An automated sampling control method and system for a quantitative weighing bin

Abstract

The application relates to the technical field of intelligent control, and discloses an automatic sampling control method and system for a quantitative weighing bin, which comprises the following steps: processing signals collected by intelligent sensors to construct an initial density distribution graph; after sampling and data processing are performed on the distribution graph, a smooth density distribution is generated; by analyzing density differences in the smooth density distribution, a component segregation area is identified, and a distribution rule thereof is extracted, so that a target segregation area is located and a depth analysis interval is determined; a quantitative relationship model between depth and component deviation is established in the interval, and a depth deviation distribution is calculated; by analyzing deviation gradients, candidate depth points are screened out, and an optimal sampling depth is determined by taking component deviation minimization as a target; finally, a sampling device is accurately positioned and sampled according to the depth value, and representative sample information is generated according to collected sample data. The method can improve the extraction accuracy of a representative area.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to an automated sampling control method and system for a quantitative weighing bin. Background Technology

[0002] Currently, in the field of bulk material storage and processing, quantitative weighing silos serve as a crucial link connecting raw material supply and production processes. The accuracy of their sampling directly determines the reliability of product quality control and the stability of enterprise economic benefits. Due to the diverse types of materials and their significant differences in properties, and the segregation that occurs during settling within the silo due to variations in particle size and density, resulting in a non-uniform composition distribution in the vertical direction, obtaining representative average samples has become a core challenge for the industry. The pressing technical challenge for the industry is how to accurately and in real-time perceive the distribution of materials within the silo and dynamically determine the optimal sampling location to overcome the effects of segregation.

[0003] In a current technology, automated sampling control of quantitative weighing bins primarily relies on mechanized sampling operations at fixed depths or preset locations. This method achieves automated execution in hardware, such as by controlling a robotic arm to grasp or drill samples at preset coordinate points via a program. However, its technical logic has fundamental shortcomings: it is entirely based on static pre-settings and lacks an intelligent sensor network capable of sensing the real-time distribution of materials and corresponding data analysis capabilities. Therefore, it cannot detect and respond to the complex segregation phenomena that occur in the bin in real time due to gravity after each feeding. Because the selection of sampling points is fixed, it is impossible to dynamically track the areas with the most significant or least representative segregation. Consequently, when segregation is severe, the collected samples may originate from localized locations with abnormal composition (such as enrichment of large particles or fine powder), thus failing to accurately reflect the average characteristics of the entire bin's material.

[0004] In summary, existing technologies suffer from low extraction accuracy of representative regions. Summary of the Invention

[0005] This invention provides an automated sampling control method and system for a quantitative weighing bin to solve the problem of low regional extraction accuracy in existing technologies.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides an automated sampling control method for a quantitative weighing bin, comprising:

[0007] Obtain an initial signal set, extract density distribution data based on the initial signal set, and construct an initial density distribution map;

[0008] Sampling is performed on the initial density distribution map to obtain a density value sequence. Anomaly detection and data transformation are performed on the density value sequence to generate a two-dimensional density distribution map. A smooth density distribution is obtained by smoothing the two-dimensional density distribution map.

[0009] Analyze the density differences in the smooth density distribution, identify component segregation regions, perform spatial mapping on the identified segregation regions to generate feature maps, and extract the distribution patterns of the segregation regions based on the feature maps;

[0010] Based on the distribution pattern, target areas with density differences exceeding a preset threshold are identified from the segregated regions. Key coordinate points are extracted from the target areas, and an initial depth range is determined. The initial depth range is then filtered to determine the depth analysis interval.

[0011] Based on the depth analysis interval, a relationship model between depth and component deviation is established. The component deviation value is calculated based on the relationship model, and the component deviation value is smoothed to obtain the depth deviation distribution.

[0012] The deviation gradient is calculated based on the depth deviation distribution. A set of candidate depth points is selected based on the deviation gradient. The optimal depth point is determined from the set of candidate depth points with the minimization of component deviation as the guide. The final sampling depth value is determined by parsing the graphical depth information of the optimal depth point.

[0013] Based on the final sampling depth value, the position deviation is calculated. Combined with the position deviation, the position of the sampling device is adjusted through feedback control. The adjusted device position is matched with the preset depth partition to obtain the material sample data at the corresponding depth. Based on the material sample data, representative sample information is generated.

[0014] Secondly, the present invention provides an automated sampling control system for a quantitative weighing bin, comprising:

[0015] The data acquisition module is used to acquire an initial signal set, extract density distribution data based on the initial signal set, and construct an initial density distribution map;

[0016] The density optimization module is used to sample the initial density distribution map to obtain a density value sequence, perform anomaly detection and data transformation on the density value sequence to generate a two-dimensional density distribution map, and obtain a smooth density distribution by performing smoothing calculation on the two-dimensional density distribution map.

[0017] The feature extraction module is used to analyze the density differences in the smooth density distribution, identify component segregation regions, perform spatial mapping on the identified segregation regions, generate feature maps, and extract the distribution patterns of the segregation regions based on the feature maps.

[0018] The interval positioning module is used to determine the target area with a density difference exceeding a preset threshold from the segregated area according to the distribution pattern, extract key coordinate points from the target area and determine the initial depth range, and filter the initial depth range to determine the depth analysis interval.

[0019] The distribution refinement module is used to establish a relationship model between depth and component deviation based on the depth analysis interval, calculate the component deviation value based on the relationship model, and smooth the component deviation value to obtain the depth deviation distribution.

[0020] The depth decision module is used to calculate the deviation gradient based on the depth deviation distribution, select a set of candidate depth points based on the deviation gradient, determine the optimal depth point from the set of candidate depth points with the minimization of component deviation as the guide, and determine the final sampling depth value by parsing the graphical depth information of the optimal depth point.

[0021] The sampling execution module is used to calculate the position deviation based on the final sampling depth value, and adjust the position of the sampling device through feedback control based on the position deviation. The adjusted device position is matched with the preset depth partition to obtain the material sample data at the corresponding depth, and representative sample information is generated based on the material sample data.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] (1) The present invention first preprocesses the initial signal set collected by the sensor, effectively eliminating measurement noise and environmental interference, and obtaining a high-quality denoised signal sequence; based on this, density distribution data is generated and an initial density distribution map is constructed. This step ensures that all subsequent analyses are based on accurate and reliable data, providing clean and reliable input for the entire analysis process, avoiding misjudgments caused by data noise from the root, and significantly improving the accuracy of subsequent fault judgment (such as segregation identification).

[0024] (2) This invention obtains a smooth density distribution that truly reflects the material distribution trend by performing anomaly detection, transformation, and smoothing calculations on density distribution data. This process filters out random fluctuations and singularities in the original data, allowing the overall pattern and local characteristics of density changes to be clearly and stably presented. This lays a solid foundation for the subsequent accurate identification of the key fault state of material segregation, making the location and analysis of abnormal component areas more reliable, thereby directly improving the accuracy of the system's judgment on abnormal material conditions in the warehouse.

[0025] (3) This invention performs in-depth analysis of smooth density distribution, identifies component segregation regions and extracts their distribution patterns, thereby accurately locating target regions with significant density differences and screening out key depth analysis intervals. This step achieves layer-by-layer focusing from macroscopic distribution to microscopic anomaly regions, ensuring that analytical resources are concentrated on the key depth ranges most likely to experience severe segregation and most significantly affect sample representativeness. This targeted depth interval positioning method greatly improves the efficiency and accuracy of subsequent depth analysis, avoids invalid calculations in areas with no or slight anomalies, and makes the location of fault (segregation) regions more accurate.

[0026] (4) Within a defined depth analysis range, this invention establishes a quantitative relationship model between depth and component deviation, and calculates the refined depth deviation distribution. This step transforms spatial distribution information into quantifiable deviation data, and reveals the continuous and stable law of deviation variation with depth through smoothing processing. By constructing this model, the system can accurately quantify the degree of component deviation of the material at each depth point, providing a direct and reliable mathematical basis for the next stage of scientifically selecting the optimal sampling point, thereby ensuring the objectivity and accuracy of sampling depth decision-making in terms of mechanism.

[0027] (5) Based on the depth deviation distribution, this invention ultimately determines the theoretically optimal sampling depth point by calculating the deviation gradient and optimizing with the goal of minimizing component deviation, and then analyzes to obtain the final sampling depth value. When this depth is located in the operation-sensitive central region, the system can dynamically adjust the position of the sampling device through feedback control and finally acquire sample data. This step ensures that the execution of the sampling action is accurately aligned with the most representative depth point obtained through the whole process analysis and calculation, and compensates for possible errors in mechanical positioning through closed-loop control. The entire closed-loop process from "data analysis" to "physical execution" fundamentally guarantees the representativeness of the collected samples and realizes the final accuracy of the sampling results. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of an automated sampling control method for a quantitative weighing bin provided in the first embodiment of the present invention;

[0029] Figure 2 This is a schematic diagram of an automated sampling control system for a quantitative weighing bin provided in the second embodiment of the present invention. Detailed Implementation

[0030] 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.

[0031] Reference Figure 1 The first embodiment of the present invention provides an automated sampling control method for a quantitative weighing bin, comprising the following steps:

[0032] S11, Obtain an initial signal set, extract density distribution data based on the initial signal set, and construct an initial density distribution map;

[0033] S12, sample the initial density distribution map to obtain a density value sequence, perform anomaly detection and data transformation on the density value sequence to generate a two-dimensional density distribution map, and obtain a smooth density distribution by performing smoothing calculation on the two-dimensional density distribution map;

[0034] S13, Analyze the density differences in the smooth density distribution, identify component segregation regions, perform spatial mapping on the identified segregation regions, generate feature maps, and extract the distribution patterns of the segregation regions based on the feature maps;

[0035] S14, Based on the distribution pattern, determine the target area where the density difference exceeds a preset threshold from the segregated area, extract key coordinate points from the target area and determine the initial depth range, and filter the initial depth range to determine the depth analysis interval;

[0036] S15, Based on the depth analysis interval, establish a relationship model between depth and component deviation, calculate the component deviation value based on the relationship model, and smooth the component deviation value to obtain the depth deviation distribution;

[0037] S16, calculate the deviation gradient based on the depth deviation distribution, select a set of candidate depth points based on the deviation gradient, determine the optimal depth point from the set of candidate depth points with the minimization of component deviation as the guide, and determine the final sampling depth value by parsing the graphical depth information of the optimal depth point.

[0038] S17. Calculate the position deviation based on the final sampling depth value. Combine the position deviation with feedback control to adjust the position of the sampling device. Match the adjusted device position with the preset depth partition to obtain material sample data at the corresponding depth. Generate representative sample information based on the material sample data.

[0039] In step S11, an initial signal set is obtained, and density distribution data is extracted based on the initial signal set to construct an initial density distribution map, including:

[0040] An initial signal set is obtained, and time-frequency conversion and feature extraction are performed on the initial signal set to obtain a frequency domain signal sequence;

[0041] If the high-frequency component of the frequency domain signal sequence is greater than a preset frequency threshold, the high-frequency component is filtered to obtain a denoised signal sequence.

[0042] Based on the denoised signal sequence, density distribution data is generated by mapping, and interpolation calculation is performed on the density distribution data to generate an initial density distribution map.

[0043] It should be noted that the initial signal set is obtained by deploying a multi-point intelligent sensor grid on the walls and bottom of the weighing bin, synchronously collecting raw physical signals (such as pressure and vibration) reflecting the material accumulation state at each point at fixed time intervals. This set forms the data foundation for all subsequent analyses. Mathematically, this set can be represented as an M×N time-domain data matrix X, where M is the total number of sensor channels and N is the number of sampling points for each channel. The feature extraction process first preprocesses each row of this matrix (i.e., the single-channel time-domain sequence), including subtracting the mean to eliminate DC offset, and applying a Hamming window function to window the signal to reduce spectral leakage. Specifically, the windowing process multiplies the sequence of length N by a Hamming window coefficient vector of the same length. The smooth transition to zero at both ends of the time domain of this window function can effectively suppress the spectral sidelobes generated by the Fourier transform.

[0044] In the frequency domain transformation stage, a Fast Fourier Transform (FFT) is performed on the windowed sequence. Essentially, this involves solving the Discrete Fourier Transform (DFT), converting the time-domain sampling points into an equal number of complex frequency points uniformly distributed between 0 and the sampling frequency. Each complex number corresponding to a frequency point contains both the amplitude and phase information of that frequency component. Subsequently, by calculating the positive square root of the sum of the squares of the real and imaginary parts of each complex frequency point, the amplitude of that frequency component is obtained, thus converting the complex spectrum into an amplitude spectrum containing only amplitude information. The horizontal axis of this amplitude spectrum represents the frequency, and the vertical axis represents the energy of the original signal at that frequency. Finally, the amplitude spectra of all channels are arranged according to the spatial location of the sensors, collectively forming the frequency domain signal sequence for subsequent frequency domain analysis of the system.

[0045] If the high-frequency components of the frequency domain signal sequence are greater than a preset frequency threshold, the high-frequency components are filtered to obtain a denoised signal sequence. Specifically, this is achieved by designing a low-pass digital filter with a cutoff frequency matched to the preset threshold, such as a Butterworth or Chebyshev filter. The filter design first determines its normalized cutoff frequency based on the preset frequency threshold, and then selects between Butterworth (maximum flat response) and Chebyshev (steeper transition band but allowing passband ripple) types based on specific requirements for passband flatness and transition band steepness. Subsequently, based on the selected type, cutoff frequency, and desired passband ripple, stopband attenuation, and other indicators, the minimum order of the filter is determined by calculation or table lookup. The system function of the digital filter, i.e., the coefficients of its difference equation, is calculated using the corresponding design formula (e.g., bilinear transformation based on an analog prototype filter). This filter significantly attenuates all components in the spectrum above the cutoff frequency. The filtered spectrum is then reconstructed back to the time domain using an inverse Fourier transform, resulting in a clean denoised signal sequence free of high-frequency interference.

[0046] The preset frequency threshold is determined by performing spectral analysis on a large number of historical signal samples. The specific method is as follows: First, sensor signals are collected under typical operating conditions and their spectrum is calculated. Then, the frequency band in which the effective signal energy corresponding to the material density distribution characteristics is mainly concentrated is analyzed, and the frequency bands of typical high-frequency noise introduced by mechanical vibration and electrical interference are identified. Finally, the threshold is set in a safe transition zone between the upper limit of the effective signal frequency band and the lower limit of the main noise frequency band. It is usually selected as 1.2 to 1.5 times the highest significant frequency component of the effective signal to ensure that all effective material information is retained while filtering out high-frequency noise to the maximum extent.

[0047] Based on the denoised signal sequence, density distribution data is generated by mapping. This is achieved by associating the coordinate position of each smart sensor with its corresponding denoised signal amplitude (converted to a density value by a calibration model), and filling it into a preset two-dimensional or three-dimensional grid matrix according to spatial positional relationships. Specifically, firstly, based on a pre-established calibration function, the denoised signal amplitude at each sensor location (usually the root mean square value of the time-domain signal or the energy of a specific frequency band) is converted into the corresponding material density value. This calibration function is determined through laboratory experiments, establishing a mapping relationship between signal characteristics and known standard material densities. Next, based on the actual geometric dimensions of the silo and the physical installation coordinates of the sensors, a regular spatial grid matrix is ​​initialized in computer memory, with each cell of the matrix corresponding to a small volume element within the silo. Then, the density value calculated for each sensor is filled into one or more grid cells in the grid matrix closest to the sensor's coordinates. Each grid cell stores the estimated density value of its corresponding spatial region, thus forming the density distribution data.

[0048] The calibration function is established through a systematic laboratory calibration process: In a controlled laboratory environment, the sensor is installed on the simulated warehouse wall, and a series of standard material samples with known precise densities are filled in; for each standard density material, the raw signal output by the sensor is collected, and after the same denoising and feature extraction process, a stable signal feature value (such as the root mean square value) is calculated; subsequently, with the signal feature value as the independent variable and the known material density as the dependent variable, regression analysis is performed on multiple sets of experimental data points using curve fitting techniques such as the least squares method, thereby fitting a continuous function (such as a linear or polynomial function), which defines a unique mapping relationship from the signal feature value to the material density value.

[0049] The initial density distribution map is generated by interpolating the density distribution data because the sensor deployment cannot completely cover the entire warehouse space and some grid cells have no direct measurement data. Therefore, spatial interpolation algorithms, such as bilinear interpolation or inverse distance weighted interpolation, are required to fill all blank cells one by one, resulting in a complete density field matrix in which each grid cell has a density estimate.

[0050] Finally, the system calls the graphics library to map the matrix values ​​into colors or grayscale, rendering them as a pixel array on a two-dimensional plane or constructing a three-dimensional volume to generate an initial density distribution map that visually reflects the spatial changes in material density within the silo. The density values ​​of these blank areas are estimated using the density values ​​of surrounding known grid points, ultimately calculating a continuous density distribution surface or volume data covering the entire silo space, and outputting it as a visualized graphic. This graphic serves as the initial density distribution map upon which subsequent analysis is based.

[0051] It is important to note that after obtaining the initial density distribution data, this method does not directly use the original discrete data for subsequent analysis. Instead, it deliberately introduces the step of "first generating an initial density distribution map, and then sampling from the map." The core purpose of this step is to solve a key practical engineering problem: the initial data comes from discretely distributed smart sensors, and its spatial resolution and uniformity cannot directly meet the high requirements for data regularity and continuity in subsequent refined analysis. Generating a continuous "initial density distribution map" through interpolation essentially constructs a standardized data field model covering the entire warehouse space. Subsequently, regularized "sampling" is performed on this model, which is not limited by the original sensor placement and obtains density value sequences according to the optimal resolution and standardized spatial locations required by the algorithm. This design ensures the analytical accuracy, result reliability, and system robustness of the entire automated sampling control method.

[0052] In step S12, sampling is performed on the initial density distribution map to obtain a density value sequence. Anomaly detection and data transformation are then performed on the density value sequence to generate a two-dimensional density distribution map. A smoothed density distribution is obtained by smoothing the two-dimensional density distribution map, including:

[0053] The initial density distribution map is sampled to obtain a density value sequence, and anomaly detection is performed on the density value sequence to filter out a valid data set;

[0054] The effective data set is transformed into a coordinate matrix and mapped to a two-dimensional data matrix. Based on the two-dimensional data matrix, a two-dimensional density distribution map is reconstructed and generated.

[0055] The two-dimensional density distribution map is smoothed to obtain a smoothed density distribution.

[0056] It should be noted that the density numerical sequence obtained by sampling the initial density distribution map is specifically achieved through a vertical stratified sampling algorithm. The specific execution process of this algorithm is as follows: First, the system divides the vertical direction into several continuous depth bands based on the actual height of the silo and the preset stratification density (e.g., one layer every 0.5 meters or 1 meter). Next, within each depth band, multiple vertical lines are evenly arranged along the horizontal direction. The number of these vertical lines is determined by the cross-sectional dimensions of the silo and the required sampling resolution. The precise three-dimensional coordinates of the sampling points are determined on each vertical line at a fixed vertical interval (e.g., 0.1 meters). Then, for each sampling point, the system maps its coordinates onto the discrete data grid corresponding to the initial density distribution map and calculates the density estimate at that coordinate location using a bilinear interpolation algorithm. This interpolation process uses a weighted average of the known density values ​​of the four nearest grid points. Finally, the density values ​​calculated from all sampling points are sorted and linearly arranged according to their depth coordinates in ascending order, thereby outputting a continuous one-dimensional density numerical sequence.

[0057] The core of anomaly detection and filtering of the one-dimensional density numerical sequence to obtain a valid data set is the application of a statistical outlier detection method. The specific process is as follows: First, the system calculates the arithmetic mean and standard deviation of the sequence to characterize the central location and dispersion of the data. Then, based on the three-standard-deviation principle in statistics, the upper and lower bounds of the anomaly detection threshold are dynamically set. Next, the system traverses each data point in the sequence, judging whether its value exceeds the threshold range. After completing the full sequence traversal, the system removes all marked outliers from the original sequence. Finally, the system reorganizes all remaining unmarked data points, i.e., those falling within the threshold range, into a new sequence according to their original depth order in the sequence. This new sequence is the filtered, effective data set free of anomaly interference.

[0058] The effective data set is transformed into a two-dimensional data matrix. This process is based on the spatial position of each effective data point in the original distribution map. Specifically, the system reads the original three-dimensional spatial coordinates from the metadata associated with each effective data point. These coordinates typically include depth values ​​(Z-axis) and two-dimensional horizontal coordinates (X-axis and Y-axis). First, the system normalizes the horizontal coordinates (X, Y) to a scalarized horizontal position value by calculating the horizontal distance to the center of the storage area or a specified reference point, or by directly selecting a representative axial coordinate (such as the X-axis). This value serves as the x-coordinate index of the two-dimensional matrix. Simultaneously, the depth value (Z) is linearly scaled and used as the y-coordinate index. Next, the system determines the number of rows (corresponding to depth resolution) and columns (corresponding to horizontal position resolution) of the two-dimensional matrix based on the range of horizontal positions and depth values ​​of all data points. Then, the system calculates the corresponding matrix row and column numbers for each data point, typically by subtracting the minimum value from its coordinate value, dividing by the resolution interval, and rounding. Finally, an empty matrix with a defined number of rows and columns is created, and the density value of each data point is... Fill the corresponding rows and columns into the matrix cells determined by the calculation, thereby generating an initial, possibly sparse, two-dimensional data matrix.

[0059] A two-dimensional density distribution map is generated by reconstructing the two-dimensional data matrix. Since the data points in the initial matrix may be unevenly distributed, a grid reconstruction technique is required. This reconstruction process is achieved through the following steps: First, the system defines a new regular grid with a resolution much higher than the original data and a fixed number of rows and columns within the target space. Next, the system traverses each grid point in the new grid. For the current grid point, within a preset search radius, it searches for all valid known data points in the original sparse matrix. For each found known point, the system calculates its straight-line distance to the current grid point and uses the inverse distance weighting algorithm to take the "reciprocal of the square of the distance" as the weight of the known point. Subsequently, the system multiplies the density values ​​of all known points by their respective weights, sums these products, and divides by the sum of all weights to calculate the estimated density value of the current new grid point. By traversing and calculating all grid points in the new grid, the system finally obtains a complete and uniform two-dimensional data matrix with an estimated density value at each location. This matrix is ​​then visualized and rendered to generate a continuous two-dimensional density distribution map.

[0060] The smoothing process first requires generating a two-dimensional Gaussian convolution kernel. Based on a preset smoothing intensity (controlled by the standard deviation parameter) and kernel size, the system calculates the Gaussian function value corresponding to each position within the kernel. This function value is determined by the distance from that position to the kernel center; the closer the distance, the larger the function value. Subsequently, the function values ​​at all positions are processed using a min-max normalization method, summing them to one. These normalized values ​​constitute the weight coefficients of the convolution kernel. During smoothing calculations, the system aligns the center of this convolution kernel sequentially with each pixel in the distribution map, aligning the kernel's coverage with a local neighborhood of that pixel. For each position, the system multiplies the original density value of each pixel in the neighborhood by... The system uses the weight coefficients corresponding to the convolution kernel positions, and then adds all these product results together. The sum is the new density value of the current center pixel after smoothing. The system traverses and calculates the new value of each pixel in the order from the top left corner to the bottom right corner of the image. For pixels at the image edges, the system uses a symmetrical filling method to expand the image boundary to ensure that the convolution operation can be performed normally. After completing the calculation of all pixels in the entire image, each pixel value in the original distribution map is replaced by its corresponding weighted average value. Finally, a density distribution map that effectively suppresses random fluctuations and has a smoother spatial transition is output, i.e., a smooth density distribution.

[0061] In step S13, the density differences in the smooth density distribution are analyzed, component segregation regions are identified, spatial mapping is performed on the identified segregation regions to generate feature maps, and the distribution patterns of the segregation regions are extracted based on the feature maps, including:

[0062] Calculate the density difference sequence of the smooth density distribution, and identify the component segregation regions based on the density difference sequence;

[0063] Data spatial mapping is performed on the component segregation regions to generate a segregation distribution feature map;

[0064] Cluster analysis is performed on the segregation distribution feature map to extract and analyze the geometric center trajectory of the clusters, thereby obtaining the distribution pattern of the segregation region.

[0065] It should be noted that the specific implementation of calculating the density difference sequence of the smooth density distribution is as follows: The system first reads the two-dimensional data matrix corresponding to the smooth density distribution. This matrix has R rows (representing the depth direction) and C columns (representing the horizontal position). The system processes the data of each column sequentially, starting from the first column. For each column, the system traverses downwards from the second row (i.e., the row with depth index 2) to the last row. During the traversal, for the currently processed i-th row, the system reads the density value of the grid point at that location. Simultaneously, read the density value of the grid point directly above it (i.e., the (i-1)th row). Next, the system calculates the difference between the two values. This difference This refers to the vertical density difference between the current depth point and the points above it; the system will calculate each The values ​​are stored in a list according to their corresponding column and row numbers (i.e., horizontal position and depth); after all rows in all columns (except the first row of each column, which is not calculated because it has no adjacent point above) have completed the above difference calculation, this list containing all values ​​is... The list of values ​​is the final density difference sequence.

[0066] Identifying component segregation regions based on the density difference sequence is accomplished by pre-setting a density difference threshold based on historical data statistics. The system first calls a pre-trained threshold model based on a large amount of historical normal operating data. This model uses statistical analysis to determine the typical fluctuation range of the absolute value of density difference in the vertical direction, and uses this to set a dynamic difference threshold. Next, the system iterates through each data point in the density difference sequence, reads its absolute difference value, and compares this absolute value with the dynamic threshold. If the absolute difference value of a data point is greater than the threshold, the system determines that a significant density abrupt change has occurred at that point, and records the row and column indices (i.e., its specific spatial location) of that point in the original smooth density distribution two-dimensional matrix as an "anomaly marker point."

[0067] After completing the traversal and labeling of the entire sequence, the system performs spatial clustering analysis on all labeled outliers: based on the two-dimensional coordinates of each labeled point, it uses depth-first search or region growing algorithms to merge spatially adjacent labeled points (such as those in the four directions of front, back, left, and right) into the same set, forming a connected region. Each such connected region represents a continuous spatial range in the smooth density distribution map where the density changes drastically; this range is ultimately identified and output by the system as a component segregation region.

[0068] The pre-training process of the threshold model is as follows: Before system deployment, a large amount of historical smooth density distribution data generated under normal, uniform material filling conditions is collected. For each set of historical data, the system calculates its density difference sequence using a method completely consistent with the online steps, and records the absolute value of the difference for all data points in the sequence. Subsequently, the system performs statistical analysis on all historical absolute values ​​of difference, calculates their mean and standard deviation, and selects "mean plus three times the standard deviation" as the initial static threshold reference line based on the normal distribution assumption. To further optimize and adapt it to dynamic conditions, the system also calculates the high percentile (e.g., 99.5 percentile) of these historical absolute values ​​of difference, compares and integrates this percentile value with the aforementioned static threshold, and finally determines a robust difference threshold. This threshold aims to cover the vast majority of normal fluctuations and classify differences exceeding this range as abnormal. This threshold will be stored as a model parameter.

[0069] The generation of a segregation distribution feature map by mapping the data space of the component segregation regions is performed using principal component analysis (PCA), a dimensionality reduction algorithm. The system first traverses each identified component segregation region, extracting multidimensional features from each grid point within that region. These features typically include the point's depth coordinates, horizontal coordinates, smoothed absolute density value, and calculated vertical density difference value. These features at each point collectively constitute a high-dimensional feature vector. Next, the system aggregates these high-dimensional vectors from all points within all segregation regions, forming a high-dimensional data matrix where rows represent points and columns represent features. Subsequently, the system performs PCA on this data matrix. This involves first demeaning each column of feature data, then calculating the covariance matrix of the processed data matrix, and performing eigenvalue decomposition on the covariance matrix to obtain a set of eigenvectors arranged in descending order of their corresponding eigenvalues. These eigenvectors are the principal component directions. The system selects the two principal component directions with the largest eigenvalues ​​as the new coordinate axes.

[0070] Next, the system performs a dot product operation between each feature vector in the original high-dimensional data matrix and the two selected principal component directions, thereby linearly projecting each high-dimensional point onto the two-dimensional plane spanned by the two principal components to obtain its coordinates in the new two-dimensional space. Finally, the system plots the new two-dimensional coordinates of all points into a scatter plot. The position of each point in the plot reflects the distribution of its original high-dimensional features in the two most representative principal component directions. This plot is the scatter distribution feature plot that can reveal the core structural features of the scatter region.

[0071] Cluster analysis is performed on the segregated distribution feature map to extract and analyze the geometric center trajectories of the clusters to obtain the distribution pattern of the segregated regions. The cluster analysis uses a density-based clustering algorithm, with the core parameters of neighborhood radius and minimum number of points determined as follows: First, the Euclidean distances between all pairs of points in the feature map are calculated, and their average value is taken as the average point spacing. Then, 1.5 times this average point spacing is taken as the neighborhood radius. The minimum number of points, Minpts, is based on the empirical formula: Minpts = max(2, ceil(ln(N))), where N is the total number of points in the feature map. The feature map is divided into several clusters using DBSCAN clustering. Next, when extracting the geometric center trajectories, for each cluster, the system calculates the average depth and horizontal level position of all its member points in the original spatial depth-horizontal coordinate system (not the dimensionality-reduced feature map coordinates). This average coordinate is the geometric center point of this cluster. Subsequently, the system sorts all clusters according to the actual depth value of the geometric center point (from top to bottom), and connects the center points of adjacent depths with line segments in sequence to form one or more trajectory polylines that traverse different depths.

[0072] Finally, by analyzing the overall direction of these trajectory lines (such as being nearly vertical, inclined, or horizontal), the concentration of depth intervals, and the number and sharpness of turning points, the system summarizes the macroscopic distribution patterns of the segregated areas, such as whether they are vertically linearly distributed along the warehouse wall, horizontally layered at specific depths, or irregularly clustered in the middle of the warehouse.

[0073] In step S14, based on the distribution pattern, target areas with density differences exceeding a preset threshold are determined from the segregated regions. Key coordinate points are extracted from the target areas, and an initial depth range is determined. The initial depth range is then filtered to determine a depth analysis interval, including:

[0074] Based on the distribution pattern, density difference values ​​are extracted from the segregated regions. If the density difference value exceeds a preset difference threshold, the corresponding target segregated region is locked.

[0075] Extract the boundary pixels of the target segregated region and obtain the key coordinate points through spatial mapping;

[0076] Based on the key coordinate points, the corresponding initial depth range is calculated, and the depth range information is subjected to extreme value filtering to determine the preliminary depth analysis interval.

[0077] It should be noted that, according to the distribution pattern, the density difference is extracted from the segregated region. The specific process is that the system traverses each identified segregated region, and for each spatial grid point in the region, it reads the density value stored in the smooth density distribution map and calculates the absolute difference between the density value of the point and the density value of the previous grid point that is adjacent to it in the vertical direction. This generates a set of density difference values ​​for the segregated region that reflects the degree of density abrupt change in the internal vertical direction.

[0078] If the density difference value exceeds the preset difference threshold and the corresponding target segregation region is locked, it is achieved by comparing each density difference value in the set with a preset threshold (e.g., 0.5 tons / cubic meter) obtained based on material characteristics and historical data statistics in real time. Once the density difference value of any grid point in a segregation region exceeds this threshold, the system marks the entire segregation region and locks it as a target segregation region that needs further refined analysis.

[0079] The specific process for setting the difference threshold is as follows: First, the system retrieves an initial threshold range (e.g., 0.4 tons / cubic meter to 0.6 tons / cubic meter) from a pre-stored empirical parameter library based on the type of material being processed (e.g., coal type) and its typical physical properties (e.g., particle size, bulk density). Then, combining the historical settling data of the current batch of materials, the system analyzes the normal fluctuation statistical characteristics of its density difference (e.g., mean and standard deviation), and selects a specific value within this initial range as the difference threshold for this judgment. The final determination of this value must ensure that it can effectively distinguish between density changes caused by normal material packing gradients and abnormal density abrupt changes caused by severe component segregation.

[0080] Extracting the boundary pixels of the target biased region and obtaining key coordinate points through spatial mapping involves using an edge detection algorithm (such as the Canny algorithm) in image processing to process the pixel region of the target biased region on the two-dimensional projection image. First, the two-dimensional image region corresponding to the target biased region is input into the Canny edge detector. Edge detection is performed through Gaussian kernel filtering and double threshold detection to classify the pixels into strong edge points, weak edge points, and non-edge points. Finally, the strong edge points are used as the determined edges, and those weak edge points adjacent to the strong edge points are connected to obtain a complete and continuous boundary pixel chain.

[0081] The system collects the two-dimensional image coordinates (row and column indices) of all these boundary pixels. Next, spatial mapping is performed: the system calls a pre-established mapping model based on on-site calibration. This model is typically based on the pixel coordinates of multiple known three-dimensional spatial calibration points in the image, and fits an affine transformation matrix using the least squares method. For each boundary pixel's two-dimensional coordinates (u, v), the system calculates its corresponding horizontal coordinates (X, Y) in the 3D coordinate system of the container using this transformation matrix, while its depth coordinate Z is directly converted from the image row number of the pixel using a preset depth-row number linear calibration relationship. Finally, the set of all calculated (X, Y, Z) coordinate points constitutes the key coordinate points for accurately describing the spatial contour of the target segregated region.

[0082] The specific setting process for the high and low thresholds employs an adaptive method to ensure stable edge detection results across different images. First, after performing non-maximum suppression, the system statistically analyzes all retained gradient magnitudes and plots their histograms. The high threshold is typically determined based on the statistical distribution of gradient magnitudes; for example, it can be set to the magnitudes of the top 30% of all non-zero gradient magnitudes after sorting by size, or it can be calculated using automatic thresholding methods such as Otsu's method. The low threshold is dynamically calculated based on a preset high-to-low threshold ratio (e.g., 3:1).

[0083] The initial depth range is calculated based on the key coordinate points. This is achieved by the system reading the depth (Z-axis) coordinate values ​​of all key coordinate points and finding the minimum value among them. and maximum value Directly divide the interval The initial depth range is determined; extreme value filtering is performed on the depth range information to determine the preliminary depth analysis interval, specifically as follows: The system first determines the initial depth range based on its upper and lower bounds. and The system generates a dense sequence of depth sampling points using linear interpolation, based on a preset physical step size. Then, it quickly indexes the stored global density difference data array using the horizontal coordinates of each sampling point and reads the corresponding vertical density difference value. Next, the system compares the density difference value of each sampling point with a secondary difference threshold pre-set according to the normal settling gradient of the material. All sampling points with density difference values ​​below this secondary threshold are considered insignificant and removed from the sequence. Finally, the system iterates through the remaining valid sampling point sequence to find the minimum and maximum depth values. The closed interval determined by these two extreme values ​​is the preliminary depth analysis interval, focusing on areas of significant density change, obtained after filtering.

[0084] The secondary difference threshold is set based on a theoretical settling model of the material under gravity and combined with operational experience. The specific setting process is as follows: First, based on the typical particle size distribution, density, and internal friction angle of the material being processed, the theoretical maximum vertical density gradient that can be generated by natural particle rearrangement under static stacking conditions is calculated using settling theory models such as the Jensen formula, and this is converted into a theoretical value of density difference per unit depth. Then, this theoretical value is multiplied by an empirical safety factor (usually 0.6 to 0.8) to obtain a conservative reference threshold. Finally, this reference threshold is comprehensively compared and fine-tuned with a large number of historical statistical distributions of actual density differences observed under normal operating conditions (such as the mean plus one standard deviation) to ensure that the final set secondary threshold can reliably distinguish between "expected density changes caused by natural particle settling" and "abnormal density mutations caused by segregation of actual components," thus retaining the depth points corresponding to the latter during screening.

[0085] In step S15, a relationship model between depth and component deviation is established based on the depth analysis interval. The component deviation value is calculated based on the relationship model, and the component deviation value is smoothed to obtain the depth deviation distribution, including:

[0086] Based on the depth analysis interval, a high-density grid point set is generated, and the corresponding depth coordinates and component concentration measurements are extracted.

[0087] By fitting the depth coordinates with the measured component concentration values, a relationship function between depth and component deviation is constructed.

[0088] Based on the relationship function, the deviation values ​​of each point in the high-density grid point set are calculated to obtain the initial deviation data sequence;

[0089] The initial deviation data sequence is smoothed to obtain a refined depth deviation distribution.

[0090] It should be noted that the generation of a high-density grid point set based on the depth analysis interval is specifically achieved by the system dividing the depth interval into dense grids in both the horizontal and vertical directions according to the pre-determined upper and lower boundaries and a preset physical resolution. For example, the grids are divided into 0.1-meter intervals in the depth and 0.2-meter × 0.2-meter grids in the horizontal plane, thereby generating a three-dimensional coordinate set containing thousands of regularly distributed points in the target space.

[0091] When extracting the corresponding depth coordinates and component concentration measurements, the following steps are performed: For each point in the high-density grid set, the system first uses its horizontal coordinates (X, Y) in its three-dimensional coordinates to quickly index and locate the corresponding data column or slice in the established smooth density distribution data field (usually a two-dimensional matrix or three-dimensional volume data). Then, based on its depth coordinates (Z), the system calculates the density value at that precise depth in the column or slice through linear interpolation. Subsequently, the system calls a density-concentration conversion model pre-established through laboratory calibration. This model defines the mathematical relationship (e.g., linear or polynomial relationship) between the density of a specific material and the content of its key components (such as ash and volatile matter). By inputting the density value read or calculated in the previous step into this model, the corresponding component concentration measurement value can be output in real time. Finally, the system pairs and stores the original depth coordinates of each grid point with its calculated component concentration value, forming a data pair set for subsequent analysis.

[0092] The process of fitting the depth coordinates and component concentration measurements to construct a relationship function between depth and component deviation is performed using a polynomial interpolation algorithm. The system first organizes all data points into two columns: one column contains the depth coordinates of all grid points, and the other column contains the corresponding component concentration measurements. Next, the system pre-determines a cubic polynomial as the basic form of the fitting function. Then, the system applies the least squares method: its goal is to find a specific set of polynomial coefficients that minimizes the overall error between the predicted concentration value calculated from the depth coordinates using this polynomial and the actual measured concentration values ​​at all points. This "overall error" is quantified as the sum of squares of the differences between the predicted and measured values ​​at all data points. The system constructs and solves a system of linear equations derived from this minimization condition to calculate the unique and optimal set of polynomial coefficients that minimizes the sum of squared errors. Substituting the optimal coefficients obtained from the solution into the preset cubic polynomial form yields a specific complete mathematical function that takes depth as input and the expected component concentration as output. This function is the constructed relationship function between depth and the expected value of component concentration, which depicts the overall trend line of component concentration variation with depth throughout the entire analytical depth range.

[0093] The system calculates the deviation values ​​of each point in the high-density grid point set according to the relationship function. First, the system sequentially traverses each point in the high-density grid point set. For the point currently being processed, the system reads its depth coordinate value and uses it as input, substituting it into the depth-component concentration relationship function constructed in the previous step. By executing the calculation logic of this function, the "expected" component concentration value based on the overall trend at this depth is obtained. Next, the system reads the actual component concentration measurement value of this point obtained through the density-concentration conversion model. Then, the system calculates the difference between these two values, that is, subtracting the expected value from the actual measurement value and taking the absolute value of the difference. The result is the local component deviation value of this grid point relative to the overall concentration change trend. The system associates and stores this calculated deviation value with the spatial coordinates of the point. After completing the traversal and calculation of all points in the set, the system extracts the deviation values ​​of all points and rearranges them in ascending order according to their corresponding depth coordinates, finally forming a continuous one-dimensional data sequence. This sequence is the initial deviation data sequence reflecting the degree of local deviation at each location.

[0094] The initial deviation data sequence is smoothed to obtain a refined depth deviation distribution. Using a moving average algorithm, the system first determines the range of neighboring points needed to calculate the average value for each point based on the depth coordinates of the data points and a preset physical window radius. Then, starting from the first point in the deviation data sequence, the system searches for all neighboring deviation data points within the preset depth window, centered on the current point, and calculates the arithmetic mean of the deviation values ​​of these points. Next, the system replaces the original deviation value of the current center point with the calculated average value. After replacing the current point, the window slides forward one data point, repeating the steps of finding neighboring points and calculating the average value until all points in the sequence have been processed. For data points where the ends of the sequence cannot form a complete window, the system uses boundary copying or mirror filling to ensure the continuity of the sequence. This process effectively suppresses high-frequency noise caused by random measurement errors or small local fluctuations, while preserving the macroscopic trend of deviation changes with depth, ultimately outputting a continuous, smoothly transitioning curve—the refined depth deviation distribution.

[0095] In step S16, the deviation gradient is calculated based on the depth deviation distribution. A set of candidate depth points is selected based on the deviation gradient. From the set of candidate depth points, the optimal depth point is determined with the minimization of component deviation as the guiding principle. The final sampling depth value is determined by parsing the graphical depth information of the optimal depth point, including:

[0096] Based on the depth deviation distribution, a deviation gradient sequence is calculated, and the minimum gradient value and the corresponding depth position in the deviation gradient sequence are identified as a set of candidate depth points.

[0097] The optimal depth point is selected from the candidate depth point set, a depth information distribution map of the optimal depth point is generated, and the central clustering degree of the depth information distribution map is analyzed.

[0098] If the central clustering degree is greater than the preset clustering degree threshold, the depth value corresponding to the optimal depth point is extracted to determine the final sampling depth value.

[0099] It should be noted that the calculation of the deviation gradient sequence based on the depth deviation distribution is achieved by the system performing point-by-point difference calculation on the obtained refined deviation data in depth order. Specifically, starting from the second data point in the sequence, the deviation value of the current depth point is subtracted from the deviation value of the previous adjacent depth point, and then divided by the depth interval between the two points to obtain a gradient value that reflects the rate of change of the deviation. By traversing the entire deviation sequence in this way, a continuous deviation gradient sequence can be generated.

[0100] Based on the minimum gradient value and its corresponding depth location identified in the deviation gradient sequence, a set of candidate depth points is established. The system iterates through each data point in the deviation gradient sequence, skipping the first and last points. For the i-th point currently being processed, the system first reads its gradient value. And calculate its absolute value. Next, this absolute value is compared with a preset gradient threshold. Simultaneously, the system performs local minima detection: it reads the gradient value g(i-1) of the previous point and the gradient value of the next point. Then check if the conditions are met. and If the absolute value of the gradient at the current point is greater than the threshold and simultaneously satisfies the conditions for a local minimum, the system determines that the point is a valid gradient minimum point. Subsequently, the system maps the point back to its corresponding original depth coordinates based on its index in the gradient sequence and records this depth value in the candidate list. After traversing the entire gradient sequence, all the recorded depth points constitute the set of candidate depth points for subsequent filtering.

[0101] The optimal depth point is selected from the set of candidate depth points by the system calculating a comprehensive score for each candidate point in the set. The scoring rules take into account three factors: the absolute value of the gradient of the candidate point (with the highest weight), the distribution density of other candidate points within a certain depth range centered on the candidate point, and the closeness of the depth of the candidate point to the depth of the maximum deviation of the whole analysis interval. The system calculates the standardized score of each candidate point on these three factors and sums them by weight. Finally, the candidate point with the highest comprehensive score is selected and determined as the optimal depth point in the current analysis interval.

[0102] The scoring calculation process is as follows: First, the three factors are normalized to eliminate the influence of dimensions. For the absolute value of the gradient, the maximum and minimum values ​​among all candidate points are used as the range, and each point value is linearly mapped to between 0 and 1. For the distribution density of neighboring candidate points, the number of other candidate points within a fixed depth range centered on the current point is counted, and the maximum and minimum values ​​are also normalized. For the proximity to the depth of the maximum deviation of the overall depth, the absolute difference between the depth of each candidate point and the depth of the maximum deviation of the overall depth is calculated, and then the difference is normalized (the smaller the difference, the higher the proximity score). Next, weights are assigned to the three normalized factors. For example, the weight of the absolute value of the gradient is set to 0.5, the weight of the neighboring point density is set to 0.25, and the weight of the proximity is set to 0.25. This weight allocation scheme is determined based on feature importance analysis. The specific principles are as follows: the absolute value of the gradient directly measures the drastic change in local component deviation and is the most direct indicator of representativeness, therefore it is given the highest weight of 0.5; the density of neighboring points reflects the clustering of candidate points in the depth direction, and high clustering means that the deviation pattern is stable near that depth, so it is given a medium weight of 0.25; the proximity to the depth of the overall maximum deviation reflects the focus on the most anomalous region, and is also given a medium weight of 0.25. Finally, the comprehensive score of each candidate point is obtained by multiplying the three normalized scores by their corresponding weights and then summing them. The system selects the candidate point with the highest comprehensive score as the optimal depth point.

[0103] The system generates a depth information distribution map of the optimal depth point and analyzes the central clustering degree of the depth information distribution map to obtain the central clustering degree. First, the system takes the optimal depth point as the center and expands it upwards and downwards by a preset depth range (e.g., 0.1 meters). It extracts the horizontal coordinates of all high-density grid points within this range and their corresponding deviation values. Then, it generates a two-dimensional scatter plot. This plot uses depth as the vertical axis (fixed to a narrow interval near the optimal depth) and horizontal position as the horizontal axis. The color or size of each scatter point represents its deviation value. This plot is the depth information distribution map.

[0104] The analysis of the central clustering degree uses a kernel density estimation algorithm. The system first collects the horizontal coordinates of all scattered points in the depth information distribution map. Next, the system uses a kernel density estimation algorithm to estimate the probability density distribution of these points on the horizontal plane. When the algorithm is executed, it first needs to determine a bandwidth parameter, which is usually automatically calculated based on the standard deviation of the horizontal coordinates of all scattered points using empirical rules (such as Scott's rule); then, the system creates a fine two-dimensional grid within the horizontal plane where the scattered points are located. For each grid point, the algorithm calculates the Euclidean distance between that point and all scattered points. Based on these distances, it uses a kernel function (such as a Gaussian kernel) to calculate the "contribution" of each scattered point to that grid point. The sum of the contributions of all scattered points yields the kernel density estimate for that grid point. After traversing all grid points, a two-dimensional density distribution surface is generated. The system searches for the region with the highest density value on this density surface, typically by identifying local peak points of density values ​​and defining a continuous region around these peak points with a density value no less than a certain percentage (e.g., 80%) of the peak value. This region is the highest density region. Finally, the system counts how many of the original scattered points have their horizontal coordinates falling within this highest density region and divides this number by the total number of scattered points. The resulting ratio is the centrality of clustering.

[0105] If the central clustering degree is greater than a preset clustering degree threshold, the depth value corresponding to the optimal depth point is extracted to determine the final sampling depth value. The specific implementation of the extraction and determination process is as follows: After verifying that the central clustering degree meets the standard, the system immediately accesses the data structure storing the "optimal depth point" object and reads its core attribute—the depth coordinate value. Next, the system formats this raw depth value, for example, by uniformly converting it to "meters" as the unit, and rounding it according to the precision required by the sampling device's communication protocol (e.g., retaining three decimal places). Then, the system encapsulates this formatted depth value into an immutable instruction parameter, writes it into the "final sampling depth value" dedicated register, and sets the "depth lock" flag.

[0106] The preset clustering threshold is set based on the analysis results of a large amount of historical sampling data. Specifically, the method for determining the preset clustering threshold is as follows: During the system's offline training phase, M sets of historical operating condition data are collected (usually M>200). Each set of data includes the optimal depth point calculated using this method and its depth information distribution map. Through external authoritative testing methods, the actual sampling representativeness corresponding to each set of data is binarized and labeled (good representativeness is recorded as 1, poor representativeness as 0).

[0107] Next, the set of central clustering values ​​Sg for all highly representative samples and the set of central clustering values ​​Sb for all poorly representative samples are calculated. The distribution of Sg is analyzed, and its 5th percentile P5 is taken. At the same time, to maximize the discriminative power, the threshold To is calculated to maximize the sum of the proportion of samples in Sg that are above the threshold and the proportion of samples in Sb that are below the threshold. The final preset clustering threshold is calculated as: Tf = max(0.70, min(P5, To)), where 0.70 is the empirical lower limit.

[0108] In step S17, based on the final sampling depth value, the position deviation is calculated. Using the position deviation, the position of the sampling device is adjusted via feedback control. The adjusted device position is then matched with a preset depth partition to obtain material sample data for the corresponding depth. Based on the material sample data, representative sample information is generated, including:

[0109] If the final sampling depth value is located in the middle region of the warehouse, calculate the positional deviation between the final sampling depth value and the boundary.

[0110] Based on the positional deviation, a control command is generated to drive the sampling device to move to the target position and record the device's movement trajectory;

[0111] The device's movement trajectory is matched with a preset depth partition to determine the target depth range, material sample data within the target depth range is extracted, and representative sample information is generated.

[0112] It should be noted that if the final sampling depth value is located in the central region of the silo, the specific implementation of calculating the positional deviation between the final sampling depth value and the boundary is as follows: the system first calls the pre-stored three-dimensional geometric model of the silo, reads the key coordinate data of the silo wall boundary, and obtains the determined final sampling depth value and its corresponding horizontal plane coordinates. Then, the system calculates the Euclidean distance from the horizontal coordinate of the sampling point to the nearest silo wall boundary (in a cylindrical silo, the radial distance to the central axis of the silo). This distance value is defined as the positional deviation, which is used to quantify the degree of offset of the sampling point relative to the ideal central region.

[0113] Based on the position deviation, a control command is generated to drive the sampling device to move to the target position and record the device's movement trajectory. This process adopts a hierarchical response control strategy. The system compares the calculated position deviation with multiple preset threshold ranges. If the deviation is less than 0.8 meters, a "direct descent" command is generated, driving the device to descend directly along the current vertical line. If the deviation is between 0.8 meters and 1.5 meters, a compensation mode is triggered. The system generates a control sequence containing multiple sub-commands. For example, the first command drives the lateral movement mechanism of the sampling device to translate a specific compensation distance (such as half of the deviation) towards the center of the chamber. The second command performs attitude fine-tuning and position confirmation. The third command locks the new vertical path and starts the descent. During the execution of the command, the device's built-in displacement sensor collects its three-dimensional spatial coordinates in real time and records them in time stamp order, forming a complete spatial movement trajectory from the starting point through horizontal compensation to vertical descent.

[0114] The specific setting process for the threshold range comprehensively considers equipment performance, accuracy requirements, and safety redundancy. First, it is based on the mechanical parameters of the sampling device, particularly the maximum stroke, positioning accuracy, and response time of the lateral movement mechanism. This determines the upper limit of the deviation applicable to the "direct descent" mode (e.g., 0.8 meters). This upper limit ensures that the device's natural hanging position falls within the allowable position error circle for representative sampling when not moving laterally. Second, based on the hopper radius and the device's maximum effective operating range, an upper limit of the deviation interval requiring the "compensation" mode is set (e.g., 1.5 meters). This value is typically less than the device's maximum theoretical stroke to provide a safety margin for positioning and attitude adjustment. All thresholds are finalized in the control parameter table after extensive positioning tests under actual no-load and simulated load conditions during the system integration phase, based on repeated calibration and optimization of the device's repeatability, trajectory smoothness, and final positioning success rate.

[0115] The system matches the device's movement trajectory with preset depth zones to determine the target depth range. Material sample data within the target depth range is extracted, and representative sample information is generated. The core of this process involves the system comparing the recorded depth coordinate sequence of the device's movement trajectory with pre-defined depth zones based on material settling characteristics (e.g., upper dispersed zone 0-2 meters, middle dense segregation zone 2-6 meters, lower compacted zone 6 meters to the bottom of the silo). By determining which preset zone the depth range covered by the main body of the trajectory falls into, the target depth range for this sampling is determined. For example, if the final sampling depth is 4.85 meters and the device trajectory... If the depth mainly falls between 2 and 6 meters, it matches the "central dense segregation zone". Then, the system controls the sampling device to perform sampling operations within the target range. Usually, a multi-point collaborative strategy is adopted, with the main depth point (4.85 meters) as the core and auxiliary sampling at the adjacent depths above and below (such as 4.70 meters and 5.00 meters) to obtain a sample combination on a depth gradient. Finally, the system integrates and analyzes the multiple raw sample data collected, calculates the average concentration and statistical distribution characteristics of its key components (such as ash), and attaches metadata such as sampling location and depth range, and packages them to generate a complete representative sample information report.

[0116] Reference Figure 2 The second embodiment of the present invention provides an automated sampling control system for a quantitative weighing bin, comprising:

[0117] The data acquisition module is used to acquire an initial signal set, extract density distribution data based on the initial signal set, and construct an initial density distribution map;

[0118] The density optimization module is used to sample the initial density distribution map to obtain a density value sequence, perform anomaly detection and data transformation on the density value sequence to generate a two-dimensional density distribution map, and obtain a smooth density distribution by performing smoothing calculation on the two-dimensional density distribution map.

[0119] The feature extraction module is used to analyze the density differences in the smooth density distribution, identify component segregation regions, perform spatial mapping on the identified segregation regions, generate feature maps, and extract the distribution patterns of the segregation regions based on the feature maps.

[0120] The interval positioning module is used to determine the target area with a density difference exceeding a preset threshold from the segregated area according to the distribution pattern, extract key coordinate points from the target area and determine the initial depth range, and filter the initial depth range to determine the depth analysis interval.

[0121] The distribution refinement module is used to establish a relationship model between depth and component deviation based on the depth analysis interval, calculate the component deviation value based on the relationship model, and smooth the component deviation value to obtain the depth deviation distribution.

[0122] The depth decision module is used to calculate the deviation gradient based on the depth deviation distribution, select a set of candidate depth points based on the deviation gradient, determine the optimal depth point from the set of candidate depth points with the minimization of component deviation as the guide, and determine the final sampling depth value by parsing the graphical depth information of the optimal depth point.

[0123] The sampling execution module is used to calculate the position deviation based on the final sampling depth value, and adjust the position of the sampling device through feedback control based on the position deviation. The adjusted device position is matched with the preset depth partition to obtain the material sample data at the corresponding depth, and representative sample information is generated based on the material sample data.

[0124] It should be noted that the automated sampling control system for a quantitative weighing bin provided in this embodiment of the invention is used to execute all the process steps of the automated sampling control method for a quantitative weighing bin in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0125] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0126] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. An automated sampling control method for a quantitative weighing bin, characterized in that, include Obtain an initial signal set, extract density distribution data based on the initial signal set, and construct an initial density distribution map; Sampling is performed on the initial density distribution map to obtain a density value sequence. Anomaly detection and data transformation are performed on the density value sequence to generate a two-dimensional density distribution map. A smooth density distribution is obtained by smoothing the two-dimensional density distribution map. Analyze the density differences in the smooth density distribution, identify component segregation regions, perform spatial mapping on the identified segregation regions to generate feature maps, and extract the distribution patterns of the segregation regions based on the feature maps; Based on the distribution pattern, target areas with density differences exceeding a preset threshold are identified from the segregated regions. Key coordinate points are extracted from the target areas, and an initial depth range is determined. The initial depth range is then filtered to determine the depth analysis interval. Based on the depth analysis interval, a relationship model between depth and component deviation is established. The component deviation value is calculated based on the relationship model, and the component deviation value is smoothed to obtain the depth deviation distribution. The deviation gradient is calculated based on the depth deviation distribution. A set of candidate depth points is selected based on the deviation gradient. The optimal depth point is determined from the set of candidate depth points with the minimization of component deviation as the guide. The final sampling depth value is determined by parsing the graphical depth information of the optimal depth point. Based on the final sampling depth value, the position deviation is calculated. Combined with the position deviation, the position of the sampling device is adjusted through feedback control. The adjusted device position is matched with the preset depth partition to obtain the material sample data at the corresponding depth. Based on the material sample data, representative sample information is generated.

2. The automated sampling control method for the quantitative weighing bin according to claim 1, characterized in that, The step of obtaining an initial signal set, extracting density distribution data based on the initial signal set, and constructing an initial density distribution map includes: An initial signal set is obtained, and time-frequency conversion and feature extraction are performed on the initial signal set to obtain a frequency domain signal sequence; If the high-frequency component of the frequency domain signal sequence is greater than a preset frequency threshold, the high-frequency component is filtered to obtain a denoised signal sequence. Based on the denoised signal sequence, density distribution data is generated by mapping, and interpolation calculation is performed on the density distribution data to generate an initial density distribution map.

3. The automated sampling control method for the quantitative weighing bin according to claim 1, characterized in that, The process involves sampling from the initial density distribution map to obtain a density value sequence, performing anomaly detection and data transformation on the density value sequence to generate a two-dimensional density distribution map, and then performing smoothing calculations on the two-dimensional density distribution map to obtain a smoothed density distribution, including: The initial density distribution map is sampled to obtain a density value sequence, and anomaly detection is performed on the density value sequence to filter out a valid data set; The effective data set is transformed into a coordinate matrix and mapped to a two-dimensional data matrix. Based on the two-dimensional data matrix, a two-dimensional density distribution map is reconstructed and generated. The two-dimensional density distribution map is smoothed to obtain a smoothed density distribution.

4. The automated sampling control method for the quantitative weighing bin according to claim 1, characterized in that, The analysis of density differences in the smooth density distribution, identification of component segregation regions, spatial mapping of the identified segregation regions to generate feature maps, and extraction of the distribution patterns of the segregation regions based on the feature maps include: Calculate the density difference sequence of the smooth density distribution, and identify component segregation regions based on the density difference sequence; Data spatial mapping is performed on the component segregation regions to generate a segregation distribution feature map; Cluster analysis is performed on the segregation distribution feature map to extract and analyze the geometric center trajectory of the clusters, thereby obtaining the distribution pattern of the segregation region.

5. The automated sampling control method for the quantitative weighing bin according to claim 1, characterized in that, The process of determining target regions with density differences exceeding a preset threshold from the segregated regions based on the distribution pattern, extracting key coordinate points from the target regions and determining an initial depth range, and then filtering the initial depth range to determine a depth analysis interval includes: Based on the distribution pattern, density difference values ​​are extracted from the segregated regions. If the density difference value exceeds a preset difference threshold, the corresponding target segregated region is locked. Extract the boundary pixels of the target segregated region and obtain the key coordinate points through spatial mapping; Based on the key coordinate points, the corresponding initial depth range is calculated, and the depth range information is subjected to extreme value filtering to determine the preliminary depth analysis interval.

6. The automated sampling control method for the quantitative weighing bin according to claim 1, characterized in that, The process of establishing a relationship model between depth and component deviation based on the depth analysis interval, calculating component deviation values ​​based on the relationship model, and smoothing the component deviation values ​​to obtain the depth deviation distribution includes: Based on the depth analysis interval, a high-density grid point set is generated, and the corresponding depth coordinates and component concentration measurements are extracted. By fitting the depth coordinates with the measured component concentration values, a relationship function between depth and component deviation is constructed. Based on the relationship function, the deviation values ​​of each point in the high-density grid point set are calculated to obtain the initial deviation data sequence; The initial deviation data sequence is smoothed to obtain a refined depth deviation distribution.

7. The automated sampling control method for a quantitative weighing bin according to claim 1, characterized in that, The process of calculating the deviation gradient based on the depth deviation distribution, selecting a set of candidate depth points based on the deviation gradient, determining the optimal depth point from the set of candidate depth points with the goal of minimizing component deviation, and determining the final sampling depth value by parsing the graphical depth information of the optimal depth point includes: Based on the depth deviation distribution, a deviation gradient sequence is calculated, and the minimum gradient value and the corresponding depth position in the deviation gradient sequence are identified as a set of candidate depth points. The optimal depth point is selected from the candidate depth point set, a depth information distribution map of the optimal depth point is generated, and the central clustering degree of the depth information distribution map is analyzed. If the central clustering degree is greater than the preset clustering degree threshold, the depth value corresponding to the optimal depth point is extracted to determine the final sampling depth value.

8. The automated sampling control method for a quantitative weighing bin according to claim 1, characterized in that, The step involves calculating the position deviation based on the final sampling depth value, adjusting the position of the sampling device using feedback control based on the position deviation, matching the adjusted device position with a preset depth partition to obtain material sample data at the corresponding depth, and generating representative sample information based on the material sample data, including: If the final sampling depth value is located in the middle region of the warehouse, calculate the positional deviation between the final sampling depth value and the boundary. Based on the positional deviation, a control command is generated to drive the sampling device to move to the target position and record the device's movement trajectory; The device's movement trajectory is matched with a preset depth partition to determine the target depth range, material sample data within the target depth range is extracted, and representative sample information is generated.

9. An automated sampling control system for a quantitative weighing bin, characterized in that, include: The data acquisition module is used to acquire an initial signal set, extract density distribution data based on the initial signal set, and construct an initial density distribution map; The density optimization module is used to sample the initial density distribution map to obtain a density value sequence, perform anomaly detection and data transformation on the density value sequence to generate a two-dimensional density distribution map, and obtain a smooth density distribution by performing smoothing calculation on the two-dimensional density distribution map. The feature extraction module is used to analyze the density differences in the smooth density distribution, identify component segregation regions, perform spatial mapping on the identified segregation regions, generate feature maps, and extract the distribution patterns of the segregation regions based on the feature maps. The interval positioning module is used to determine the target area with a density difference exceeding a preset threshold from the segregated area according to the distribution pattern, extract key coordinate points from the target area and determine the initial depth range, and filter the initial depth range to determine the depth analysis interval. The distribution refinement module is used to establish a relationship model between depth and component deviation based on the depth analysis interval, calculate the component deviation value based on the relationship model, and smooth the component deviation value to obtain the depth deviation distribution. The depth decision module is used to calculate the deviation gradient based on the depth deviation distribution, select a set of candidate depth points based on the deviation gradient, determine the optimal depth point from the set of candidate depth points with the minimization of component deviation as the guide, and determine the final sampling depth value by parsing the graphical depth information of the optimal depth point. The sampling execution module is used to calculate the position deviation based on the final sampling depth value, and adjust the position of the sampling device through feedback control based on the position deviation. The adjusted device position is matched with the preset depth partition to obtain the material sample data at the corresponding depth, and representative sample information is generated based on the material sample data.

Images