Silo landing leg stress monitoring and early warning method and system

Abstract

The invention relates to the technical field of stress monitoring, in particular to a silo landing leg stress monitoring and early warning method and system, and the method comprises the steps: building a silo coordinate system, obtaining the coordinates of each silo landing leg, setting a label for each landing leg, and obtaining the density data of a material and the weight data of a silo; collecting multi-source characteristic data in real time, marking a discharging event based on a discharging valve position signal, and outputting an event time window; in the unloading event time window, collecting material image data in the silo in real time, and calculating and obtaining a grid mass field and a material gravity center position based on the material image data and density data. In the early warning link, the supporting leg stress data collected in real time and the model prediction stress are compared one by one, the source of stress deviation can be effectively recognized, when the deviation exceeds the corresponding threshold value, the system can further judge whether the abnormity belongs to a structural problem or stress unbalance loading caused by unloading according to the deviation characteristics, and the accuracy and the reliability of the system are improved. Therefore, more targeted anomaly identification is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of force monitoring, in particular to a silo leg force monitoring and early warning method and system. BACKGROUND

[0002] In the existing field of silo structure health monitoring, the leg force monitoring system usually arranges strain gauges or pressure sensors at each leg to collect load data of the supporting structure in real time to determine the force balance state of the silo operation. When the system detects that the stress of a certain leg exceeds the set threshold, an alarm or early warning is triggered to prevent safety accidents such as silo tilting and foundation damage caused by abnormal structural stress.

[0003] However, in actual operation, the unloading process of the silo has significant dynamic and nonlinear characteristics. The existing force monitoring system often fails to fully consider the stress change mechanism under unloading conditions, resulting in false monitoring results, false alarms, and even missed real early warning opportunities. For example, during the unloading stage, the opening sequence and rhythm of different unloading ports directly affect the distribution of material in the silo. When a unilateral unloading valve is opened in advance, the center of gravity of the material will shift to one side in a short time, causing the axial pressure of the leg on that side to rise sharply. The existing monitoring system fails to identify this temporary unloading-induced load imbalance and incorrectly determines it as an abnormal structural load, triggering false alarms. In addition, local blockage or bridging often occurs during unloading. When the bridge suddenly collapses, the material flow impacts the bottom plate, forming a transient stress wave. The existing system uses a fixed time window filtering algorithm, which is difficult to distinguish between short impact signals and real overload signals, thus also causing false alarms. SUMMARY

[0004] To solve the above problems, the present application provides a silo leg force monitoring and early warning method and system.

[0005] The present application adopts the following technical solution, a silo leg force monitoring and early warning method, comprising:

[0006] Establish a silo coordinate system, obtain the coordinates of each silo leg, and set a label for each leg. Obtain the density data of the material and the weight data of the silo;

[0007] Construct a multi-source feature data synchronous collection system containing unloading valve position signals, material flow velocity, and leg strain stress. Collect multi-source feature data in real time. Based on the unloading valve position signals, label the unloading event and output the event time window;

[0008] When in the unloading event time window, collect material image data in the silo in real time. Based on the material image data and the density data, calculate and obtain the grid mass field and the material center of gravity position;

[0009] Input the current material flow velocity, grid mass field, center of gravity position, silo weight data, and silo leg coordinates into the pre-built leg force prediction model, and output the predicted force data of the leg.

[0010] By collecting real-time force data on the outriggers Predicted force data for each outrigger Comparative analysis is performed to generate early warning instructions.

[0011] As a further description of the above technical solution: the method for obtaining the grid mass field and the position of the material's center of gravity includes:

[0012] Several depth camera units are arranged around the top of the silo. Each depth camera unit integrates an active light source component and a depth camera. The active light source component is used to project a preset coded grating onto the material surface. The depth camera collects the grating deformation map reflected from the material surface to obtain preliminary depth point cloud data.

[0013] Using the calibrated camera extrinsic parameters, the depth point cloud data of each depth camera is projected onto the silo coordinate system to obtain the point cloud set corresponding to each depth camera. Based on the iterative nearest point registration algorithm, the geometric consistency of the overlapping area of ​​the point cloud is corrected, and on this basis, the material surface height field data of the overall material surface geometry reconstruction is obtained.

[0014] The bottom surface of the silo is discretized into a grid. Based on the material surface height field data, the volume and unit mass of each grid are obtained to obtain the grid mass field.

[0015] The location of the material's center of gravity is obtained based on grid mass field calculations.

[0016] As a further description of the above technical solution: the method for correcting the geometric consistency of overlapping areas in the point cloud based on the iterative nearest point registration algorithm, and obtaining the overall material surface geometry reconstruction material surface height field data on this basis includes:

[0017] Step 1: Set the point cloud data obtained by any depth camera unit as the reference point cloud. , denoted as: , Indicating the first point cloud The coordinates of a 3D point are used to set the point cloud data obtained from other depth camera units as the point cloud to be registered. , denoted as: , Indicating the first point cloud The coordinates of a three-dimensional point;

[0018] Step 2, in the reference point cloud and the point cloud to be matched In the middle, the corresponding point pair set is obtained by the nearest point search, denoted as: C=( , ); wherein, C is the corresponding point pair set, ∈ , that is, the first point in the reference point cloud; ∈ , that is, the first point in the point cloud to be registered;

[0019] Step 3, based on the obtained corresponding point pair set C, an error measurement function E for measuring the difference between two points is constructed, and the total error is calculated, and the expression of the error measurement function E is: . is the total error, represents a 3x3 rotation matrix for describing spatial rotation, represents a 3x1 translation vector for describing spatial offset, represents the point after rotation R and displacement ;

[0020] Step 4, the error measurement function E is minimized to obtain a set of optimal parameters , wherein, represents the optimal rotation matrix, represents the optimal translation vector;

[0021] Step 5, the optimal rotation matrix and the optimal translation vector are applied to all points in the point cloud to be registered, so that the transformed point cloud to be registered is obtained;

[0022] Step 6, after obtaining the point cloud to be registered , the nearest point search is performed again with the reference point cloud , and a new corresponding point pair set C' is obtained, and steps 3-5 are repeated three times until the total error change of adjacent two iterations is less than a preset error threshold or a preset maximum iteration number is reached.

[0023] Step 7, the point cloud to be registered after geometric consistency correction is combined with the reference point cloud , and the complete and consistent reconstructed material surface height field data is obtained.

[0024] As a further description of the above technical solution: the training method of the leg force prediction model comprises:

[0025] Pre-collect H group training data, H is a positive integer greater than 0, the training data includes the grid mass field of the material at the current time, the center of gravity position, the weight data of the silo and the coordinates of the silo leg and the corresponding predicted force data of the leg.

[0026] The gradient boosting regression tree model is used as the leg force prediction model, the initial hyperparameters are set, and the collected training data is divided into training set, validation set and test set according to the preset proportion;

[0027] The training set is used to train the model, the mean square error is used as the loss function, a simple model is initialized, and decision trees are iteratively constructed, each new tree aims to fit the negative gradient of the loss function of the previous model on the training set, the leaf node weight of the new tree is optimized by gradient descent method, and it is integrated into the existing model with a preset learning rate; The Bayesian optimization method is used to optimize the hyperparameters to optimize the performance on the validation set;

[0028] Every 20 iterations, the performance index of the model is calculated using the validation set, and when the root mean square error on the validation set decreases by less than a preset threshold for several consecutive times, the model performance is considered to be stable, and the training is stopped;

[0029] The test set is used to evaluate the trained model, and the model performance evaluation meets the standard, and is deployed and applied.

[0030] As a further description of the above technical solution: the multi-source feature data synchronous acquisition system includes travel switches or potentiometer type position sensors, mass flow meters and three-dimensional force sensors;

[0031] The multi-source feature data includes unloading valve position signals, flow velocity and force data of each leg.

[0032] As a further description of the above technical solution: the method for generating a warning instruction comprises:

[0033] Compare the real-time force scalar of each leg with its preset threshold, and calculate the overrun ratio;

[0034] When the overrun ratio is greater than or equal to 1, it is determined to be abnormal, and a prediction force comparison analysis is performed to generate a warning instruction;

[0035] When the overrun ratio is less than 1, no warning instruction is generated.

[0036] As a further description of the above technical solution: the method for generating a warning instruction by comparing the predicted force comprises:

[0037] Based on the predicted stress data, the stress deviation is calculated, and the obtained stress deviation is compared and analyzed with the preset deviation threshold value, when the stress deviation is greater than or equal to the deviation threshold value, a structural abnormality early warning instruction is generated;

[0038] When the stress deviation is less than the deviation threshold value, an unloading partial load abnormality instruction is generated.

[0039] As a further description of the above technical solution: the method for establishing the silo coordinate system is to take the center point of the center bottom end of the silo as the origin, take two horizontal lines perpendicular to each other as the horizontal and vertical coordinates, and take the vertical center line of the silo as the vertical coordinate to establish the silo coordinate system.

[0040] As a further description of the above technical solution: the stress data of the support leg ={ 、 、 }, =1, 2, …, N, wherein N is the number of support legs, the stress data of the kth support leg at time t, 、 、 are the stress components in X, Y and Z directions respectively;

[0041] predicted stress data ={ 、 、 }, the predicted stress data of the kth support leg at time t, 、 、 are the stress components in X, Y and Z directions respectively.

[0042] The silo support leg stress monitoring and early warning system is used to realize the silo support leg stress monitoring and early warning method, and the system comprises:

[0043] A data acquisition module establishes a silo coordinate system, acquires the coordinates of each silo support leg, sets a label for each support leg, acquires the density data of the material and the weight data of the silo;

[0044] A parameter acquisition module constructs a multi-source feature data same-frequency acquisition system containing unloading valve position signals, material flow speed, and strain stress of each support leg, acquires multi-source feature data in real time, labels unloading events and outputs event time windows based on unloading valve position signals;

[0045] A data analysis module acquires material image data in the silo in real time when in the unloading event time window, calculates and acquires grid mass field and material barycenter position based on the material image data and the density data;

[0046] a stress prediction module, which inputs the grid mass field of the material at the current time, the center of gravity position, the weight data of the silo, and the coordinates of the silo legs into a pre-constructed leg stress prediction model, and outputs predicted stress data of the legs;

[0047] a warning generation module, which compares and analyzes the real-time collected leg stress data with the predicted stress data of each leg to generate a warning instruction.

[0048] Beneficial effects:

[0049] In the above technical solution, the leg stress prediction model is constructed based on the grid mass field, the material center of gravity position, the silo weight data, and the leg spatial coordinates, which can predict the leg stress under different working conditions. In the warning link, the real-time collected leg stress data is compared with the model predicted stress on a leg-by-leg basis, which can effectively identify the source of the stress deviation. When the deviation exceeds the corresponding threshold, the system can further judge whether the abnormality belongs to a structural problem or a stress deviation caused by unloading, thereby realizing more targeted abnormality identification. Compared with the traditional alarm mode based on a single threshold trigger, this method significantly reduces the possibility of false positives and false negatives, and improves the reliability of the warning result. BRIEF DESCRIPTION OF DRAWINGS

[0050] The present application will be further explained below in conjunction with the drawings and embodiments:

[0051] Figure 1 A flowchart of the silo leg stress monitoring and warning method provided for Embodiment 1 of the present application;

[0052] Figure 2 A flowchart of the method for obtaining the grid mass field and the material center of gravity position provided for Embodiment 1 of the present application;

[0053] Figure 3 A flowchart of the method for obtaining the overall material surface geometry reconstruction material surface height field data provided for Embodiment 1 of the present application;

[0054] Figure 4 A module connection diagram of the silo leg stress monitoring and warning system provided for Embodiment 2 of the present application. DETAILED DESCRIPTION

[0055] In order to make the technical means, creative features, purposes and effects realized by the present application easy to understand, the present application will be further described below in conjunction with specific drawings. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0056] Embodiment 1

[0057] Please refer to Figures 1-3 The embodiment of the application provides a technical scheme: a silo support leg stress monitoring and early warning method, comprising:

[0058] The center point of the center bottom end of the silo is taken as an origin, two horizontal lines perpendicular to each other are taken as horizontal and vertical coordinates, and the vertical center line of the silo is taken as a vertical coordinate to establish a silo coordinate system;

[0059] Based on the established silo coordinate system, the coordinates of each silo support leg are obtained, and a label is set for each support leg, and the density data of the material and the weight data of the silo are obtained;

[0060] It should be noted that the weight data of the silo includes the total weight of the material and the weight of the silo structure itself, which is collected and obtained through a preset weighing sensor, and the density data of the material can be directly obtained by obtaining the type of the stored goods in the silo.

[0061] Optionally, a digital label can be set for each support leg, and the labels are set to 1, 2, 3, 4, …, respectively, according to the number of support legs.

[0062] A multi-source feature data synchronous collection system including a discharge valve position signal, a material flow speed, and a strain stress of each support leg is constructed, multi-source feature data is collected in real time, a discharge event is labeled and an event time window is output based on the discharge valve position signal;

[0063] The multi-source feature data synchronous collection system includes a travel switch or a potentiometer type position sensor, a mass flowmeter, and a three-dimensional force sensor.

[0064] The multi-source feature data includes a discharge valve position signal, a material flow speed, and a stress data of each support leg.

[0065] By analyzing the discharge valve position signal to identify the discharge behavior and generate an event time window, the system can accurately distinguish the discharge stage from the non-discharge stage.

[0066] It should be noted that the discharge valve position signal is collected and obtained by a travel switch or a potentiometer type position sensor, which is rigidly installed at each discharge valve driving shaft or valve rod to directly reflect the valve opening angle or displacement.

[0067] The material flow speed is collected and obtained by a mass flowmeter, which is installed at the outlet of the hopper.

[0068] The strain and stress of each support leg are collected and obtained by a three-dimensional force sensor, which is arranged on each support leg.

[0069] It should be noted that through multi-source data same frequency acquisition, the data misplacement caused by asynchronous sensor timing can be effectively avoided, and the recognition accuracy of the unloading behavior is improved. The event time window is used to filter non-unloading interference factors, so that the subsequent stress prediction and material distribution calculation occur in the unloading related period, and the temporary eccentric load signal generated by unloading in the traditional system is misjudged as a structural failure.

[0070] Specifically, through multi-source data same frequency acquisition, the data misplacement caused by asynchronous sensor timing can be effectively avoided, and the recognition accuracy of the unloading behavior is improved. The event time window is used to filter non-unloading interference factors, so that the subsequent stress prediction and material distribution calculation occur in the unloading related period, and the temporary eccentric load signal generated by unloading in the traditional system is misjudged as a structural failure.

[0071] When in the unloading event time window, the material image data in the silo is collected in real time, and the grid mass field and the material barycenter position are calculated and obtained based on the material image data and the density data;

[0072] The method for obtaining the grid mass field and the material barycenter position comprises:

[0073] A plurality of depth camera units are arranged on the top of the silo along the circumference, the depth camera unit is integrated with an active light source assembly and a depth camera, the active light source assembly is used for projecting a preset coded grating to the material surface, and the grating deformation image reflected by the material surface is collected by the depth camera, so as to obtain preliminary depth point cloud data;

[0074] The depth point cloud data of each depth camera is projected into the silo coordinate system by using the calibrated camera external parameters, the point cloud set corresponding to each depth camera is obtained, the geometric consistency correction is performed on the point cloud overlapping area based on the iterative closest point registration algorithm, and the material surface height field data of the overall material surface geometric morphology reconstruction is obtained on this basis;

[0075] It should be noted that for each depth camera, the internal and external parameter calibration is completed in advance, the position and orientation information of the camera in the silo global coordinate system is obtained, the external parameter includes a rotation matrix and a translation vector, and is used to describe the rigid transformation relationship between the camera coordinate system and the silo global coordinate system;

[0076] The point cloud is converted from the camera coordinate system to the silo global coordinate system by using the calibrated external parameter, and the method comprises the following rigid transformation operation on each point cloud coordinate: first, multiply the point by the rotation matrix to adjust the direction, and then add the translation vector to complete the position alignment, that is, each point is transformed into the coordinate in the silo global coordinate system.

[0077] The bottom surface of the silo is discretized into a grid, the volume and unit mass of each grid in the grid are obtained based on the material surface height field data, and the grid mass field is obtained ;

[0078] The volume is calculated by the following formula: ; wherein, , is the row number and column number of the grid, is the volume at the time moment at the discrete grid at the row, the column, is the material surface height value at the time moment at the discrete grid at the row, the column, is the area of the discrete grid at the row, the column.

[0079] The unit mass is calculated by the following formula: ; is the grid mass field at the time moment at the discrete grid at the row, the column, is the material density data.

[0080] The material barycenter position is calculated based on the grid mass field.

[0081] The calculation formula for calculating the material barycenter position is:

[0082] ;

[0083] ;

[0084] ;

[0085] ; wherein, , is the horizontal coordinate of the grid center at the row, the column of the discrete grid, is the vertical coordinate of the grid center at the row, the column of the discrete grid, is the vertical coordinate of the grid center at the row, the column of the discrete grid; , , are the horizontal coordinate, vertical coordinate and vertical coordinate of the material barycenter position, respectively.

[0086] Specifically, depth imaging technology is used to obtain the three-dimensional shape of the material inside the silo during the unloading stage, and the bottom surface of the silo is discretized into a grid. The mass of each grid is calculated to obtain the grid mass field and the position of the material's center of gravity in real time.

[0087] By constructing a material surface height field using real 3D point clouds, the traditional single-point material position or experience-based estimation mode is no longer relied upon, making the material distribution expression more refined and accurate. The precise calculation of the center of gravity position can intuitively reflect the material offset trend during the unloading process, providing key input for predicting the force on the outriggers, and significantly improving the accuracy of the prediction model and its adaptability to uneven unloading.

[0088] Methods for correcting geometric consistency in overlapping point cloud regions based on iterative nearest-point registration algorithms, and obtaining overall material surface geometry reconstruction data based on this, include:

[0089] Step 1: Set the point cloud data obtained by any depth camera unit as the reference point cloud. , denoted as: , Indicating the first point cloud The coordinates of a 3D point are used to set the point cloud data obtained from other depth camera units as the point cloud to be registered. , denoted as: , Indicating the first point cloud The coordinates of a three-dimensional point;

[0090] Step 2, in the reference point cloud Point cloud to be matched In this context, the set of corresponding point pairs is obtained through nearest-neighbor search, denoted as: C = ( , ); where C is the set of corresponding point pairs, ∈ That is, the first point in the reference point cloud One point; ∈ That is, the first point in the point cloud to be registered One point;

[0091] It should be noted that the corresponding point pairs are determined based on the principle of closest point in three-dimensional space, that is, for point clouds. any point q in sk Searching for point clouds The point with the smallest distance from it is taken as p. s ;

[0092] Step 3: Based on the obtained set of corresponding point pairs C, construct an error metric function E to measure the difference between two points, and calculate the overall error. The expression for the error metric function E is: . is the total error, is a 3x3 rotation matrix, which describes a spatial rotation, is a 3x1 translation vector, which describes a spatial offset, is a point after rotation R and displacement .

[0093] It is noted that the meaning of the total error is that if all the corresponding points can be aligned with the reference points as much as possible after the same rotation and translation transformation, the error will be smaller.

[0094] Step 4, the error metric function E is minimized to obtain a set of optimal parameters , wherein, is the optimal rotation matrix, is the optimal translation vector;

[0095] It is noted that the step of minimizing includes:

[0096] The geometric centroid position is calculated for all corresponding point pairs (the centroid is the vector average of all point coordinates);

[0097] The point cloud is respectively translated to the centroid, and the rotation is solved with the centroid as the center;

[0098] The rotation relationship that makes the two point sets closest is obtained based on the singular value decomposition method;

[0099] The optimal translation vector is calculated by substituting the obtained rotation relationship back into the centroid coordinates.

[0100] Step 5, the optimal rotation matrix and the optimal translation vector are applied to the points in all the point clouds to be registered , so that the transformed point cloud to be registered is obtained;

[0101] Step 6, after obtaining the point cloud to be registered , the nearest point search is performed again with the reference point cloud , a new corresponding point pair set C' is obtained, and steps 3-5 are repeated until the change in the total error of the adjacent two iterations is less than a preset error threshold or a preset maximum number of iterations is reached;

[0102] Step 7, the point cloud to be registered after geometric consistency correction is combined with the reference point cloud to obtain complete and consistent reconstructed material surface height field data;

[0103] The point cloud to be registered With the reference point cloud The merged expression is: . Indicates a set union operation.

[0104] Input the flow velocity of the material at the current moment, the grid mass field, the center of gravity position, the weight data of the silo, and the coordinates of the silo leg into the pre-constructed leg force prediction model, and output the predicted force data of the leg;

[0105] The training method of the leg force prediction model comprises:

[0106] Pre-collect H sets of training data, H being a positive integer greater than 0, and the training data including the grid mass field, the center of gravity position, the weight data of the silo, and the coordinates of the silo leg at the current moment, and the corresponding predicted force data of the leg.

[0107] A gradient boosting regression tree model is used as the leg force prediction model, and initial hyperparameters are set: the initial hyperparameters include: the number of decision trees is 100-200, the maximum depth of a single tree is 5-8 (to control overfitting and avoid learning noise features), the minimum number of samples for node splitting is 10, and the maximum number of features considered during splitting is 3;

[0108] The collected training data is divided into a training set, a validation set, and a test set according to a preset ratio; optionally, the division ratio is 6:3:1.

[0109] The training set is used to train the model, the mean square error is used as the loss function, a simple model is initialized, decision trees are iteratively constructed, each new tree aims to fit the negative gradient of the loss function of the previous model on the training set, the leaf node weights of the new tree are optimized by the gradient descent method, and the new tree is integrated into the existing model with a preset learning rate; the hyperparameters are optimized by the Bayesian optimization method, and the optimization range includes: the number of trees is 20-50, the learning rate is 0.01-0.1, and the regularization coefficient is 0.05-0.2;

[0110] Every 20 trees are iterated, the performance indicators of the model are calculated once using the validation set, for example, the root mean square error, when the root mean square error on the validation set decreases by less than a preset threshold (for example, 0.01) for a plurality of consecutive times (for example, 3 consecutive times), it is considered that the model performance has tended to be stable, and the training is stopped.

[0111] The trained model is evaluated using the test set, and the model performance evaluation meets the standard (the test set precision rate is ≥90%), and the model is deployed and applied;

[0112] By comparing the real-time collected leg force data with the predicted force data of each leg performing comparative analysis to generate early warning instructions;

[0113] leg force data ={ 、 、 }, =1, 2, …, N;

[0114] wherein N is the number of legs, the force data of the kth leg at time t, 、 、 are the force components in X, Y, and Z directions, respectively;

[0115] predicted force data ={ 、 、 }, =1, 2, …, N;

[0116] wherein, the predicted force data of the kth leg at time t, 、 、 are the force components in X, Y, and Z directions, respectively;

[0117] The method for generating early warning instructions comprises:

[0118] comparing the real-time force scalar of each leg with its preset threshold value, and calculating the overrun ratio;

[0119] The calculation formula is: ; wherein is the overrun ratio, is the preset threshold value, is the total force of the leg; wherein, ;

[0120] When the overrun ratio is ≥1, it is determined to be abnormal, and the predicted force comparative analysis is performed to generate early warning instructions;

[0121] When the overrun ratio is <1, no early warning instructions are generated;

[0122] The method for performing predicted force comparative analysis to generate early warning instructions comprises:

[0123] based on the predicted force data of each leg at time , calculating the force deviation;

[0124] The calculation formula of the force deviation is:

[0125]

[0126] The acquired stress deviation is compared and analyzed with a preset deviation threshold value, and when the stress deviation is greater than or equal to the deviation threshold value, a structural abnormality early warning instruction is generated;

[0127] When the stress deviation is less than the deviation threshold value, an unloading partial load abnormality instruction is generated.

[0128] It should be noted that each preset threshold value in the above analysis process is set by a person skilled in the art according to the actual situation or obtained through a large amount of data simulation.

[0129] Specifically, through the steps of coordinate system establishment, multi-source data same frequency collection, three-dimensional material reconstruction, prediction model calculation and deviation comparison and analysis, fine monitoring of the running state of the silo is realized. Compared with the traditional method of only relying on the threshold value of the support leg strain gauge, the present method can simultaneously consider the unloading behavior, material distribution, center of gravity deviation and self weight of the silo and other key factors, and significantly improves the stress state interpretation ability.

[0130] In the case of complex unloading conditions, uneven material flow, local blockage and arch collapse, the present method can still accurately determine the source of stress change, avoid false alarm caused by misjudging unloading partial load as structural abnormality, and identify the real existing abnormal stress or structural risk, thereby improving the accuracy and reliability of the early warning.

[0131] Embodiment 2

[0132] Please refer to Figure 4 The embodiment of the present application provides a technical scheme: a silo support leg stress monitoring and early warning system for realizing the silo support leg stress monitoring and early warning method, the system comprising:

[0133] A data acquisition module establishes a silo coordinate system, acquires the coordinates of each silo support leg, and sets a label for each support leg, and acquires the density data of the material and the weight data of the silo;

[0134] A parameter acquisition module constructs a multi-source feature data same frequency collection system including unloading valve position signal, material flow speed and strain stress of each support leg, and acquires multi-source feature data in real time, based on the unloading valve position signal, labels the unloading event and outputs the event time window;

[0135] A data analysis module acquires material image data in the silo in real time when in the unloading event time window, and calculates and acquires the grid mass field and the material center of gravity position based on the material image data and the density data;

[0136] The stress prediction module inputs the grid mass field, the center of gravity position of the material at the current time, the weight data of the silo, and the coordinates of the silo support leg into a pre-constructed support leg stress prediction model, and outputs the predicted stress data of the support leg;

[0137] The early warning generation module compares the real-time collected support leg stress data with the predicted stress data of each support leg performs comparative analysis, and generates a warning instruction.

[0138] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the claims and their equivalents.

Claims

1. A method for monitoring and early warning of stress on silo outriggers, characterized in that, include: Establish a silo coordinate system, obtain the coordinates of each silo support leg, set a label for each support leg, and obtain the density data of the material and the weight data of the silo. Construct a multi-source feature data synchronous acquisition system that includes unloading valve position signal, material flow velocity, and strain stress of each support leg; collect multi-source feature data in real time; and label unloading events and output event time windows based on unloading valve position signal. When the unloading event time window is in progress, real-time material image data in the silo is collected, and the grid mass field and the position of the material center of gravity are calculated based on the material image data and density data. Input the current material flow velocity, grid mass field, center of gravity position, silo weight data, and silo leg coordinates into the pre-built leg force prediction model, and output the predicted force data of the leg. By collecting real-time force data on the outriggers Predicted force data for each outrigger Comparative analysis is performed to generate early warning instructions.

2. The method for monitoring and early warning of stress on silo outriggers according to claim 1, characterized in that, The method for obtaining the grid mass field and the position of the material's center of gravity includes: Several depth camera units are arranged around the top of the silo. Each depth camera unit integrates an active light source component and a depth camera. The active light source component is used to project a preset coded grating onto the material surface. The depth camera collects the grating deformation map reflected from the material surface to obtain preliminary depth point cloud data. Using the calibrated camera extrinsic parameters, the depth point cloud data of each depth camera is projected onto the silo coordinate system to obtain the point cloud set corresponding to each depth camera. Based on the iterative nearest point registration algorithm, the geometric consistency of the overlapping area of ​​the point cloud is corrected, and on this basis, the material surface height field data of the overall material surface geometry reconstruction is obtained. The bottom surface of the silo is discretized into a grid. Based on the material surface height field data, the volume and unit mass of each grid are obtained to obtain the grid mass field. The location of the material's center of gravity is obtained based on grid mass field calculations.

3. The method for monitoring and early warning of stress on silo outriggers according to claim 2, characterized in that, Methods for correcting geometric consistency in overlapping point cloud regions based on iterative nearest-point registration algorithms, and obtaining overall material surface geometry reconstruction data based on this, include: Step 1: Set the point cloud data obtained by any depth camera unit as the reference point cloud. , denoted as: , Indicating the first point cloud The coordinates of a 3D point are used to set the point cloud data obtained from other depth camera units as the point cloud to be registered. , denoted as: , Indicating the first point cloud The coordinates of a three-dimensional point; Step 2, in the reference point cloud and the point cloud to be matched In this context, the set of corresponding point pairs is obtained through nearest-neighbor search, denoted as: C = ( , ); where C is the set of corresponding point pairs, ∈ That is, the first point in the reference point cloud One point; ∈ That is, the first point in the point cloud to be registered One point; Step 3: Based on the obtained set of corresponding point pairs C, construct an error metric function E to measure the difference between two points, and calculate the overall error. The expression for the error metric function E is: ; For the total error, This represents a 3×3 rotation matrix used to describe spatial rotations. This represents a 3×1 translation vector used to describe spatial offset. Indicates the point After rotation R and displacement The point obtained later; Step 4: Minimize the error metric function E to obtain a set of optimal parameters. ,in, Represents the optimal rotation matrix. Represents the optimal translation vector; Step 5, select the optimal rotation matrix. and optimal translation vector Acting on all points in the point cloud to be registered This allows us to obtain the transformed point cloud to be registered. ; Step 6, obtain the point cloud to be registered. Then, compared with the reference point cloud Perform the nearest point search again to obtain a new set of corresponding point pairs C'. Repeat steps 3-5 until the total error change between two adjacent iterations is less than the preset error threshold or the preset maximum number of iterations is reached. Step 7: Register the point cloud after geometric consistency correction. With reference point cloud By merging the data, we obtain complete and consistent reconstructed material surface height field data.

4. The method for monitoring and early warning of stress on silo outriggers according to claim 1, characterized in that, The training method for the outrigger force prediction model includes: H sets of training data are collected in advance, where H is a positive integer greater than 0. The training data includes the grid mass field of the material at the current moment, the position of the center of gravity, the weight data of the silo, the coordinates of the silo legs, and the predicted force data of the corresponding legs. The gradient boosting regression tree model was used as the outrigger force prediction model. Initial hyperparameters were set, and the collected training data were divided into training set, validation set and test set according to a preset ratio. The model is trained using the training set, with mean squared error as the loss function. A simple model is initialized, and a decision tree is built iteratively. Each new tree is designed to fit the negative gradient of the loss function of the preceding model on the training set. The weights of the leaf nodes of the new tree are optimized using gradient descent and integrated into the existing model with a preset learning rate. The hyperparameters are tuned using Bayesian optimization to optimize performance on the validation set. Every 20 trees are iterated, the model's performance metrics are calculated using the validation set. When the root mean square error on the validation set decreases by less than a preset threshold for several consecutive iterations, the model's performance is considered to have stabilized, and training is stopped. The trained model is evaluated using a test set. Once the model's performance meets the evaluation criteria, it is deployed and applied.

5. The method for monitoring and early warning of stress on silo outriggers according to claim 1, characterized in that, The multi-source feature data synchronous acquisition system includes a limit switch or potentiometer-type position sensor, a mass flow meter, and a three-dimensional force sensor; The multi-source feature data includes unloading valve position signal, material flow velocity, and force data of each support leg.

6. The method for monitoring and early warning of stress on silo outriggers according to claim 1, characterized in that, The method for generating early warning instructions includes: The real-time force scalar of each outrigger is compared with its preset threshold to calculate the over-limit ratio; When the over-limit ratio is ≥1, it is judged as abnormal, and a predictive force comparison analysis is performed to generate an early warning command; When the over-limit ratio is less than 1, no warning instruction will be generated.

7. The method for monitoring and early warning of stress on silo outriggers according to claim 6, characterized in that, The method for generating early warning commands through predictive force comparison analysis is as follows: Based on the predicted force data, the force deviation is calculated, and the obtained force deviation is compared and analyzed with the preset deviation threshold. When the force deviation is greater than or equal to the deviation threshold, a structural anomaly warning instruction is generated. When the force deviation is less than the deviation threshold, an abnormal unloading load command is generated.

8. The method for monitoring and early warning of stress on silo outriggers according to claim 1, characterized in that, The method for establishing the silo coordinate system is as follows: take the center point at the bottom of the silo as the origin, take two horizontal lines that are perpendicular to each other on the horizontal plane as the horizontal and vertical coordinates, and take the vertical center line of the silo as the vertical coordinate to establish the silo coordinate system.

9. The method for monitoring and early warning of stress on silo outriggers according to claim 1, characterized in that, The force data of the outrigger ={ , , }, =1, 2, ..., N, where N is the number of legs. Force data of the k-th outrigger at time t , , These are the force components in the X, Y, and Z directions, respectively. Predicted force data ={ , , }, =1, 2, ..., N, where, Predicted force data for the k-th outrigger at time t , , These are the force components in the X, Y, and Z directions, respectively.

10. A silo outrigger stress monitoring and early warning system, used to implement the silo outrigger stress monitoring and early warning method according to any one of claims 1-9, characterized in that, The system includes: The data acquisition module establishes a silo coordinate system, acquires the coordinates of each silo support leg, sets a label for each support leg, and acquires the density data of the material and the weight data of the silo. The parameter acquisition module constructs a multi-source feature data synchronous acquisition system that includes unloading valve position signal, material flow velocity, and strain stress of each leg. It collects multi-source feature data in real time, and based on the unloading valve position signal, it marks unloading events and outputs event time windows. The data analysis module collects material image data in the silo in real time during the unloading event time window, and calculates the grid mass field and the position of the material center of gravity based on the material image data and density data. The stress prediction module takes the current material's grid mass field, center of gravity position, silo weight data, and silo leg coordinates as input to the pre-built leg stress prediction model and outputs the predicted stress data of the legs. The early warning generation module uses real-time collected outrigger force data. Predicted force data for each outrigger Comparative analysis is performed to generate early warning instructions.

Images