A vehicle automatic sampling system for industrial raw materials

Abstract

The present invention discloses a vehicle automatic sampling system for industrial raw materials, comprising: a stopping module, used for stopping the vehicle when the vehicle loaded with industrial raw materials enters the sampling area; a laser radar scanning module, used for scanning the vehicle with a laser radar to generate a three-dimensional coordinate point cloud of the carriage; a sampling target point generation module, used for randomly generating a number of preset target points in the three-dimensional coordinate point cloud of the carriage to construct a preset target point set; a correction reference coordinate point generation module, used for inputting the preset target point set into a preset deviation optimization model to generate corresponding correction reference coordinate points; an automatic sampling module, used for controlling the sampler drill rod point by point according to the correction reference coordinate points to go to the preset target point for sampling; the present invention significantly reduces the deviation between the actual execution point and the preset target point, thereby improving the execution accuracy of the automatic sampling system.

Description

Technical Field

[0001] The invention relates to a sampling system, in particular to an automatic vehicle sampling system for industrial raw materials. Background Art

[0002] When steel companies purchase industrial raw materials such as coal and iron ore powder, these raw materials usually need to be transported to the factory by car. In order to ensure the accuracy of production quality and raw material parameters, the laboratory needs to sample the raw materials when the car enters the factory to analyze key parameters such as moisture and ash. Sampling is usually completed by a sampler, and the initial car sampling is done manually by entering the vehicle. This method has the following disadvantages due to its over-reliance on manual operation: Human interference: The sampling results are easily affected by the operator's experience and behavior, lacking consistency and reliability; Poor working environment: Manual entry into the car compartment for sampling poses safety hazards, high labor intensity, and low work efficiency.

[0003] The patent document with patent publication number CN110687071A discloses a feed raw material automatic sampling and identification system and method based on machine vision, which improves the detection efficiency through remote detection module and data storage module. However, in the actual sampling process, in order to improve sampling efficiency and safety, the existing technology gradually develops towards unattended automation. With the development of automation technology, the first generation of samplers came into being. The sampler is a single-point drill rod sampling device. The drill rod is fixed and cannot be moved. Sampling can only be completed by manually commanding the vehicle to move to different positions. The second generation sampler has achieved full automation. The drill rod can move horizontally and vertically above the carriage, and the sampling range covers different areas of the carriage. This type of equipment monitors the carriage through a top-mounted camera, uses video images to delineate the carriage area, and randomly generates sampling points in the carriage, avoiding most of the human factors in the first generation of equipment. However, the second generation sampler will still cause execution deviations due to external interference. Specifically: vibration interference: due to the shaking of the vehicle when it is parked, the execution path of the drill rod may deviate, causing the actual execution point to deviate from the preset target point; uneven material distribution: the height difference of the piled raw materials may affect the force of the drill rod during the descent process, further exacerbating the deviation; environmental fluctuations: small interference in the external environment (such as wind or resonance of mechanical devices) may cause instability in the sampling path. Therefore, the defect in the prior art is the lack of an optimization mechanism for dynamic deviations in the sampling execution process. Summary of the invention

[0004] In view of the deficiencies of the prior art, the present invention provides a vehicle automatic sampling system for industrial raw materials, which solves the technical problems raised in the background technology through dynamic sampling deviation optimization.

[0005] To achieve the above objectives, the present invention discloses the following technical solutions:

[0006] A vehicle automatic sampling system for industrial raw materials, the sampling system comprising:

[0007] A stopping module is used to stop a vehicle loaded with industrial raw materials when the vehicle enters the sampling area; illustratively, the stopping module sets a gate at the entrance of the sampling area;

[0008] A laser radar scanning module, used to scan the vehicle using a laser radar to generate a three-dimensional coordinate point cloud of the vehicle compartment;

[0009] A sampling target point generation module is used to randomly generate a number of preset target points in the three-dimensional coordinate point cloud of the carriage to construct a preset target point set;

[0010] A correction reference coordinate point generation module is used to input a preset target point set into a preset deviation optimization model to generate corresponding correction reference coordinate points;

[0011] The correction reference coordinate point is a correction intermediate point generated point by point by the deviation optimization model according to the coordinates of the preset target point, and is used to correct the execution path of the sampling machine drill rod;

[0012] The automatic sampling module is used to control the sampling machine drill rod point by point according to the calibrated reference coordinate points to go to the preset target point for sampling.

[0013] In some embodiments, the modeling step of the deviation optimization model includes:

[0014] S1. Obtain a training set containing N sampling optimization samples;

[0015] S2. Input a training set containing N sample optimization samples into a multi-layer perception model. The multi-layer perception model extracts a number of sample optimization samples in batches for forward propagation to output a single-point prediction deviation vector. After iterative training, a matching degree optimization model is obtained.

[0016] S3, inputting the training set containing N sample optimization samples into the matching degree optimization model to generate the optimal prediction deviation vector of the first actual execution point;

[0017] S4, adding the optimal prediction deviation vector of the first actual execution point and its coordinates dimension by dimension to generate a correction reference coordinate point for the sampling machine drill rod to move to the preset target point;

[0018] The expression of the correction reference coordinate point is: ;

[0019] represents the correction reference coordinate point of the kth point, represents the actual execution point of the kth point, represents the optimal prediction deviation vector of the kth point;

[0020] S5, obtaining the execution loss of the sampler drill pipe in the actual sampling operation;

[0021] S6. Minimize the total execution loss of N sampled optimization samples, and obtain the deviation optimization model after iterative training.

[0022] In some embodiments, obtaining a training set including N sampling optimization samples includes:

[0023] S1-1. Use laser radar to obtain the three-dimensional coordinate point cloud of the carriage;

[0024] The expression of the three-dimensional coordinate point cloud is: ;

[0025] C is the three-dimensional coordinate point cloud, are the X-axis coordinate, Y-axis coordinate and Z-axis coordinate of the i-th point respectively, and M is the total number of three-dimensional point cloud coordinates;

[0026] S1-2, selecting a number of preset target points in the three-dimensional coordinate point cloud, and aggregating the plurality of preset target points into a preset target point set;

[0027] The expression of the preset target point set is: ;

[0028] is the preset target point set, which represents all randomly selected preset target points; are the X-axis coordinate, Y-axis coordinate and Z-axis coordinate of the k-th preset target point, respectively, and N is the total number of preset target points;

[0029] S1-3, obtaining several actual execution points of the sampling machine drill rod, and summarizing them to obtain an actual execution point set;

[0030] The expression of the actual execution point set is: ;

[0031] is the actual execution point set, indicating the sampling point actually reached by the sampling machine drill rod; are the X-axis coordinate, Y-axis coordinate and Z-axis coordinate of the kth actual execution point, respectively; N is the total number of actual execution points, which is equal to the total number of preset target points;

[0032] S1-4, calculating the true deviation vector of each point between the preset target point set and the actual execution point set point by point;

[0033] The expression of the true deviation vector is:

[0034] ;

[0035] is the kth true deviation vector, including the coordinate deviations in the X-axis, Y-axis and Z-axis directions; is the coordinate deviation component of the kth true deviation vector on the X-axis, Y-axis and Z-axis;

[0036] S1-5, summarizing each point-by-point calculated true deviation vector into a true deviation vector set;

[0037] The expression of the true deviation vector set is: represents the set of true deviation vectors;

[0038] S1-6, generating continuous sampling numbers based on the sampling order of the actual execution points, and standardizing the sampling numbers to obtain sampling number features; wherein the sampling numbers are generated based on the X-axis coordinate values ​​of the actual execution points sorted from small to large;

[0039] S1-7, performing vector concatenation of each preset target point, sampling sequence number feature and the global feature of the 3D point cloud after dimensionality reduction to generate a sampling optimization input vector;

[0040] The expression of the sampling optimization input vector is:

[0041] ;

[0042] represents the input features of the kth sampling optimization sample, Represents the global features of the 3D point cloud after dimensionality reduction, including n eigenvalues; Represents the kth sampling sequence feature; This means that for every k, there is ;

[0043] S1-8, using the sampling optimization input vector as the input feature and the true deviation vector as the target label to construct the sampling optimization sample;

[0044] The expression of the sampling optimization sample is: ; D represents the training set containing N sampling optimization samples.

[0045] In some of the embodiments, a matching degree optimization model is obtained after iterative training, including:

[0046] S2-1, inputting the sampled optimized input vectors in the sampled optimized samples into the input layer one by one for feature encoding to obtain the encoded input vector; wherein the input layer is a fully connected layer;

[0047] S2-2, extracting features from the encoded input vector, and performing high-dimensional feature interaction through the hidden layer to obtain an input vector of the hidden layer;

[0048] S2-3, mapping the input vector of the hidden layer to the prediction deviation vector through the output layer;

[0049] S2-4, calculating the matching loss between the predicted deviation vector and the true deviation vector;

[0050] The matching loss function is: ;

[0051] in, is the matching loss value, which is used to measure the matching degree between the predicted deviation vector and the true deviation vector; N is the total number of sampling optimization samples, , , Optimize the sample prediction deviation vector for the kth sample and the coordinate deviation components on the X-axis, Y-axis and Z-axis;

[0052] S2-5, performing back propagation to update the model parameters of the multi-layer perception model to obtain updated model parameters;

[0053] S2-6, iteratively execute S2-1 and S2-5 until the minimum matching loss is obtained;

[0054] S2-7. The model that minimizes the matching loss is regarded as the matching optimization model.

[0055] In some embodiments, obtaining the execution loss of the sampling machine drill pipe in the actual sampling operation includes:

[0056] S5-1, control the sampling machine drill rod to move according to the generated calibration reference coordinate point and perform the sampling operation;

[0057] S5-2, record the coordinates of the actual execution point of the sampling machine drill rod;

[0058] S5-3, calculating the execution loss between the actual execution point and the correction reference coordinate point;

[0059] The execution loss is characterized by: the L2 norm of the actual execution coordinate point and the correction reference coordinate point;

[0060] The expression of the L2 norm is:

[0061] ;

[0062] in, Indicates the distance between the calibration reference coordinate point and the actual execution point. represents the L2 norm, represents the correction reference coordinate point of the kth sampling optimization sample; , , They represent the X-axis coordinate, Y-axis coordinate, and Z-axis coordinate of the actual execution point respectively. , , They represent the X-axis, Y-axis, and Z-axis coordinates of the calibration reference coordinate point, respectively.

[0063] In some embodiments, minimizing the total execution loss of N sample optimization samples and obtaining a deviation optimization model after iterative training includes:

[0064] S6-1, calculating the execution loss between each actual execution point and the correction reference coordinate point;

[0065] S6-2, summing the execution loss between each actual execution point and the correction reference coordinate point to obtain the total execution loss of N sample optimization samples;

[0066] The expression of the total execution loss is: ; represents the total execution loss;

[0067] S6-3. Calculate the gradient according to the total execution loss and update the model parameters of the matching optimization model;

[0068] S6-4. Iterate and execute S6-1 and S6-3 until the total execution loss is minimized.

[0069] In some of the embodiments, the vehicle is scanned using a laser radar to generate a three-dimensional coordinate point cloud of the vehicle compartment, and a number of preset target points are randomly generated within the three-dimensional coordinate point cloud of the vehicle compartment to construct a preset target point set, and also includes: using a laser radar to locate the spatial coordinates of the vehicle compartment in real time.

[0070] In some of the embodiments, the real-time positioning of the spatial coordinates of the vehicle compartment using a laser radar includes: using a point cloud segmentation algorithm to filter point cloud data other than the vehicle compartment to generate a three-dimensional coordinate point cloud of the vehicle compartment.

[0071] The present invention provides a vehicle automatic sampling system for industrial raw materials, which has the following beneficial effects:

[0072] The present invention introduces a matching loss function and takes the matching degree between the predicted deviation vector and the real deviation vector as the core optimization target of the system sampling deviation. In the iterative optimization process, based on minimizing the matching loss and dynamically updating the model parameters, the deviation distribution law can be gradually captured, and the dynamic error compensation during the execution of the sampler drill rod can be realized. The error accumulation problem caused by environmental fluctuations, mechanical vibrations and other factors in the traditional method is solved, so that the deviation between the actual execution point and the preset target point is significantly reduced, and the execution accuracy of the automatic sampling system is further improved.

[0073] Furthermore, the present invention generates correction reference coordinate points point by point through the deviation optimization model to provide the intermediate points of dynamic path correction. The correction reference coordinate points not only integrate the information of the preset target point and the predicted deviation vector, but also effectively avoid the deviation accumulation of the sampler drill rod caused by static planning or a single deviation correction scheme by adjusting the execution path of the guide drill rod in real time.

[0074] Furthermore, the present invention adopts a point-by-point optimization strategy, combines the deviation distribution law in the training phase and the real-time deviation feedback in the execution phase, and dynamically adjusts the sampling path of each step. By correcting the point-by-point guidance of the reference coordinate point, the deviation of the current point can be gradually corrected, and finally the global optimization of multiple points can be achieved to ensure that the sampling path always approaches the target trajectory. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 This is a structural block diagram of a vehicle automatic sampling system for industrial raw materials according to the present invention;

[0076] Figure 2 A sampling flow chart of a vehicle automatic sampling system for industrial raw materials according to the present invention;

[0077] Figure 3 A modeling flow chart of the deviation optimization model of the present invention;

[0078] Figure 4 This is a schematic diagram of the positions of the laser radar and the vehicle described in the present invention. DETAILED DESCRIPTION

[0079] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0080] First, the prior art and related concepts involved in the embodiments of the present invention are described:

[0081] PointNet structure: It is a deep learning model designed specifically for three-dimensional point cloud data. Its uniqueness lies in its ability to process unordered point cloud data directly without converting the point cloud data into voxel grids or depth maps.

[0082] Multilayer Perceptron Model: Multilayer Perceptron (MLP) is a feedforward neural network consisting of an input layer, a hidden layer, and an output layer. The neurons in each layer are connected to the previous layer through weights, and nonlinear characteristics are introduced through activation functions, so that it can handle complex nonlinear mapping problems.

[0083] LiDAR point cloud recognition: LiDAR point cloud recognition technology obtains the three-dimensional point cloud data of the target object through the LiDAR device. The point cloud data consists of the coordinates (X, Y, Z) of discrete points in space, describing the three-dimensional geometric structure of the object. Combined with the point cloud segmentation algorithm, it can extract the area of ​​interest from the complex scene and filter out irrelevant point cloud data.

[0084] Point cloud segmentation algorithm: The point cloud segmentation algorithm is used to further process the three-dimensional point cloud data of the car body collected by the LiDAR and separate the car body area from the background area. By screening the point set with higher density in the car body area, the three-dimensional point cloud range of the car body is generated. The result of point cloud segmentation is directly used to generate preset target points and as part of the input features of the deviation optimization model, helping the model to better adapt to actual application scenarios.

[0085] Example 1: Please refer to Figure 1 to Figure 2 The present invention provides a vehicle automatic sampling system for industrial raw materials, the sampling system comprising:

[0086] A stopping module, used to stop a vehicle loaded with industrial raw materials when it enters a sampling area;

[0087] A laser radar scanning module, used to scan the vehicle using a laser radar to generate a three-dimensional coordinate point cloud of the vehicle compartment;

[0088] A sampling target point generation module is used to randomly generate a number of preset target points in the three-dimensional coordinate point cloud of the carriage to construct a preset target point set;

[0089] A correction reference coordinate point generation module is used to input a preset target point set into a preset deviation optimization model to generate corresponding correction reference coordinate points;

[0090] The correction reference coordinate point is a correction intermediate point generated point by point by the deviation optimization model according to the coordinates of the preset target point, and is used to correct the execution path of the sampling machine drill rod;

[0091] The automatic sampling module is used to control the sampling machine drill rod point by point according to the calibrated reference coordinate points to go to the preset target point for sampling.

[0092] In this embodiment, the vehicle is scanned using a laser radar to generate a three-dimensional coordinate point cloud of the vehicle compartment, and a number of preset target points are randomly generated within the three-dimensional coordinate point cloud of the vehicle compartment to construct a preset target point set, and also includes: using a laser radar to locate the spatial coordinates of the vehicle compartment in real time.

[0093] Furthermore, the real-time positioning of the spatial coordinates of the carriage using a laser radar includes: using a point cloud segmentation algorithm to filter point cloud data outside the carriage to generate a three-dimensional coordinate point cloud of the carriage.

[0094] Specifically, because the purpose of the sampling machine drill rod is to find the carriage loaded with industrial raw materials, and the range of the carriage in the sampling area can be basically determined through exploration during the debugging period, the preset recognition area within this range can filter out tires or other parts that do not need to be recognized, and only recognize the area where the carriage is located. The recognition process uses the laser radar point cloud recognition technology in the prior art to filter out irrelevant positions such as tires and front of the car. The laser radar point cloud recognition technology collects three-dimensional point cloud data through the laser radar device, and combined with the point cloud segmentation algorithm, it can filter out irrelevant point cloud data (such as wheels, front of the car, etc.), and only retain the point cloud area of ​​the carriage.

[0095] It should be noted that, in this embodiment, the laser radar can be arranged above the sampling area, and it actually only needs to collect the two-dimensional coordinates of the horizontal plane of the industrial raw materials in the car. However, in order to avoid the influence of the concave and convex surface of the stacked industrial raw materials, collecting three-dimensional point clouds may be the best choice.

[0096] Example 2: Participation Figures 1 to 4 The technical solution of this embodiment 2 is different from that of embodiment 1 in that a modeling method of the deviation optimization model described in embodiment 1 is disclosed, and the modeling method includes:

[0097] S1. Obtain a training set containing N sampling optimization samples;

[0098] The S1 specifically includes:

[0099] S1-1. Use laser radar to obtain the three-dimensional coordinate point cloud of the carriage;

[0100] The expression of the three-dimensional coordinate point cloud is: ;

[0101] C is the three-dimensional coordinate point cloud, are the X-axis coordinate, Y-axis coordinate and Z-axis coordinate of the i-th point respectively, and M is the total number of three-dimensional point cloud coordinates;

[0102] S1-2, selecting a number of preset target points in the three-dimensional coordinate point cloud, and aggregating the plurality of preset target points into a preset target point set;

[0103] The expression of the preset target point set is: ;

[0104] is the preset target point set, which represents all randomly selected preset target points; are the X-axis coordinate, Y-axis coordinate and Z-axis coordinate of the k-th preset target point, respectively, and N is the total number of preset target points;

[0105] refer to Figure 4 , preset target points, that is, randomly generated sampling points. The rule for randomly generated sampling points is to directly randomly select N points according to a uniform distribution, or to perform pseudo-random sampling based on area division, or to perform random extraction based on some preset rules; for example, constrain the area range determined by experience, and select preset target points from the coordinate area where defective raw material samples are easily detected.

[0106] S1-3, obtaining several actual execution points of the sampling machine drill rod, and summarizing them to obtain an actual execution point set;

[0107] The expression of the actual execution point set is: ;

[0108] is the actual execution point set, indicating the sampling point actually reached by the sampling machine drill rod; are the X-axis coordinate, Y-axis coordinate and Z-axis coordinate of the kth actual execution point, respectively; N is the total number of actual execution points, which is equal to the total number of preset target points;

[0109] S1-4, calculating the true deviation vector of each point between the preset target point set and the actual execution point set point by point;

[0110] The expression of the true deviation vector is:

[0111] ;

[0112] is the kth true deviation vector, including the coordinate deviations in the X-axis, Y-axis and Z-axis directions; is the coordinate deviation component of the kth true deviation vector on the X-axis, Y-axis and Z-axis;

[0113] S1-5, summarizing each point-by-point calculated true deviation vector into a true deviation vector set;

[0114] The expression of the true deviation vector set is: represents the set of true deviation vectors;

[0115] S1-6, generating continuous sampling numbers based on the sampling order of the actual execution points, and standardizing the sampling numbers to obtain sampling number features; wherein the sampling numbers are generated based on the X-axis coordinate values ​​of the actual execution points sorted from small to large;

[0116] S1-7, performing vector concatenation of each preset target point, sampling sequence number feature and the global feature of the 3D point cloud after dimensionality reduction to generate a sampling optimization input vector;

[0117] The expression of the sampling optimization input vector is: ;

[0118] represents the input features of the kth sampling optimization sample, Represents the global features of the 3D point cloud after dimensionality reduction, including n eigenvalues; Represents the kth sampling sequence feature; This means that for every k, there is ;

[0119] Specifically, the global feature of the 3D point cloud after dimensionality reduction refers to extracting a global feature vector F that can represent the geometric structure and spatial information of the entire 3D coordinate point cloud from the 3D coordinate point cloud obtained by the lidar. ,in, ; The dimensionality reduction method can use the PointNet structure to extract the global features of the reduced 3D point cloud.

[0120] S1-8, using the sampling optimization input vector as the input feature and the true deviation vector as the target label to construct the sampling optimization sample;

[0121] The expression of the sampling optimization sample is: ; D represents the training set containing N sampling optimization samples.

[0122] The modeling approach also includes:

[0123] S2. Input the training set containing N sampled optimization samples into the multi-layer perception model. The multi-layer perception model extracts a number of sampled optimization samples in batches for forward propagation to output a single-point prediction deviation vector. After iterative training, a matching degree optimization model is obtained.

[0124] The S2 specifically includes:

[0125] S2-1, inputting the sampled optimized input vectors in the sampled optimized samples into the input layer one by one for feature encoding to obtain the encoded input vector; wherein the input layer is a fully connected layer;

[0126] Specifically, the encoded input vector refers to the original sampled optimized input vector converted into a representation vector in a high-dimensional space through the mapping process of the input layer. The mapping process includes linear transformation and nonlinear activation processing of input features (such as coordinate features, sampling sequence features, and 3D point cloud global features) to make them adaptable to feature extraction in subsequent hidden layers.

[0127] S2-2, extracting features from the encoded input vector, and performing high-dimensional feature interaction through the hidden layer to obtain an input vector of the hidden layer;

[0128] Specifically, feature extraction of the encoded input vector means that high-dimensional feature interactions are performed on the encoded input vector by stacking multiple layers of nonlinear mapping and parameter learning in the hidden layer, thereby extracting higher-order and more discriminative feature representations. The feature extraction method of the hidden layer uses a multi-layer perceptron structure, including matrix operations (such as the product of the weight matrix of the fully connected layer) and the application of activation functions (such as ReLU) to capture the nonlinear relationship of the input features and form the input vector of the hidden layer.

[0129] S2-3, mapping the input vector of the hidden layer to the prediction deviation vector through the output layer;

[0130] S2-4, calculating the matching loss between the predicted deviation vector and the true deviation vector;

[0131] The matching loss function is: ;

[0132] in, is the matching loss value, which is used to measure the matching degree between the predicted deviation vector and the true deviation vector; N is the total number of sampling optimization samples, , , Optimize the sample prediction deviation vector for the kth sample and the coordinate deviation components on the X-axis, Y-axis and Z-axis;

[0133] Specifically, the matching loss function minimizes the difference between the predicted deviation vector output by the model and the actual true deviation vector. By minimizing the loss function, the distribution law of the actual deviation of the sampler drill pipe can be gradually learned.

[0134] Furthermore, automatic sampling of industrial raw materials requires high-precision sampling, and the deviation of the sampler drill rod is usually affected by complex environmental factors such as mechanical equipment vibration deviation and execution accuracy deviation, resulting in the actual execution point often having a significant deviation from the preset target point, thereby forming a real deviation vector.

[0135] Therefore, in the process of automatic sampling of raw materials, the formation of real deviation is extremely difficult to avoid. The matching loss function can help the model capture the complex laws of the real deviation of the stroke. The set matching loss can make the sampler drill rod gradually approach the realization of "preset target point = actual execution point", that is, the equipment execution point accurately covers the position of the preset target point.

[0136] S2-5, performing back propagation to update the model parameters of the multi-layer perception model to obtain updated model parameters;

[0137] This step calculates the matching loss of the current batch of sample optimization samples through the back-propagation algorithm, and calculates the gradient of the matching loss with respect to the model parameters, and uses an optimization algorithm (such as gradient descent) on the gradient to update the model parameters.

[0138] S2-6, iteratively execute S2-1 and S2-5 until the minimum matching loss is obtained;

[0139] This step extracts batch samples from the training set in sequence and repeats steps S2-1 to S2-5. Each batch iteration executes steps S2-1 to S2-5 until the model converges and outputs a prediction deviation vector that minimizes the matching loss. Of course, minimizing the matching loss can be regarded as a convergence condition for determining the convergence of the model. To determine whether the convergence condition is reached, that is, to what extent the matching loss is minimized, a general convergence condition can be used, such as the maximum number of iterations, the upper limit of the training time, or the change in the matching loss is less than a predetermined threshold.

[0140] S2-7. The model that minimizes the matching loss is regarded as the matching optimization model.

[0141] In this embodiment, the model parameters when the matching loss is minimized during the training process can be selected and saved as the model parameters of the matching optimization model, that is, the multi-layer perception model when the matching loss is minimized is exported as the matching optimization model.

[0142] The modeling approach also includes:

[0143] S3. Input the training set containing N sample optimization samples into the matching degree optimization model to generate the optimal prediction deviation vector of the first actual execution point.

[0144] Specifically, the training set of the sampled optimization samples is input point by point into the matching optimization model. First, the first sampled optimization input vector (corresponding to the coordinates and other features of the first preset target point) is used as input, and the optimal prediction deviation vector of the first actual execution point is generated after forward propagation; subsequently, the remaining sampled optimization input vectors are input point by point, and a new optimal prediction deviation vector is generated in combination with the previous correction reference coordinate point until the calculation of the optimal prediction deviation vectors of all N points is completed.

[0145] The modeling approach also includes:

[0146] S4, adding the optimal prediction deviation vector of the first actual execution point and its coordinates dimension by dimension to generate a correction reference coordinate point for the sampling machine drill rod to move to the preset target point;

[0147] Furthermore, the correction reference coordinate point is a correction intermediate point generated by the deviation optimization model, and its function is to provide a reference for path correction for the sampler drill rod, and ultimately guide it to the preset target point.

[0148] The expression of the correction reference coordinate point is: ;

[0149] represents the correction reference coordinate point of the kth point, represents the actual execution point of the kth point, represents the optimal prediction deviation vector of the kth point;

[0150] In this embodiment, during the automatic sampling of industrial raw materials by vehicles, the actual sampling point of the drill rod of the sampler will always be affected by external factors (such as equipment accuracy, vehicle shaking or uneven material distribution), causing it to deviate from the preset target point. In order to correct this deviation, an optimal prediction deviation vector is used to provide sampling optimization guidance. Specifically, after the kth sampling, the coordinate vector of the actual execution point and the components of the optimal prediction deviation vector in each dimension are summed to obtain a correction reference coordinate point, which is regarded as a corrected three-dimensional coordinate containing comprehensive information of the current actual sampling point and the optimal prediction deviation vector; its goal is to provide a guiding direction for sampling that is closer to the preset target point, and to avoid the recurrence of the current deviation by gradually optimizing the deviation of each point.

[0151] To go further, "avoiding the recurrence of the current deviation" means that the "optimal prediction deviation vector" of each point is optimized based on the current actual execution point and the preset target point, and the optimization direction is to make the prediction of the optimal prediction deviation vector more accurate and ultimately point to the direction with the smallest deviation. Specifically, if the current true deviation (optimal prediction deviation vector) repeatedly points to the same direction, it means that the optimization process does not correct the sampling trend, but continuously accumulates deviations. In order to avoid this situation, the optimization process dynamically adjusts the prediction deviation.

[0152] The modeling approach also includes:

[0153] S5. Obtain the execution loss of the sampler drill pipe in the actual sampling operation.

[0154] The S5 specifically includes:

[0155] S5-1, control the sampling machine drill rod to move according to the generated calibration reference coordinate point and perform the sampling operation;

[0156] S5-2, record the coordinates of the actual execution point of the sampling machine drill rod;

[0157] S5-3, calculating the execution loss between the actual execution point and the correction reference coordinate point;

[0158] The execution loss is characterized by: the L2 norm of the actual execution coordinate point and the correction reference coordinate point;

[0159] The expression of the L2 norm is:

[0160] ;

[0161] in, Indicates the distance between the calibration reference coordinate point and the actual execution point. represents the L2 norm, represents the correction reference coordinate point of the kth sampling optimization sample; , , They represent the X-axis coordinate, Y-axis coordinate, and Z-axis coordinate of the actual execution point respectively. , , They represent the X-axis, Y-axis, and Z-axis coordinates of the calibration reference coordinate point, respectively.

[0162] The modeling approach also includes:

[0163] S6. Minimize the total execution loss of N sampled optimization samples, and obtain the deviation optimization model after iterative training.

[0164] Specifically, by minimizing the total execution loss of N sampling optimization samples, that is, the sum of the losses between all the correction reference coordinate points and the corresponding actual execution coordinate points, the correction reference coordinate points gradually guide the system-controlled sampling trajectory of the sampler drill rod to approach the preset target point. With each iteration, the model parameters of the matching optimization model are gradually adjusted, and the output optimal prediction deviation vector can also be based on the execution path of the sampling optimization sample, that is, optimization is performed at each point, and the sampling error is reduced point by point, which ultimately makes the calculation of the system's correction reference coordinate points more accurate, ensuring that the overall sampling performance of the system is optimal.

[0165] The S6 specifically includes:

[0166] S6-1, calculating the execution loss between each actual execution point and the correction reference coordinate point;

[0167] S6-2, summing the execution loss between each actual execution point and the correction reference coordinate point to obtain the total execution loss of N sample optimization samples;

[0168] The expression of the total execution loss is: ; represents the total execution loss;

[0169] S6-3. Calculate the gradient according to the total execution loss and update the model parameters of the matching optimization model;

[0170] S6-4. Iterate and execute S6-1 and S6-3 until the total execution loss is minimized.

[0171] In summary, the strategy of dynamic point-by-point deviation optimization in the present invention enables the system to have higher environmental adaptability. No matter under complex conditions such as vehicle vibration, uneven distribution of raw materials or fluctuation of mechanical equipment precision, it can maintain high sampling accuracy and execution efficiency, which significantly improves the robustness of the system.

[0172] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.) means.

[0173] The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., DVD ), or semiconductor media. The semiconductor media may be a solid state drive.

[0174] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not performed. Another point, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interface, indirect coupling or communication connection of the device or unit, which can be electrical, mechanical or other forms.

[0175] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A vehicle automatic sampling system for industrial raw materials, characterized in that: The sampling system comprises: A stopping module, used to stop a vehicle loaded with industrial raw materials when it enters a sampling area; A laser radar scanning module, used to scan the vehicle using a laser radar to generate a three-dimensional coordinate point cloud of the vehicle compartment; A sampling target point generation module is used to randomly generate a number of preset target points in the three-dimensional coordinate point cloud of the carriage to construct a preset target point set; A correction reference coordinate point generation module is used to input a preset target point set into a preset deviation optimization model to generate corresponding correction reference coordinate points; The correction reference coordinate point is a correction intermediate point generated point by point by the deviation optimization model according to the coordinates of the preset target point, and is used to correct the execution path of the sampling machine drill rod; An automatic sampling module is used to control the sampling machine drill rod point by point according to the calibrated reference coordinates to go to the preset target point for sampling; The modeling steps of the deviation optimization model include: S1. Obtain a training set containing N sampling optimization samples; S2. Input a training set containing N sampled optimization samples into a multi-layer perception model. The multi-layer perception model extracts a number of sampled optimization samples in batches for forward propagation to output a single-point prediction deviation vector. After iterative training, a matching degree optimization model is obtained. S3, inputting the training set containing N sample optimization samples into the matching degree optimization model to generate the optimal prediction deviation vector of the first actual execution point; S4, adding the optimal prediction deviation vector of the first actual execution point and its coordinates dimension by dimension to generate a correction reference coordinate point for the sampling machine drill rod to move to the preset target point; The expression of the correction reference coordinate point is: ; represents the correction reference coordinate point of the kth point, represents the actual execution point of the kth point, represents the optimal prediction deviation vector of the kth point; S5, obtaining the execution loss of the sampling machine drill pipe in the actual sampling operation; S6. Minimize the total execution loss of N sampled optimization samples, and obtain the deviation optimization model after iterative training.

2. The vehicle automatic sampling system for industrial raw materials according to claim 1, characterized in that: Get a training set containing N sampling optimization samples, including: S1-1. Use laser radar to obtain the three-dimensional coordinate point cloud of the carriage; The expression of the three-dimensional coordinate point cloud is: ; C is the three-dimensional coordinate point cloud, are the X-axis coordinate, Y-axis coordinate and Z-axis coordinate of the i-th point respectively, and M is the total number of three-dimensional point cloud coordinates; S1-2, selecting a number of preset target points in the three-dimensional coordinate point cloud, and aggregating the plurality of preset target points into a preset target point set; The expression of the preset target point set is: ; is the preset target point set, which represents all randomly selected preset target points; are the X-axis coordinate, Y-axis coordinate and Z-axis coordinate of the k-th preset target point, respectively, and N is the total number of preset target points; S1-3, obtaining several actual execution points of the sampling machine drill rod, and summarizing them to obtain an actual execution point set; The expression of the actual execution point set is: ; is the actual execution point set, indicating the sampling point actually reached by the sampling machine drill rod; are the X-axis coordinate, Y-axis coordinate and Z-axis coordinate of the kth actual execution point, respectively; N is the total number of actual execution points, which is equal to the total number of preset target points; S1-4, calculating the true deviation vector of each point between the preset target point set and the actual execution point set point by point; The expression of the true deviation vector is: ； is the kth true deviation vector, including the coordinate deviations in the X-axis, Y-axis and Z-axis directions; is the coordinate deviation component of the kth true deviation vector on the X-axis, Y-axis and Z-axis; S1-5, summarizing each point-by-point calculated true deviation vector into a true deviation vector set; The expression of the true deviation vector set is: represents the set of true deviation vectors; S1-6, generating continuous sampling numbers based on the sampling order of the actual execution points, and standardizing the sampling numbers to obtain sampling number features; wherein the sampling numbers are generated based on the X-axis coordinate values ​​of the actual execution points sorted from small to large; S1-7, performing vector concatenation of each preset target point, sampling sequence number feature and the global feature of the 3D point cloud after dimensionality reduction to generate a sampling optimization input vector; The expression of the sampling optimization input vector is: ； represents the input features of the kth sampling optimization sample, Represents the global features of the 3D point cloud after dimensionality reduction, including n eigenvalues; Represents the kth sampling sequence feature; This means that for every k, there is ; S1-8, using the sampling optimization input vector as the input feature and the true deviation vector as the target label to construct the sampling optimization sample; The expression of the sampling optimization sample is: ; D represents the training set containing N sampling optimization samples.

3. The vehicle automatic sampling system for industrial raw materials according to claim 2, characterized in that: After iterative training, the matching optimization model is obtained, including: S2-1, inputting the sampled optimized input vectors in the sampled optimized samples into the input layer one by one for feature encoding to obtain the encoded input vector; wherein the input layer is a fully connected layer; S2-2, extracting features from the encoded input vector, and performing high-dimensional feature interaction through the hidden layer to obtain an input vector of the hidden layer; S2-3, mapping the input vector of the hidden layer to the prediction deviation vector through the output layer; S2-4, calculating the matching loss between the predicted deviation vector and the true deviation vector; The matching loss function is: ; in, is the matching loss value, which is used to measure the matching degree between the predicted deviation vector and the true deviation vector; N is the total number of sampling optimization samples, , , Optimize the sample prediction deviation vector for the kth sample and the coordinate deviation components on the X-axis, Y-axis and Z-axis; S2-5, performing back propagation to update the model parameters of the multi-layer perception model to obtain updated model parameters; S2-6, iteratively execute S2-1 and S2-5 until the minimum matching loss is obtained; S2-7. The model that minimizes the matching loss is regarded as the matching optimization model.

4. The vehicle automatic sampling system for industrial raw materials according to claim 1, characterized in that: Obtain the execution loss of the sampler drill pipe in the actual sampling operation, including: S5-1, control the sampling machine drill rod to move according to the generated calibration reference coordinate point and perform the sampling operation; S5-2, record the coordinates of the actual execution point of the sampling machine drill rod; S5-3, calculating the execution loss between the actual execution point and the correction reference coordinate point; The execution loss is characterized by: the L2 norm of the actual execution coordinate point and the correction reference coordinate point; The expression of the L2 norm is: ； in, Indicates the distance between the calibration reference coordinate point and the actual execution point. represents the L2 norm, represents the correction reference coordinate point of the kth sampling optimization sample; , , They represent the X-axis coordinate, Y-axis coordinate, and Z-axis coordinate of the actual execution point respectively. , , They represent the X-axis, Y-axis, and Z-axis coordinates of the calibration reference coordinate point, respectively.

5. The vehicle automatic sampling system for industrial raw materials according to claim 1, characterized in that: Minimize the total execution loss of N sample optimization samples, and obtain the deviation optimization model after iterative training, including: S6-1, calculating the execution loss between each actual execution point and the correction reference coordinate point; S6-2, summing the execution loss between each actual execution point and the correction reference coordinate point to obtain the total execution loss of N sample optimization samples; The expression of the total execution loss is: ; represents the total execution loss; S6-3. Calculate the gradient according to the total execution loss and update the model parameters of the matching optimization model; S6-4. Iterate and execute S6-1 and S6-3 until the total execution loss is minimized.

6. The vehicle automatic sampling system for industrial raw materials according to claim 1, characterized in that: The method further includes: using a laser radar to scan the vehicle, generate a three-dimensional coordinate point cloud of the vehicle compartment, and randomly generate a number of preset target points within the three-dimensional coordinate point cloud of the vehicle compartment to construct a preset target point set, and also includes: using a laser radar to locate the spatial coordinates of the vehicle compartment in real time.

7. The vehicle automatic sampling system for industrial raw materials according to claim 6, characterized in that: The real-time positioning of the spatial coordinates of the carriage using a laser radar includes: using a point cloud segmentation algorithm to filter point cloud data other than the carriage to generate a three-dimensional coordinate point cloud of the carriage.

Images