time using a formula. The calculation logic is as follows: Where t is the time point number, and These represent the grain flow rates into and out of the warehouse, respectively; PE represents the grain bulk density; and ST represents the bottom area of ​​the grain silo. The settling rate of the grain pile. To update the time step, the high-density grain pile surface point cloud and real-time height values ​​are embedded into the 3D model of the grain silo to obtain a 3D geometric model.

[0046] Extract the preset respiration heat coefficient from the database, and obtain the grain's own respiration heat based on the grain's bulk density and respiration heat coefficient. The calculation logic is as follows: ,in To preset the respiratory heat coefficient, To determine the carbon dioxide emission rate of grain, the maximum specific growth rate under preset optimal conditions was extracted from the database. The real-time specific growth rate of mold can be obtained through a formula. The calculation logic of its formula is as follows: ,in and The growth rate is a function of temperature and water activity, where Te and ar are temperature and water, respectively, calculated using real-time specific growth rate and mold biomass concentration. The calculation logic for obtaining heat generated by mold growth is as follows: Where YQ is the heat production rate coefficient; it should be noted that during the growth and reproduction of mold, it metabolizes and releases heat by decomposing organic matter in the grain. The heat production is directly proportional to the biomass and specific growth rate of the mold, and the maximum specific growth rate is obtained from microbiology literature; by integrating the heat of respiration of the grain itself and the heat production of mold growth, the heat source term HF is obtained. Embedding the heat source term into the temperature conduction equation, the heat conduction equation of the grain pile temperature field is obtained, and its calculation logic is as follows: Where Pa and Ce are the effective bulk density and effective specific heat capacity of the grain pile, respectively. For effective thermal conductivity, The term represents heat conduction, describing heat transfer caused by the temperature gradient. Kriging interpolation is used to fuse data points on grain moisture content, generating a moisture content distribution field (DC) for the grain pile. This distribution field is then embedded into the moisture diffusion formula to obtain the effective moisture diffusion coefficient. The calculation logic of its formula is as follows: ,in The effective moisture diffusion coefficient is then embedded into the heat conduction equation of the grain pile temperature field to obtain a heat and moisture coupling transfer model.

[0047] The three-dimensional spatial extent of the grain pile is obtained based on a three-dimensional geometric model. This three-dimensional spatial extent is divided into a uniform three-dimensional grid, with each grid being a cube. The porosity of the three-dimensional grid is then obtained. Its calculation logic is as follows: Where YC is the total solid volume of grain particles within the grid, TY is the total volume of the 3D grid, and ce is the grid number. The porosity of all 3D grids is assembled into a 3D array to obtain a non-uniform porosity field. It should be noted that the total solid volume of grain particles is obtained by acquiring the particle size distribution of the corresponding actual grain variety and simulating it using a three-dimensional geometric model; the gas diffusion coefficient of the grain pile is obtained through porosity, and its calculation logic is as follows: ,in Tortuosity describes the degree of curvature of the gas diffusion path within pores, derived from empirical formulas. , Here, DU is an empirical constant, representing the preset binary molecular diffusion coefficient of gas in air, extracted from a database. A gas diffusion convection model is constructed based on porosity and the gas diffusion coefficient, with the following construction logic: ,in The molar concentration of the gas in the pores. For diffusion term, For concentration gradient;

[0048] The corresponding pest density is obtained by acquiring the corresponding 3D grid, and the average pest density of the grain pile is calculated by averaging the pest densities. Based on the average pest density JO and the real-time mold specific growth rate, the average pest density of the grain pile is calculated. A behavior model of the grain pile is established, and the logic for constructing the model is as follows: ,in and These are the response functions for temperature and moisture, respectively;

[0049] Based on a three-dimensional geometric model, several three-dimensional meshes are divided. Through a co-evolution algorithm, all models are unified into a state tensor in several three-dimensional meshes. The dimensions of the state tensor include the spatial coordinates, time dimension, and channel dimension corresponding to the three-dimensional mesh. The channel dimension is divided into several sub-channels, and the state information of different models is embedded into the corresponding sub-channels to obtain the equipment attribute channel, grain pile physical field channel, and behavior state channel. The state tensor is mapped to the digital twin space to obtain a multi-dimensional digital twin model of the grain warehouse. The full evaluation state of the grain warehouse is output through the multi-dimensional digital twin model of the grain warehouse.

[0050] In this application, based on step three, the grain pile state field and equipment state field at the next moment are obtained by rolling prediction of the overall state of the grain silo using the Fourier neural operator model. Then, the root cause of the anomaly is located through a causal inference algorithm. The calculation logic is as follows:

[0051] It should be noted that the Fourier neural operator model selects historical grain warehouse data, constructs training samples using the sliding window method, and each sample contains several input tensors for consecutive time steps, corresponding to the state field labels for future time steps. The AdamW optimizer is used for iterative training for several rounds.

[0052] The full assessment state of the grain warehouse is input into the Fourier neural operator model, and the full assessment state of the grain warehouse is mapped to a high-dimensional feature space to obtain a three-dimensional spatial dimension. The three-dimensional spatial dimension is then transformed by a Fourier convolutional layer to obtain high-frequency modal features. The high-frequency modal features are then decoded by a decoding layer to output the prediction results of the grain pile state field and equipment state field at the next moment.

[0053] The difference between the data in the grain pile state field and the preset grain pile data in the corresponding database is used to obtain the grain pile data error. If the absolute value of the grain pile data error is greater than the preset grain pile error threshold, the corresponding three-dimensional grid coordinates and grain pile abnormal parameters are marked. The difference between the data in the equipment state field and the preset equipment data in the corresponding database is used to obtain the equipment data error. If the absolute value of the equipment data error is greater than the preset equipment error threshold, the corresponding equipment is marked as an abnormal candidate, and the equipment number and corresponding physical coordinates are recorded.

[0054] By using equipment operating parameters as intervention variables, abnormal grain pile parameters as outcome variables, and external temperature and humidity, mold, and pests as confounding variables, a causal graph is generated.

[0055] Extract the variable data associated with the abnormal regions and construct a structured dataset. Calculate the causal effect of each intervention variable H on the abnormal outcome S. The calculation logic is as follows: ,in Let do be the expectation operator, and do be the causal interference predictor. To force intervention variables to take outliers, The variables for mandatory intervention are taken to normal values; the variables are sorted from largest to smallest according to their causal effects, and the first variable is set as the abnormal root cause. The fault location corresponding to the abnormal root cause is obtained based on the three-dimensional grid coordinates. The fault location and abnormal root cause results are output and sent to the controller. For example, if the temperature in the grain pile R5 area rises abnormally, the root cause is the abnormal decrease in the speed of the ventilation fan F3.

[0056] In this application, based on step four, the controller obtains the optimal maintenance strategy for the current grain silo by modifying the droplet optimization algorithm based on the fault location and root cause results. The specific steps are as follows:

[0057] Based on the root causes of failures and grain warehouse operation and maintenance specifications, an executable set of maintenance strategy solutions is constructed. The execution target, resource consumption, effect, and time consumption of each maintenance strategy are clearly marked. By modifying the droplet optimization algorithm, the set of maintenance strategy solutions is transformed into a droplet population. The objective function value of each droplet is obtained, and the flow rate is updated until a preset number of iterations is reached. The calculation logic for the objective function value is as follows: ,in The total maintenance cost of implementing the strategy, To maintain the total cost of fully executing the strategy, Total time consumed The total time taken to execute all actions sequentially. The effectiveness of fault elimination is represented by the percentage of grain temperature in area R5 that has returned to normal, with 1 indicating complete recovery. To maximize the fault elimination effect 1, where b is the scheme number, the calculation logic for its update flow rate is as follows: ,in Let v be the number of iterations and v be the flow rate of the water droplet. for -1 The objective function value of the droplet is used; the smaller the value, the better the strategy. Calculate the objective function for each updated droplet, select the optimal droplet with the smallest current objective function, record its corresponding strategy, and execute it.

[0058] All formulas described in this invention are dimensionless numerical calculation expressions. Dimensionlessness can be achieved using conventional methods such as standardization, which will not be elaborated upon here. The formulas are all based on extensive measured data from grain warehouse operations and maintenance, obtained through software simulation and fitting. They can reflect the true patterns of changes in the physical field of the grain warehouse, equipment operation, and grain condition evolution. The preset parameters in the formulas (such as the preset conduction coefficient of the grain pile, the heat production and yield coefficient, etc.) can be flexibly set by those skilled in the art according to the actual grain variety, grain warehouse specifications, and operating environment.

[0059] The embodiments of the present invention can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software form, it can be presented in whole or in part as a computer program product, which includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer (which may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device), all or part of the processes or functions described in the embodiments of the present invention, such as grain warehouse data acquisition, digital twin model construction, fault prediction, and maintenance strategy generation, are generated.

[0060] The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another (e.g., via wired or wireless means from one website, computer, server, or data center to another). The computer-readable storage medium may be any available medium accessible to a computer, or a data storage device such as a server or data center that includes one or more sets of available media. Available media include magnetic media (such as floppy disks, hard disks, and magnetic tapes), optical media (such as DVDs), or semiconductor media (such as solid-state drives).

[0061] It should be understood that in the various embodiments of the present invention, the sequence number of each process does not represent the order of execution. The execution order should be determined by its function and internal logic (such as completing data acquisition before model building, or realizing fault prediction before generating maintenance strategies), and should not constitute a limitation on the implementation process of the embodiments of the present invention.

[0062] Those skilled in the art will recognize that the algorithmic steps described in conjunction with the embodiments of this invention, such as Chebyshev graph convolutional completion, co-evolutionary modeling, Fourier neural operator prediction, causal inference localization, and water droplet optimization, can be implemented through electronic hardware or a combination of computer software and electronic hardware. The implementation of these functions depends on the specific application requirements of grain warehouse operation and maintenance and the design constraints of intelligent operation and maintenance. Professional technicians can adopt different methods to implement the functions for different application scenarios, but such implementation should not be considered beyond the scope of protection of this invention.

[0063] In the several embodiments provided by this invention, it should be understood that the disclosed grain warehouse digital twin interactive system and method can be implemented in other forms. For example, the division of the model modules is only a logical functional division, and in actual implementation, there may be other division methods (such as integrating the thermal-humidity coupling transfer model with the gas diffusion convection model), or multiple modules or components may be integrated into another system, or some features may be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between modules can be implemented through interfaces, which can be electrical, mechanical, or other forms.

[0064] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0065] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (personal computer, server, network device, etc.) to execute all or part of the steps (such as data acquisition, model building, fault prediction, strategy generation, etc.) of the grain warehouse digital twin interaction method described in the various embodiments of the present invention.

[0066] The aforementioned storage media include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, and optical disks. The above descriptions are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions (such as algorithm parameter adjustments, model module optimizations, etc.) that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A digital twin interaction method for grain warehouse operation and maintenance, characterized in that, Includes the following steps: Step 1: Collect grain warehouse environmental parameters, grain status parameters and equipment operating parameters through a sensor network deployed in the physical grain warehouse. Then, based on the sparse completion algorithm of the grain pile physical map, complete the data missing caused by sparse or offline sensors in the edge area of ​​the grain warehouse to obtain high-quality interactive data. Step 2: Based on high-quality interactive data, establish geometric models, thermal-humidity coupling and transfer models, gas diffusion models, and behavioral models respectively. Map all models into a multi-dimensional digital twin model of the grain warehouse through a co-evolution algorithm and output the full assessment status of the grain warehouse. Step 3: Use Fourier neural operators to perform rolling prediction of the overall state of the grain warehouse to obtain the state field of the grain pile and the state field of the equipment at the next moment, and then use the causal inference algorithm to locate the root cause of the abnormality of the grain warehouse. Step 4: Based on the fault location and root cause results, the controller obtains the optimal maintenance strategy for the current grain warehouse by modifying the Waterdrop Optimization Algorithm.

2. The digital twin interaction method for grain warehouse operation and maintenance according to claim 1, characterized in that, The specific steps for establishing a three-dimensional geometric model and a heat and humidity coupling transfer model based on high-quality interactive data are as follows: Based on BIM or 3D laser scanning technology, the data information of the grain silo and its equipment is assigned a globally unique identifier and semantically linked with the static information in the database to obtain a 3D model of the grain silo. The surface image of the grain pile is processed by a multi-view stereo matching algorithm to generate a high-density grain pile surface point cloud. Based on the flow data of grain pile entering and leaving the silo, the height value of the grain pile voxels is updated in real time by a formula. The high-density grain pile surface point cloud and the real-time height value are embedded into the 3D model of the grain silo to obtain a 3D geometric model. The preset respiration heat coefficient of the database is extracted, and the respiration heat of the grain itself is obtained based on the grain bulk density and respiration heat coefficient. The maximum specific growth rate under the preset optimal conditions in the database is extracted, and the heat production of mold growth is obtained by real-time mold specific growth rate and mold biomass concentration. The respiration heat of the grain itself and the heat production of mold growth are integrated to obtain the heat source term. The heat source term is embedded into the temperature conduction equation to obtain the heat conduction equation of the grain pile temperature field. The data points of grain moisture content are fused by Kriging interpolation to generate the moisture content distribution field of the grain pile. The moisture content distribution field is embedded into the moisture diffusion formula to obtain the effective moisture diffusion coefficient. Then, the effective moisture diffusion coefficient is embedded into the heat conduction equation of the grain pile temperature field to obtain the heat and moisture coupling transfer model.

3. The digital twin interaction method for grain warehouse operation and maintenance according to claim 2, characterized in that, The specific steps for establishing a gas diffusion convection model and behavior model are as follows: The three-dimensional spatial range of the grain pile is obtained based on the three-dimensional geometric model. The three-dimensional spatial range is divided into uniform three-dimensional grids, each of which is a cube. The porosity of the three-dimensional grids is obtained. The porosities of all three-dimensional grids are assembled into a three-dimensional array to obtain a non-uniform porosity field. The gas diffusion coefficient of the grain pile is obtained through porosity. A gas diffusion convection model is constructed based on porosity and gas diffusion coefficient. The corresponding pest density is obtained by acquiring the density of each three-dimensional grid, and the average pest density of the grain pile is obtained by averaging the density of each pest. A grain pile behavior model is established based on the average pest density and the real-time mold growth rate.

4. The digital twin interaction method for grain warehouse operation and maintenance according to claim 3, characterized in that, All models are mapped to a multidimensional digital twin model of the grain warehouse using a co-evolutionary algorithm, outputting the overall assessment status of the grain warehouse, specifically: Based on a three-dimensional geometric model, several three-dimensional meshes are divided. Through a co-evolution algorithm, all models are unified into a state tensor in several three-dimensional meshes. The dimensions of the state tensor include the spatial coordinates, time dimension, and channel dimension corresponding to the three-dimensional mesh. The channel dimension is divided into several sub-channels, and the state information of different models is embedded into the corresponding sub-channels to obtain the equipment attribute channel, grain pile physical field channel, and behavior state channel. The state tensor is mapped to the digital twin space to obtain a multi-dimensional digital twin model of the grain warehouse. The full evaluation state of the grain warehouse is output through the multi-dimensional digital twin model of the grain warehouse.

5. A digital twin interaction method for grain warehouse operation and maintenance according to claim 1, characterized in that, The grain silo's overall state is predicted using a Fourier neural operator model to obtain the grain pile state field and equipment state field at the next moment. The calculation logic is as follows: The full assessment state of the grain warehouse is input into the Fourier neural operator model, and the full assessment state of the grain warehouse is mapped to a high-dimensional feature space to obtain a three-dimensional spatial dimension. The three-dimensional spatial dimension is then transformed by a Fourier convolutional layer to obtain high-frequency modal features. The high-frequency modal features are then decoded by a decoding layer to output the prediction results of the grain pile state field and equipment state field at the next moment. The difference between the data in the grain pile state field and the preset grain pile data in the corresponding database is used to obtain the grain pile data error. If the absolute value of the grain pile data error is greater than the preset grain pile error threshold, the corresponding three-dimensional grid coordinates and grain pile abnormal parameters are marked. The difference between the data in the equipment state field and the preset equipment data in the corresponding database is used to obtain the equipment data error. If the absolute value of the equipment data error is greater than the preset equipment error threshold, the corresponding equipment is marked as an abnormal candidate, and the equipment number and corresponding physical coordinates are recorded.

6. A digital twin interaction method for grain warehouse operation and maintenance according to claim 5, characterized in that, The root cause of anomalies is located using a causal inference algorithm, and its calculation logic is as follows: By using equipment operating parameters as intervention variables, abnormal grain pile parameters as outcome variables, and external temperature and humidity, mold, and pests as confounding variables, a causal graph is generated. Extract variable data associated with abnormal regions and construct a structured dataset. Calculate the causal effect of each intervention variable on the abnormal outcome, sort them from largest to smallest causal effect, set the first variable as the root cause of the abnormality, obtain the fault location corresponding to the root cause of the abnormality based on three-dimensional grid coordinates, output the fault location and the root cause of the abnormality, and send them to the controller.

7. A digital twin interaction method for grain warehouse operation and maintenance according to claim 1, characterized in that, By deploying a sensor network within physical grain silos, environmental parameters, grain condition parameters, and equipment operating parameters are collected. The specific steps are as follows: Environmental data is collected in real time by environmental sensors, including temperature, humidity, and gas concentration data at multiple points within the silo. Grain condition data, including internal temperature, humidity, biomass concentration, and moisture content, is collected in real time by temperature and humidity sensors, fluorescence biosensors, and moisture content sensors. Images of the grain pile surface are captured in real time by high-definition industrial cameras. Strain sensors and demodulators monitor strain changes at various points within the silo in real time, obtaining the lateral pressure distribution and structural deformation of the silo, thus obtaining silo structural data. Data information from various silo equipment, including fans, refrigeration units, level gauges, and conveyors, is read by the equipment controller. This data includes rotational speed, operating power, voltage, vibration frequency, and measurement accuracy deviation. All data are integrated into comprehensive silo data.

8. A digital twin interaction method for grain warehouse operation and maintenance according to claim 7, characterized in that, A sparse completion algorithm based on the physical map of grain piles completes the data missing caused by sparse or offline sensors in the edge area of ​​grain silos, obtaining high-quality interactive data. The specific steps are as follows: The grain pile is evenly divided into grids of equal size and set as graph nodes of the grain pile physical graph. Then, physical sensors are set as observation nodes. The observation nodes include spatial coordinates and real-time readings. Virtual nodes are created at any location inside the grain pile where there are no sensors. The weights of the physical association edges between nodes are obtained. The grain pile physical graph is constructed using graph nodes and physical association edge weights. By using Chebyshev graph convolution, the observed node data in the physical graph of the grain pile is propagated to the virtual node data, resulting in virtual node data. This virtual node data is then embedded into the comprehensive data of the grain warehouse to obtain high-quality interactive data.

9. A digital twin interaction method for grain warehouse operation and maintenance according to claim 1, characterized in that, Based on the fault location and root cause analysis, the controller obtains the optimal maintenance strategy for the current grain silo by modifying the droplet optimization algorithm. The specific steps are as follows: Based on the root causes of failures and grain warehouse operation and maintenance specifications, an executable set of maintenance strategy solutions is constructed. The execution objects, resource consumption, effects, and time consumption of each maintenance strategy are clearly marked. By modifying the droplet optimization algorithm, the set of maintenance strategy solutions is transformed into a droplet population. The objective function value of each droplet is obtained, and the flow rate is updated until the preset number of iterations is reached. The objective function is calculated for each updated droplet, and the optimal droplet with the smallest current objective function is selected. Its corresponding strategy solution is recorded and executed.

Images