An AI-driven virtual-real fusion operation and maintenance system for grain storage scenes

Abstract

The application discloses a kind of AI driven type virtual-real fusion operation and maintenance systems for grain storage scene, and relates to grain storage technical field.The application includes cloud AI, control hub, monitoring console, pole winding device, sensing terminal and flat warehouse robot, several winding devices are installed on the side of pole, several poles are fixed in the grain depot and close to the side of wall, data line is wound on the side of winding device, several sensing terminals are electrically connected by data line, the port of cloud AI is connected with the port of control hub through internet to establish two-way communication, the output of sensing terminal is electrically connected with the input of control hub, the application can convert the state information obtained by sensing terminal into intuitive three-dimensional model by establishing digital twin model, and the real-time state of grain pile can be checked by operation and maintenance personnel through monitoring console, and operation and maintenance plan can be made by cloud AI, so as to improve the decision efficiency of grain maintenance and the timeliness of abnormal early warning.

Description

Technical Field

[0001] This invention relates to the field of grain storage technology, specifically to an AI-driven virtual-physical fusion operation and maintenance system for grain storage scenarios. Background Technology

[0002] Grain storage refers to storing grain in warehouses or other grain storage facilities to ensure the quantity and quality of grain and meet future needs. Existing grain storage facilities mainly include flat warehouses, shallow round warehouses, vertical silos and multi-story warehouses. Daily operation and maintenance for flat warehouse scenarios mainly rely on traditional sensor data collection and manual monitoring. Traditional sensors are mainly deployed on the walls or ceilings of grain warehouses.

[0003] The data obtained by the above deployment methods can only reflect the actual situation outside the grain. The data dimensions are too one-sided and cannot fully and deeply understand the actual state inside the grain. Furthermore, manual monitoring or simple data monitoring cannot intuitively extract features from the data and cannot meet the increasingly detailed warehouse operation and maintenance management. More detailed deployment is needed to broaden the scope of data collection. Compared with traditional data analysis methods, the data analysis advantages of cloud AI can realize automatic deployment of operation and maintenance plans or proactive detection of anomalies. Therefore, developing an AI-driven virtual-physical integrated operation and maintenance system for grain storage scenarios is a technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an AI-driven virtual-physical fusion operation and maintenance system for grain storage scenarios, which solves the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: an AI-driven virtual-real fusion operation and maintenance system for grain storage scenarios, including cloud AI, control center, monitoring console, poles, winding devices, sensing terminals, and leveling robots. Several winding devices are installed around the poles, and several poles are fixed inside the grain storage and close to the wall. Data cables are wound around the winding devices, and several sensing terminals are electrically connected through data cables.

[0006] The cloud AI's port establishes bidirectional communication with the control center's port via the Internet. The output of the sensing terminal is electrically connected to the input of the control center. The output of the control center is electrically connected to the inputs of the monitoring station and the winding device, respectively. The flattening robot's port establishes bidirectional communication with the control center's port via a wireless local area network.

[0007] The control center establishes a perception model, which acquires the state information of the grain pile and uploads it to the cloud AI. This state information includes temperature and humidity data, pressure data, vibration data, and spatial data. The control center then builds a digital twin model and transmits it to the monitoring station for display. The cloud AI analyzes the state information to obtain a comprehensive anomaly score. The Maintenance Priority Index (MPI) is used to develop maintenance plans and establish abnormal events. These plans and events are then sent to the control center and displayed on the monitoring console for maintenance personnel to view.

[0008] Furthermore, the control center builds a perception model, including a leveling robot completing the grain surface leveling operation when grain is transferred in, the control center controlling the leveling robot to deploy the perception layer, and the control center controlling the winding device to store the perception layer when grain is transferred out.

[0009] The perception model consists of several perception layers stacked vertically. When enough grain is transferred into the grain silo, the leveling robot performs another grain surface leveling operation. When the height of the leveled grain is again equal to the distance between the upper and lower perception layers, the control center controls the leveling robot to deploy the perception layers again. The perception layers are repeatedly deployed from bottom to top to form the perception model. Each perception layer consists of several perception terminals arranged in a grid. Each perception terminal integrates a temperature and humidity sensor, a pressure sensor, a vibration sensor, and a wireless module. The distance between the upper and lower perception layers, the adjacent distance between the uprights, and the adjacent distance between perception terminals on a single data line are all equal. In the perception model, the equal distance between adjacent uprights ensures that the spacing of the perception terminals in the x-axis direction is equal, and the equal spacing of the perception terminals on a single data line ensures that... Equal spacing between sensing terminals along the y-axis and equal spacing between upper and lower sensing layers ensure equal spacing between sensing terminals along the z-axis. The spacing between sensing terminals ranges from 1 meter to 3 meters. Spacing less than 1 meter or greater than 3 meters between sensing terminals increases the calculation error of the wireless module. Spacing less than 1 meter reduces the variation in signal connection strength, while spacing greater than 3 meters decreases the signal stability of the wireless module connection, potentially leading to disconnection or interference. Furthermore, excessively small spacing between sensing layers in the grain pile results in insufficient grain height, meaning the grain thickness above a single sensing layer is too thin. The driving mechanism of the leveling robot is relatively large, and when moving on the grain surface, the driving mechanism may penetrate the grain and damage the sensing layer underneath, rendering the covering grain ineffective in isolating the leveling robot from the sensing layer.

[0010] Furthermore, the control center controls the deployment of the sensing layer on the flattening robot, including a tow hook on one side of the tail of the flattening robot, and a hanging ring matching the tow hook at the end of the data cable. The control center sends instructions to the flattening robot via a wireless local area network to execute the unfolding of the reel-up device on the pole.

[0011] The control center sends instructions again to the flattening robot via wireless local area network to repeat the unfolding of the winding device, and unfolds all the winding devices on all the poles in sequence to complete the deployment of the perception layer.

[0012] Furthermore, the control center for grain transfer controls the winding device to house the sensing layer. This sensing layer transmits pressure data to the control center. Specifically, the sensing terminal acquires pressure data through an internal pressure sensor, and then transmits the pressure data to the control center via a data line. The control center then uses a formula based on the pressure data... Calculate the grain height H above the area where each sensing terminal is located, where P is the vertical pressure value obtained by the pressure sensor built into the sensing terminal. The bulk density of the grain is expressed in kilograms per cubic meter, and the value varies for different grain types. g is the gravitational acceleration, which is generally taken as 9.8 N / kg. K is the grain pressure correction coefficient, which is dimensionless. Since grain is non-fluid and has internal friction, the measured pressure is not exactly equal to the theoretical hydrostatic pressure. Therefore, the correction coefficient K is needed to make appropriate corrections. When the grain height is less than 40 cm or less than 30% of the distance between the upper and lower sensing layers, the minimum value is taken. This means that the driving force of the winding device can overcome the resistance caused by the surrounding grain squeezing the sensing terminal. The control center executes the winding process and controls the corresponding winding device to complete the winding process.

[0013] During the winding process, the control center sends a command to the winding device. The winding device reverses the winding of the data cable. The data cable synchronously drives the corresponding sensing terminal to move and be stored in the winding device. The winding device completes the winding process and the winding process stops.

[0014] Once all the winding devices at the same horizontal level have completed the winding process and the sensing layer has been stored, the leveling robot will unfold the grain again when it is loaded next time.

[0015] Furthermore, the spatial data is obtained in the following ways:

[0016] The control center executes the signal conversion process. It sends spatial measurement commands to each sensing terminal via data lines. Each sensing terminal sends a wireless connection request via its wireless module. The sensing terminal feeds back its own identification number and the signal strength of the established connection to the external terminal to the control center via data lines. The sensing terminal's identification number is used for identification; each terminal's number is unique. The control center then uses a formula... Calculate the distance d between each sensing terminal, where G is the signal strength received by the sensing terminal in dBm, and Q is the reference distance. The signal connection strength at the location is the factory parameter calibration, and n is the path loss exponent, the specific value of which is determined according to the environment where the wireless module is located.

[0017] The control center repeatedly executes the signal conversion process until all signals connected to the sensing terminals have completed distance calculations. A spatial coordinate system is established, with the unit being meters. The origin is taken as the first sensing terminal unfolded by the first winding device in the bottom sensing layer. The coordinates of each sensing terminal are calculated sequentially. The control center merges the coordinates of all sensing terminals into a dataset and marks the dataset as spatial data.

[0018] Furthermore, the control center establishes a digital twin model, including establishing a blank model based on WebGL, inputting spatial data into the blank model and aligning and expanding it to obtain a first model, marking each sensing terminal as a monitoring point, binding temperature and humidity data with the corresponding monitoring points to generate a 3D heat map and marking it as a second model, binding pressure data with the corresponding monitoring points to obtain a pressure distribution map and marking it as a third model, establishing a fourth model based on vibration data, using Fourier transform to extract the spectrum from the vibration data, setting a vibration threshold in the spectrum, which needs to be manually set by maintenance personnel in the monitoring console, when the value in the vibration data is greater than the vibration threshold, the corresponding monitoring point in the fourth model displays a point diffusion waveform, and the vibration energy distribution inside the grain pile is intuitively displayed by displaying the point diffusion waveform;

[0019] The control center superimposes and merges the first model, the second model, the third model, and the fourth model to obtain a digital twin model.

[0020] Furthermore, the cloud-based AI establishes abnormal events, including the cloud-based AI calculating a comprehensive abnormality score for each sensing terminal based on state information. The details are as follows:

[0021] The cloud-based AI records the status information uploaded by the control center and creates curves for each sensing terminal for each data point, namely, a temperature and humidity curve, a pressure curve, a vibration curve, and a spatial curve. The temperature and humidity curve reflects the temperature and humidity changes of the corresponding sensing terminal, the pressure curve reflects the pressure changes, the vibration curve reflects the vibration amplitude changes, and the spatial curve reflects the displacement amplitude changes. A historical moving average baseline value is calculated for each curve. Each graph represents time-series data, and the historical standard deviation of each data point is calculated. This is used to measure the reasonableness of fluctuations. A weight coefficient E is set for each data item in the status information. The weight coefficient E can dynamically adjust the priority of different types of data. Specific values ​​are set based on the actual storage conditions in the grain warehouse, according to the formula... Calculate the overall anomaly score for each sensing terminal. ,in Let j be the weight value of the j-th data category. The current measured value of this data, combined with the anomaly score. The degree to which the current value deviates from the normal range can be quantified. The cloud-based AI sets an anomaly threshold for each data point. After training on the data based on the state information, the cloud-based AI intelligently sets the specific value of the anomaly threshold for each data point. When the overall anomaly score is calculated... When the value exceeds the abnormal threshold, the cloud AI creates an abnormal event and sends the sensing terminal number that triggered the abnormal threshold and the abnormal event to the control center, which then forwards it to the monitoring console for display in the digital twin model.

[0022] Furthermore, the cloud-based AI formulates a maintenance plan, including recording the running time of each sensing terminal. Cumulative time without abnormal events Cloud-based AI sets the probability of abnormal events being triggered by sensing terminals within the next 7 days. According to the formula The Maintenance Priority Index (MPI) for each sensing terminal is calculated. Cloud-based AI then ranks all sensing terminals according to their MPI values ​​from highest to lowest to generate a maintenance plan. Higher MPI values ​​indicate higher maintenance priority. It is obtained by cloud AI through data training. The dataset used for data training consists of the state information of all sensing terminals. The cloud AI is the open-source deepseek-r1 general large language model.

[0023] Furthermore, after the winding device is fully unfolded, the leveling robot marks the unfolded grain surface as a restricted area. At this time, the data cable is exposed on the grain surface. When the driving mechanism of the leveling robot passes over the data cable, it will damage the deployment of the data cable and the sensing terminal. When grain is transferred in again, the leveling robot performs a leveling operation on the grain surface from the inside out. The leveled area is marked as a normal area instead of a restricted area. Grain is generally transported into the grain warehouse by a grain conveyor. After the transport is completed, the grain will form a conical grain pile. At this time, the leveling robot is controlled to perform a leveling operation from the inside out starting from the top of the grain pile. The leveling operation of the leveling robot is existing technology and will not be described in detail here. During the process of the leveling robot gradually flattening the conical grain pile, the flattened grain has a certain thickness, which can isolate the leveling robot from the sensing layer. Therefore, the marked restricted area can be changed back to a normal area.

[0024] Furthermore, during the deployment of the reeling device, the flattening robot moves to the area where the pole is located, inserts its tail hook into the hanging ring at the end of the data cable, and moves towards the wall away from the pole. The hook pulls the hanging ring, causing the data cable to move, and the data cable drives the reeling device to rotate clockwise, sequentially pulling out the sensing terminals. During its movement, the flattening robot determines its position within the grain warehouse using wireless local area network positioning. The flattening robot's wireless positioning is also based on UWB technology. Positioning base stations are set up on the inner wall of the grain warehouse. The flattening robot obtains its position information by analyzing the wireless signal connection strength between itself and each positioning base station. Since UWB positioning technology has long been used in flattening robots, the specific positioning implementation principle will not be elaborated here. When the flattening robot determines that it has reached the wall away from the pole based on its own positioning, it completes the deployment of the reeling device on a single pole.

[0025] The present invention has the following beneficial effects:

[0026] 1. By deploying a sensing layer using a leveling robot, comprehensive and detailed status information can be accurately obtained from inside the grain pile. Compared with existing data collection methods, this invention broadens the types and scope of data collection, making grain status monitoring more comprehensive. At the same time, this invention utilizes a leveling robot in a non-operational state to lay sensing terminals, automatically judging the storage conditions of the winding device based on pressure data, realizing the automatic deployment and storage of sensing terminals without increasing construction costs, reducing manual input, and improving deployment efficiency.

[0027] 2. By establishing a digital twin model, the status information obtained by the sensing terminal can be transformed into an intuitive three-dimensional model. Maintenance personnel can view the real-time status of the grain pile through the monitoring console. Compared with traditional data analysis and charts, the data feature expression of this invention is simpler and more intuitive. By formulating maintenance plans through cloud AI, the decision-making efficiency of grain maintenance and the timeliness of abnormal warnings can be improved.

[0028] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0029] 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. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a block diagram of an AI-driven virtual-real fusion operation and maintenance system for grain storage scenarios according to the present invention.

[0031] Figure 2 This is a schematic diagram showing the deployment of the pole, winding device, and sensing terminal of the present invention;

[0032] Figure 3 For the present invention Figure 2 A magnified view of a portion of point A in the middle.

[0033] The attached diagram lists the components represented by each number as follows:

[0034] In the diagram: 1-pole, 2-rewinding device, 3-data cable, 4-sensing terminal. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only 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 within the scope of protection of the present invention.

[0036] Please see Figure 1-3 The present invention provides a technical solution: an AI-driven virtual-real fusion operation and maintenance system for grain storage scenarios, including cloud AI, control center, monitoring console, pole 1, winding device 2, sensing terminal 4 and leveling robot. Several winding devices 2 are installed around the pole 1, several poles 1 are fixed in the grain storage and close to the wall. Data cables 3 are wound around the winding device 2, and several sensing terminals 4 are electrically connected through the data cables 3.

[0037] like Figure 1 As shown, the cloud AI port establishes bidirectional communication with the control center port via the Internet, the output of the sensing terminal 4 is electrically connected to the input of the control center, the output of the control center is electrically connected to the input of the monitoring station and the winding device 2 respectively, and the flattening robot port establishes bidirectional communication with the control center port via a wireless local area network.

[0038] The control center builds a perception model to acquire the state information of the grain pile and uploads it to the cloud AI. This state information includes temperature and humidity data, pressure data, vibration data, and spatial data. The control center then creates a digital twin model and transmits this data to the monitoring station for display. Temperature and humidity data can monitor whether heat or mold is occurring inside the grain; pressure data can identify the compactness of the grain distribution, and combined with spatial data, it can predict areas within the grain pile at risk of collapse; vibration data can identify pest activity within the grain silo, enabling early warning and preventative protection. The cloud AI analyzes the state information to obtain a comprehensive anomaly score. The Maintenance Priority Index (MPI) is used to develop maintenance plans and establish abnormal events. These plans and events are then sent to the control center and displayed on the monitoring console for maintenance personnel to view.

[0039] Among them, the control center builds a perception model, including the leveling robot completing the grain surface leveling operation when grain is transferred in, the control center controlling the leveling robot to deploy the perception layer, and the control center controlling the winding device 2 to store the perception layer when grain is transferred out.

[0040] The perception model consists of several perception layers stacked one on top of the other. When enough grain is transferred into the grain silo, the leveling robot performs another grain surface leveling operation. When the height of the leveled grain is again equal to the distance between the upper and lower perception layers, the control center controls the leveling robot to deploy the perception layer again. The perception layer is repeatedly deployed from bottom to top to form the perception model. The perception model can obtain the status information of the detection points inside the grain pile in real time. Each perception terminal 4 serves as a detection point, which can comprehensively and timely obtain the internal situation of the grain pile, facilitating cloud-based AI data analysis, maintenance planning, and abnormal event detection in grain storage. The perception layer consists of several perception terminals 4 arranged in a grid. The perception terminal 4 integrates temperature and humidity sensors, pressure sensors, vibration sensors, and wireless modules. The distance between the upper and lower perception layers, the adjacent distance between the uprights 1, and the adjacent distance between perception terminals 4 on a single data line 3 are all equal. In the perception model, the equal distance between adjacent uprights 1 ensures that the distance between perception terminals 4 is equal in the x-axis direction, the equal distance between perception terminals 4 on a single data line 3 ensures that the distance between perception terminals 4 is equal in the y-axis direction, and the equal distance between the upper and lower perception layers ensures that the distance between perception terminals 4 is equal in the y-axis direction. The four sensing terminals are evenly spaced along the z-axis. This even spacing helps reduce the calculation error of UWB technology and improves the accuracy of spatial data calculation. The spacing between the four sensing terminals is 1-3 meters. Spacing less than 1 meter or more than 3 meters between the four sensing terminals will increase the calculation error of the wireless module. If the spacing is less than 1 meter, the variation in signal connection strength will be smaller. For example, the variation in signal connection strength between a spacing of 90 cm and 85 cm may be the same within a spacing range of less than 1 meter, making the spatial data meaningless. If the spacing is greater than 3 meters, the signal stability of the wireless module connection will decrease, and there may be a risk of disconnection or interference. Therefore, the spatial data is also meaningless. At the same time, if the spacing between the sensing layers in the grain pile is too small, the grain height will be too low, that is, the thickness of the grain above a single sensing layer will be too thin. The driving mechanism of the leveling robot is large. When moving on the grain surface, the driving mechanism may pass through the grain and damage the sensing layer covering the grain. The covering grain cannot isolate the leveling robot from the sensing layer. Therefore, it is necessary to set a minimum spacing between the sensing layers to avoid the leveling robot interfering with the sensing layer already deployed below the grain.

[0041] like Figure 2-3As shown, to facilitate a clear and intuitive understanding of the positional relationship between the sensing terminal 4, the data cable 3, and the winding device 2, in Figure 3 The three-dimensional volume of the sensing terminal 4 is magnified. The actual volume of the sensing terminal 4 is only 3 cubic centimeters. The cross-sectional area of ​​the sensing terminal 4 is only slightly larger than that of the data cable 3. The sensing terminal 4 is small enough to be successfully rolled up and stored by the winding device 2. The temperature and humidity sensor, pressure sensor, vibration sensor and wireless module are all the smallest existing IC components integrated inside the sensing terminal 4. The data cable 3 provides power and data transmission to the sensing terminal 4. The wireless module in the sensing terminal 4 is only used to measure spatial data, that is, to obtain the signal connection strength between each wireless module, and then transmits it to the control center through the data cable 3. The control center calculates the spatial data between the sensing terminals 4 based on the signal connection strength.

[0042] Among them, the control center controls the deployment of the perception layer of the flattening robot, including a tow hook on one side of the tail of the flattening robot, and a hanging ring matching the tow hook at the end of the data cable 3. The control center sends instructions to the flattening robot through the wireless local area network to execute the unfolding of the winding device 2 on the pole 1.

[0043] The control center sends instructions again to the flattening robot via wireless local area network to repeat the unfolding of the winding device 2, and unfolds all the winding devices 2 on all the uprights 1 in sequence, completing the deployment of the perception layer.

[0044] The grain transfer control center controls the winding device 2, which houses the sensing layer. This sensing layer transmits pressure data to the control center. Specifically, the sensing terminal 4 acquires pressure data through an internal pressure sensor, and then transmits the pressure data to the control center via data line 3. The control center then uses the pressure data to perform a calculation using a formula. Calculate the grain height H above the area where each sensing terminal 4 is located, where P is the vertical pressure value obtained by the pressure sensor built into the sensing terminal 4. The bulk density of the grain is expressed in kilograms per cubic meter (kg / m³). The value varies depending on the grain type; for example, wheat ranges from 750-800 N / m³, and corn ranges from 700-750 N / m³. g represents gravitational acceleration, typically 9.8 N / kg. K is a dimensionless grain pressure correction coefficient. Because grain is non-fluid and exhibits internal friction, the measured pressure may not perfectly equal the theoretical hydrostatic pressure. Therefore, a correction coefficient K is needed. The system's preset value range is 0.85-1. When the grain height is less than 40 cm or less than 30% of the distance between the upper and lower sensing layers, the minimum value is used. This indicates that the driving force of the winding device 2 during rotation can overcome the resistance caused by the surrounding grain pressing on the sensing terminal 4. The control center executes the winding process, controlling the corresponding winding device 2 to complete the winding process. The following is a calculation example but does not represent the optimal result. With g=9.8N / kg, K=0.9, the height of the sensing layer spacing is 200cm, and the threshold for triggering the winding process is H<40cm, the above example data is substituted into the formula to calculate the height H≈37.79cm, which is less than the threshold of 40cm. The control center judges that the grain resistance is within the safe range and sends an instruction to the winding device 2 to start the winding process.

[0045] When the winding process is executed, the control center sends a command to the winding device 2. The winding device 2 reverses the winding of the data line 3. The data line 3 synchronously drives the corresponding sensing terminal 4 to move and be stored in the winding device 2. The winding device 2 completes the winding process and the winding process stops.

[0046] Once all the winding devices 2 at the same horizontal level have completed the winding process and the sensing layer has been stored, the leveling robot will unfold them again when the grain is loaded next time.

[0047] Specifically, the spatial data is obtained through the following methods:

[0048] The control center executes the signal conversion process. It sends spatial measurement commands to each sensing terminal 4 via data line 3. Each sensing terminal 4 sends a wireless connection request via its wireless module. The sensing terminal 4 feeds back its own identification number and the signal strength of the established connection to the external terminal via data line 3. The identification number of each sensing terminal 4 is unique and used for identification. The control center then uses a formula... Calculate the distance d between each sensing terminal 4, where G is the signal strength received by the sensing terminal 4 in dBm, and Q is the reference distance. The signal strength at this location is based on factory specifications and is referenced at a certain distance. The distance is 1 meter, and n is the path loss exponent, the specific value of which is determined based on the environment in which the wireless module is located. The grain silo is an unobstructed, open space, so n is set to 4. The following is a calculation example but does not represent the optimal result: Q = -40dBm, n = 4.0. G = -80dBm, substituting into the formula, we get d = 10m;

[0049] The control center repeatedly executes the signal conversion process until all signals connected to the sensing terminals 4 have completed distance calculation. A spatial coordinate system is established, with the unit being meters. The sensing terminal 4, which is unfolded by the first winding device 2 in the bottommost sensing layer, is taken as the origin, with the coordinates of the origin being (0, 0). The coordinates of each sensing terminal 4 are calculated sequentially. The control center merges the coordinates of all sensing terminals 4 into a dataset and marks the dataset as spatial data.

[0050] The control center establishes a digital twin model, including a blank model based on WebGL, into which spatial data is input, aligned, and expanded to obtain the first model. Each sensing terminal is marked as a monitoring point. Temperature and humidity data are bound to the corresponding monitoring points to generate a 3D heat map, which is marked as the second model. Pressure data is bound to the corresponding monitoring points to obtain a pressure distribution map, which is marked as the third model. A fourth model based on vibration data is established. Fourier transform is used to extract the spectrum from the vibration data. A vibration threshold is set in the spectrum. The vibration threshold needs to be manually set by maintenance personnel in the monitoring console. When the value in the vibration data is greater than the vibration threshold, the corresponding monitoring point in the fourth model displays a point diffusion waveform. The distribution of vibration energy inside the grain pile is displayed intuitively by displaying the point diffusion waveform.

[0051] The control center superimposes and merges the first model, the second model, the third model, and the fourth model to obtain a digital twin model.

[0052] Among these, cloud-based AI establishes abnormal events, including cloud-based AI calculating a comprehensive abnormality score for each sensing terminal 4 based on state information. The details are as follows:

[0053] The cloud-based AI records the status information uploaded by the control center and creates curves for each sensing terminal 4 for each data point, namely, a temperature and humidity curve, a pressure curve, a vibration curve, and a spatial curve. The temperature and humidity curve reflects the temperature and humidity changes of the corresponding sensing terminal 4, the pressure curve reflects the pressure changes of the corresponding sensing terminal 4, the vibration curve reflects the vibration amplitude changes of the corresponding sensing terminal 4, and the spatial curve reflects the displacement amplitude changes of the corresponding sensing terminal 4. The historical moving average baseline value of each curve is calculated. Each graph represents time-series data, and the historical standard deviation of each data point is calculated. This is used to measure the reasonableness of fluctuations. A weight coefficient E is set for each data item in the status information. The weight coefficient E can dynamically adjust the priority of different types of data. Specific values ​​are set based on the actual storage conditions in the grain warehouse, according to the formula... Calculate the overall anomaly score for each sensing terminal 4. ,in Let J be the weight value of the j-th data category, and J be the total number of data categories. The current measured value of this data, combined with the anomaly score. The degree to which the current value deviates from the normal range can be quantified. The cloud-based AI sets an anomaly threshold for each data point. After training on the data based on the state information, the cloud-based AI intelligently sets the specific value of the anomaly threshold for each data point. When the overall anomaly score is calculated... When the value exceeds the abnormal threshold, the cloud-based AI generates an abnormal event and sends the number of the sensing terminal 4 that triggered the abnormal threshold, along with the abnormal event, to the control center. The control center then forwards this information to the monitoring console for display in the digital twin model. The following is a calculation example, but it does not represent the optimal result. The temperature of sensing terminal 4, located at coordinates (10, 25), increases by 2 degrees Celsius. The historical average temperature is known. =20 degrees Celsius, standard deviation =0.5 degrees Celsius, weighting coefficient E=0.6, comprehensive anomaly score =2.4, if the cloud AI sets the abnormal threshold to 2.0, then the cloud AI will trigger the creation of an abnormal event.

[0054] Among these measures, cloud-based AI develops maintenance plans, including recording the running time of each sensing terminal 4. Cumulative time without abnormal events The cloud-based AI sets the probability of the sensing terminal 4 triggering abnormal events within the next 7 days. According to the formula The Maintenance Priority Index (MPI) for each sensing terminal 4 is calculated. The cloud-based AI then sorts all sensing terminals 4 according to their MPI values ​​from highest to lowest to obtain a maintenance plan. A higher MPI value indicates a higher priority for maintenance. The dataset used for training is the state information of all sensing terminals 4, and the cloud AI is the open-source DeepSeek-r1 general-purpose large language model.

[0055] After the winding device 2 is fully extended, the leveling robot marks the extended grain surface as a no-entry zone. At this time, the data cable 3 is exposed on the grain surface. When the driving mechanism of the leveling robot passes through the data cable 3, it will damage the deployment of the data cable 3 and the sensing terminal 4, causing the entire line of sensing terminals 4 on the corresponding data cable 3 to malfunction. Therefore, it is necessary to mark it as a no-entry zone to prevent the leveling robot from interfering with the already deployed sensing terminals 4. When grain is transferred in again, the leveling robot will perform a leveling operation on the grain surface from the inside out when the grain pile is transferred in. The leveled area will then be removed from the no-entry zone. The restricted area is marked as a normal area. When grain is transferred in, it is generally transported into the grain warehouse by a grain conveyor. After the transport is completed, the grain will form a cone-shaped grain pile. At this time, the leveling robot is controlled to start from the top of the grain pile and perform the grain leveling operation from the inside to the outside. The leveling operation of the leveling robot is existing technology and will not be described in detail here. During the process of the leveling robot gradually flattening the cone-shaped grain pile, the flattened grain has a certain thickness, which can isolate the leveling robot from the sensing layer. Therefore, the marked restricted area can be changed back to a normal area, which is convenient for the next deployment of a higher-level sensing layer.

[0056] When the winding device 2 is deployed, the flattening robot moves to the area where the pole 1 is located, inserts its tail hook into the hanging ring at the end of the data cable 3, and moves towards the wall away from the pole 1. The hook pulls the hanging ring, causing the data cable 3 to move. The data cable 3 drives the winding device 2 to rotate forward, pulling out the sensing terminal 4 in sequence. During the movement, the flattening robot determines its position in the grain warehouse through wireless local area network positioning. The wireless positioning of the flattening robot is also based on UWB technology. Positioning base stations are set on the inner wall of the grain warehouse. The flattening robot obtains its own position information by analyzing the wireless signal connection strength between itself and each positioning base station. Since UWB positioning technology has long been used in flattening robots, the specific positioning implementation principle will not be elaborated here. When the flattening robot determines that it has reached the wall away from the pole 1 based on its own positioning, it completes the deployment of the winding device 2 on a single pole 1.

[0057] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An AI-driven virtual-real fusion operation and maintenance system for grain storage scenarios, comprising cloud AI, control center, monitoring console, uprights (1), winding devices (2), sensing terminals (4), and a leveling robot, wherein several winding devices (2) are installed around the uprights (1), several uprights (1) are fixed inside the grain storage and close to the wall, and data cables (3) are wound around the winding devices (2), and several sensing terminals (4) are electrically connected through the data cables (3); characterized in that: The cloud AI port establishes bidirectional communication with the control center port via the Internet. The output of the sensing terminal (4) is electrically connected to the input of the control center. The output of the control center is electrically connected to the input of the monitoring station and the winding device (2). The flattening robot port establishes bidirectional communication with the control center port via a wireless local area network. The control center establishes a perception model, which acquires the state information of the grain pile and uploads it to the cloud AI. This state information includes temperature and humidity data, pressure data, vibration data, and spatial data. The control center then builds a digital twin model and transmits it to the monitoring station for display. The cloud AI analyzes the state information to obtain a comprehensive anomaly score. The Maintenance Priority Index (MPI) is used to develop maintenance plans and establish abnormal events. These plans and events are then sent to the control center and displayed on the monitoring console.

2. The AI-driven virtual-physical fusion operation and maintenance system for grain storage scenarios according to claim 1, characterized in that, The control center builds a perception model, including the flattening robot completing the grain surface leveling operation when the grain is transferred in, the control center controlling the flattening robot to deploy the perception layer, and the control center controlling the winding device (2) to store the perception layer when the grain is transferred out. The sensing model is composed of several sensing layers stacked on top of each other. Each sensing layer is composed of several sensing terminals (4) arranged in a grid. Each sensing terminal (4) integrates a temperature and humidity sensor, a pressure sensor, a vibration sensor and a wireless module. The spacing between the upper and lower layers of the sensing layer, the adjacent distance between the poles (1) and the adjacent distance between the sensing terminals (4) are all equal.

3. The AI-driven virtual-physical fusion operation and maintenance system for grain storage scenarios according to claim 2, characterized in that, The control center controls the deployment of the sensing layer of the flattening robot, including a tow hook on one side of the tail of the flattening robot, and a hanging ring matching the tow hook at the end of the data line (3). The control center sends instructions to the flattening robot through the wireless local area network to execute the unfolding of the winding device (2) on the pole (1). The control center sends instructions again to the flattening robot via wireless local area network to repeat the unfolding operation of the winding device (2), and unfolds all the winding devices (2) on all the uprights (1) in sequence to complete the deployment of the perception layer.

4. The AI-driven virtual-physical fusion operation and maintenance system for grain storage scenarios according to claim 2, characterized in that, When the grain is transferred out, the control center controls the winding device (2) to receive the sensing layer, including the sensing layer transmitting pressure data to the control center, and the data line (3) transmitting the pressure data to the control center. The control center then uses a formula based on the pressure data. Calculate the grain height H above the area where each sensing terminal (4) is located, where P is the vertical pressure value acquired by the sensing terminal (4). The bulk density of the grain is g, the gravitational acceleration is K, the grain pressure correction coefficient is K, the control center executes the winding process, and controls the winding device (2) to complete the winding process. When the winding process is executed, the control center sends an instruction to the winding device (2), the winding device (2) reverses the winding of the data line (3), the data line (3) synchronously drives the sensing terminal (4) to move and be stored in the winding device (2), the winding device (2) completes the winding process, and the winding process stops; Once all the winding devices (2) at the same horizontal level have completed the winding process, the sensing layer will be stored.

5. The AI-driven virtual-physical fusion operation and maintenance system for grain storage scenarios according to claim 1, characterized in that, The spatial data is obtained in the following ways: The control center executes the signal conversion process. The control center sends spatial measurement instructions to each sensing terminal (4) via data line (3). Each sensing terminal (4) sends a wireless connection request to the outside world via its wireless module. The sensing terminal (4) feeds back its own number and the signal connection strength for establishing an external connection to the control center via data line (3). The number of the sensing terminal (4) is used for identification; each sensing terminal (4)'s number is unique. The control center determines the connection according to the formula... Calculate the distance d between each sensing terminal (4), where G is the signal connection strength received by the sensing terminal (4), and Q is the reference distance. The signal connection strength at the location, where n is the path loss exponent; The control center repeatedly executes the signal conversion process until all the signals connected to the sensing terminals (4) have completed the distance calculation. A spatial coordinate system is established, with the sensing terminal (4) unfolded by the first winding device (2) in the bottom sensing layer as the origin. The coordinates of each sensing terminal (4) are calculated in turn. The control center merges the coordinates of all sensing terminals (4) into a dataset and marks the dataset as spatial data.

6. The AI-driven virtual-physical fusion operation and maintenance system for grain storage scenarios according to claim 1, characterized in that, The control center establishes a digital twin model, including establishing a blank model of the control center, inputting spatial data into the blank model and aligning and unfolding it to obtain a first model, marking each sensing terminal (4) as a monitoring point, binding temperature and humidity data with the corresponding monitoring point to generate a 3D heat map and marking it as a second model, binding pressure data with the corresponding monitoring point to obtain a pressure distribution map and marking it as a third model, establishing a fourth model based on vibration data, extracting the spectrum map from the vibration data, setting a vibration threshold in the spectrum map, and when the value in the vibration data is greater than the vibration threshold, the corresponding monitoring point in the fourth model displays a point diffusion waveform; The control center superimposes and merges the first model, the second model, the third model, and the fourth model to obtain a digital twin model.

7. The AI-driven virtual-physical fusion operation and maintenance system for grain storage scenarios according to claim 1, characterized in that, The cloud-based AI establishes abnormal events, including the cloud-based AI calculating a comprehensive abnormal score for each sensing terminal (4) based on state information. The details are as follows: The cloud-based AI records the status information uploaded by the control center, and establishes a curve for each sensing terminal (4) for each data item, calculating the historical moving average benchmark value of each curve. Calculate the historical standard deviation for each data point. Set the weight coefficient E for each data item in the status information, according to the formula. Calculate the overall anomaly score for each sensing terminal (4). ,in Let j be the weight value of the j-th data category. For the current measured value of this data, the cloud-based AI sets an anomaly threshold for each data point, and then calculates the overall anomaly score. When the value is greater than the abnormal threshold, the cloud AI establishes an abnormal event and sends the number of the sensing terminal (4) that triggered the abnormal threshold and the abnormal event to the control center, which then forwards it to the monitoring station to be displayed in the digital twin model.

8. The AI-driven virtual-physical fusion operation and maintenance system for grain storage scenarios according to claim 1, characterized in that, The cloud-based AI formulates a maintenance plan, including recording the running time of each sensing terminal (4) separately. Cumulative time without abnormal events The cloud-based AI sets the probability of the sensing terminal (4) triggering abnormal events within the next 7 days. According to the formula The maintenance priority index (MPI) of each sensing terminal (4) is calculated. The cloud AI then arranges all sensing terminals (4) in descending order of their MPI values ​​to obtain a maintenance plan. The dataset used for training is the state information of all sensing terminals (4) obtained by cloud AI through data training. The cloud AI is a general large language model.

9. The AI-driven virtual-physical fusion operation and maintenance system for grain storage scenarios according to claim 3, characterized in that, After the winding device (2) is fully unfolded, the leveling robot marks the unfolded grain surface as a restricted area. When the grain is transferred in again, the leveling robot performs a grain surface leveling operation in the grain pile, and the leveled area is marked from the restricted area to the normal area.

10. The AI-driven virtual-physical fusion operation and maintenance system for grain storage scenarios according to claim 3, characterized in that, When the winding device (2) is deployed, the flattening robot moves to the area where the pole (1) is located, inserts its tail hook into the hanging ring at the end of the data cable (3), moves the flattening robot toward the wall on the other side away from the pole (1), the hook pulls the hanging ring to drive the data cable (3) to move, the data cable (3) drives the winding device (2) to rotate forward and pull out the sensing terminal (4) in sequence. When the flattening robot determines that it has reached the wall on the other side away from the pole (1) based on its own positioning, it completes the deployment of the winding device (2) on a single pole (1).

Images