Steel raw material weight measuring system and method based on three-dimensional scanning and dynamic density correction

Abstract

The invention provides a steel raw material weight measurement system based on three-dimensional scanning and dynamic density correction, and aims to solve the problems of low efficiency, large error and the like of a traditional weighing method and a volume estimation method in steel plant raw material weight measurement. According to the method, geometric point cloud data and spectral characteristics of a raw material pile are obtained through a multi-mode three-dimensional scanning module (including laser radar and multi-spectral imaging), and accurate classification of various raw materials such as iron ore, waste steel and flux is achieved in combination with raw material classification and a density database. Meanwhile, through a dynamic density correction model, factors such as raw material types, impurity proportions and moisture content are comprehensively considered, density parameters are dynamically adjusted, and the weight measurement precision is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the intelligent weight detection technology of iron and steel metallurgy raw materials, and specifically relates to a steel raw material weight measurement system and method based on three-dimensional scanning and dynamic density correction. Background Art

[0002] At present, the weight measurement of raw materials in steel plants mainly relies on traditional weighing methods and volume estimation methods, but there are significant defects: the traditional weighing method requires batch weighing of raw materials such as iron ore and scrap steel, with low efficiency and unable to monitor the dynamic changes of the storage yard in real time. For fluxes, due to their fine particles and easy mixing, the weighing error often exceeds 5%; although the volume estimation method calculates the weight through the pile volume conversion, the density of iron ore is affected by mineral types and impurities, resulting in a large static model error, and for scrap steel, due to the mixed materials and irregular shapes, the pile density fluctuates greatly. The existing technology lacks a comprehensive solution that integrates three-dimensional scanning, material identification, and dynamic density correction, and it is difficult to meet the high-precision and real-time measurement requirements of multi-raw material scenarios in steel plants. There are problems such as low efficiency and large errors. Therefore, it is imperative to develop a raw material dynamic metering system with real-time perception-intelligent correction-global linkage. Summary of the Invention

[0003] The purpose of the present invention is to provide a steel raw material weight measurement system and method based on three-dimensional scanning and dynamic density correction for the problems in the weight measurement of existing steel enterprises that are difficult to meet the high-definition and real-time measurement requirements of multi-raw material scenarios in steel plants.

[0004] To achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0005] A steel raw material weight measurement system based on three-dimensional scanning and dynamic density correction includes the following modules: a multi-modal three-dimensional scanning module, a raw material classification and density database module, a volume calculation module, a dynamic density correction module, and an automated output module;

[0006] Among them, the multi-modal three-dimensional scanning module consists of a lidar and a hyperspectral imaging device. The lidar is used to collect the three-dimensional geometric point cloud data of the raw material pile; the hyperspectral imaging device identifies the raw material type and surface impurities through spectral features;

[0007] The raw material classification and density database module stores the dynamic density parameters of iron ore, scrap steel, and fluxes;

[0008] The volume calculation module calculates the volume of the raw material pile through a spatial voxel segmentation algorithm based on the point cloud data obtained by the multi-modal three-dimensional scanning module and generates a three-dimensional volume model;

[0009] The dynamic density correction module uses a prediction algorithm with multi-parameter fusion for density prediction;

[0010] The automated output module integrates 3D scanning data, dynamic density correction values, and raw material classification information, performs volume calculations, and automatically generates the required reports.

[0011] A further preferred solution, wherein the steps of the prediction algorithm for multi-parameter fusion include the following:

[0012] (a) Establish a feature vector space: The input parameters include raw material type coding, spectral impurity concentration, infrared moisture value, and historical density deviation data;

[0013] (b) Train a density correction model: Use the gradient boosting decision tree algorithm, with the actual weighing data as the supervision signal to optimize the model parameters;

[0014] (c) Dynamically adjust coefficients: Real-time update the model weights through a Kalman filter, and dynamically adjust the iron ore SiO2 compensation coefficient, scrap inclusion weight factor, and moisture attenuation coefficient.

[0015] A method for measuring the weight of steel raw materials using a steel raw material weight measurement system with 3D scanning and dynamic density correction, including the following steps:

[0016] The first step, hardware deployment: Deploy the vehicle-mounted lidar array on the top of the raw material yard inspection vehicle to ensure 360° non-blind spot scanning; Install the multi-spectral industrial camera synchronously with the lidar, with the lens facing the surface of the raw material pile, and adjust the focal length to 4K resolution;

[0017] The second step, data acquisition and processing: Start the lidar and multi-spectral camera. The lidar scans the raw material pile at a frequency of 10Hz to generate point cloud data with the number of points per frame ≥ 1 million; The multi-spectral camera collects images in the red band, green band, blue band, near-infrared band, and short-infrared band, with an image resolution of 3840×2160; Data synchronization and registration use timestamp synchronization and dynamic registration algorithms to align the point cloud and spectral data and eliminate mechanical vibration errors;

[0018] The third step, extract the spectral characteristics of iron oxides and impurities, identify iron oxides and impurities using the spectral characteristics, and analyze the reflectance attenuation ratio of the 1450 nm and 1940 nm bands in the near-infrared spectrum to determine the moisture content;

[0019] The fourth step, input the iron oxide reflectance and impurity absorption peak spectral characteristics extracted in the third step into a pre-trained multi-task classification model to classify the types of raw materials and impurities;

[0020] The fifth step, volume calculation: Call the volume calculation module to generate a 3D voxel model and output the volume of the raw material pile;

[0021] The sixth step, retrieve the initial density: According to the raw material classification result, retrieve the density value of the corresponding raw material from the density database;

[0022] Step 7, execute the correction logic: Obtain the impurity content and moisture content through the spectral analysis technology of the multispectral imaging device, encode the hematite type as [1, 0, 0], combine the historical density deviation data, and use the formula: (original value - minimum value) / (maximum value - minimum value) to construct the standardized feature vector [0.82, 0.67, 0.43]; call the XGBoost model optimized by the gradient boosting decision tree for inference, and output the density correction coefficient; calculate the corrected density: reference density × density correction coefficient × moisture attenuation factor;

[0023] Step 8, weight calculation and report generation: Final weight = corrected density × volume, and the automated output module (105) outputs a PDF report.

[0024] Compared with the prior art, the present invention has the following advantages:

[0025] First, synchronized acquisition by lidar and multispectral camera is adopted to achieve 360° real-time scanning of the yard, and the measurement time is compressed from the traditional 35 - 60 minutes to within 10 minutes, greatly improving the efficiency;

[0026] Second, identify iron ore, scrap steel, impurity types and moisture content through multispectral, and use the gradient boosting decision tree dynamic correction model to reduce the density error of the mixed raw materials from 7.8% to 2.3%;

[0027] The present invention is applicable to the weight measurement of steel plant raw material incoming inspection and yard inventory management. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is the overall architecture diagram of the system and method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] As Figure 1 shown, a steel raw material weight measurement system based on three-dimensional scanning and dynamic density correction described in this embodiment includes the following modules: multimodal three-dimensional scanning module 101, raw material classification and density database module 102, volume calculation module 103, dynamic density correction module 104, automated output module 105;

[0030] Among them, the multimodal three-dimensional scanning module 101 is composed of a lidar and a multispectral imaging device. The lidar is used to collect three-dimensional geometric point cloud data of the raw material pile; the multispectral imaging device identifies the raw material type and surface impurities through spectral features;

[0031] The raw material classification and density database module 102 stores the dynamic density parameters of iron ore, scrap steel, and flux;

[0032] The volume calculation module 103 calculates the volume of the raw material pile based on the point cloud data obtained by the multimodal three-dimensional scanning module 101 through a spatial voxel segmentation algorithm and generates a three-dimensional volume model;

[0033] The dynamic density correction module 104 uses a prediction algorithm that fuses multiple parameters to predict density;

[0034] The steps of the prediction algorithm that fuses multiple parameters include the following:

[0035] (a) Establish a feature vector space: The input parameters include the raw material type code, spectral impurity concentration, infrared moisture value, and historical density deviation data;

[0036] (b) Train the density correction model: Use the gradient boosting decision tree algorithm and take the actual weighing data as the supervision signal to optimize the model parameters;

[0037] (c) Dynamically adjust the coefficients: Use a Kalman filter to update the model weights in real time and dynamically adjust the iron ore SiO2 compensation coefficient, scrap inclusion weight factor, and moisture attenuation coefficient.

[0038] The automated output module 105 integrates three-dimensional scanning data, dynamic density correction values, and raw material classification information, and performs volume calculations to automatically generate the required reports.

[0039] The method for measuring the weight of steel raw materials by the three-dimensional scanning and dynamic density correction steel raw material weight measurement system includes the following steps:

[0040] The first step, hardware deployment: Deploy the vehicle-mounted lidar array on the top of the raw material yard detection vehicle to ensure 360° non-blind spot scanning; Install the multispectral industrial camera synchronously with the lidar, with the lens facing the surface of the raw material pile, and adjust the focal length to 4K resolution;

[0041] The second step, data collection and processing: Start the lidar and the multispectral camera. The lidar scans the raw material pile at a frequency of 10Hz to generate point cloud data with a single-frame point count ≥ 1 million; The multispectral camera collects images in the red band, green band, blue band, near-infrared band, and short-infrared band, with an image resolution of 3840×2160; Time stamp synchronization and dynamic registration algorithms are used for data synchronization and registration to align the point cloud and spectral data and eliminate mechanical vibration errors;

[0042] The third step, extract the spectral characteristics of iron oxides and impurities, use the spectral characteristics to identify iron oxides and impurities, and analyze the reflectance attenuation ratio in the 1450 nm and 1940 nm bands of the near-infrared spectrum to determine the moisture content;

[0043] In the fourth step, input the reflectivity of the iron oxide and the spectral characteristics of the impurity absorption peak extracted in the third step into the pre-trained multi-task classification model to classify the types of raw materials and impurities;

[0044] In the fifth step, volume calculation: Call the volume calculation module 103 to generate a three-dimensional voxel model and output the volume of the raw material pile;

[0045] In the sixth step, retrieve the initial density: According to the raw material classification result, retrieve the density value of the corresponding raw material from the density database 102;

[0046] In the seventh step, execute the correction logic: Obtain the impurity content and moisture content through the spectral analysis technology of the multi-spectral imaging device. Encode the hematite type as [1,0,0]. Combine the historical density deviation data and use the formula: (original value - minimum value) / (maximum value - minimum value) to construct a standardized feature vector [0.82,0.67,0.43]; Call the XGBoost model optimized by the gradient boosting decision tree for inference and output the density correction coefficient; Calculate the corrected density: reference density × density correction coefficient × moisture attenuation factor;

[0047] In the eighth step, weight calculation and report generation: Final weight = corrected density × volume, and the automated output module 105 outputs a PDF report.

Claims

1. A steel raw material weight measurement system based on three-dimensional scanning and dynamic density correction, characterized in that It includes the following modules: multimodal three-dimensional scanning module (101), raw material classification and density database module (102), volume calculation module (103), dynamic density correction module (104), and automated output module (105); Among them, the multimodal three-dimensional scanning module (101) consists of a lidar and a multispectral imaging device. The lidar is used to collect three-dimensional geometric point cloud data of the raw material pile; the multispectral imaging device identifies the raw material type and surface impurities through spectral features; The raw material classification and density database module (102) stores the dynamic density parameters of iron ore, scrap steel, and fluxes; The volume calculation module (103) calculates the volume of the raw material pile through a spatial voxel segmentation algorithm based on the point cloud data obtained by the multimodal three-dimensional scanning module (101), and generates a three-dimensional volume model; The dynamic density correction module (104) uses a prediction algorithm for multi-parameter fusion to predict density; The automated output module (105) integrates three-dimensional scanning data, dynamic density correction values, and raw material classification information, and performs volume calculation to automatically generate the required report.

2. The steel raw material weight measuring system based on three-dimensional scanning and dynamic density correction according to claim 1, characterized in that Among them, the steps of the prediction algorithm for multi-parameter fusion include the following: (a) Establish a feature vector space: The input parameters include raw material type coding, spectral impurity concentration, infrared moisture value, and historical density deviation data; (b) Train the density correction model: Use the gradient boosting decision tree algorithm, with the actual weighing data as the supervision signal to optimize the model parameters; (c) Dynamically adjust the coefficients: Real-time update the model weights through a Kalman filter, and dynamically adjust the SiO2 compensation coefficient of iron ore, the inclusion weight factor of scrap steel, and the moisture attenuation coefficient.

3. A method for measuring the weight of steel raw materials using the three-dimensional scanning and dynamic density correction steel raw material weight measurement system according to claim 1 or 2, characterized in that: It includes the following steps: The first step, hardware deployment: Deploy the vehicle-mounted lidar array on the top of the raw material yard detection vehicle to ensure 360° non-blind spot scanning; synchronously install the multispectral industrial camera with the lidar, with the lens facing the surface of the raw material pile, and adjust the focal length to 4K resolution; The second step, data collection and processing: Start the lidar and the multispectral camera. The lidar scans the raw material pile at a frequency of 10Hz to generate point cloud data with a single-frame point count ≥ 1 million; the multispectral camera collects images in the red, green, blue, near-infrared, and short-infrared bands, with an image resolution of 3840×2160; Data synchronization and registration use timestamp synchronization and dynamic registration algorithms to align the point cloud and spectral data and eliminate mechanical vibration errors; The third step, extract the spectral features of iron oxides and impurities, use the spectral features to identify iron oxides and impurities, and analyze the reflectance attenuation ratio of the 1450 nm and 1940 nm bands in the near-infrared spectrum to determine the moisture content; The fourth step, input the iron oxide reflectance and impurity absorption peak spectral features extracted in the third step into a pre-trained multi-task classification model to classify the types of raw materials and impurities; The fifth step, volume calculation: Call the volume calculation module (103) to generate a three-dimensional voxel model and output the volume of the raw material pile; The sixth step, retrieve the initial density: According to the raw material classification result, retrieve the density value of the corresponding raw material from the density database (102); Step 7, execute the correction logic: Obtain the impurity content and moisture content through the spectral analysis technology of the multispectral imaging device, encode the hematite type as [1, 0, 0], combine the historical density deviation data, and use the formula: (original value - minimum value) / (maximum value - minimum value) to construct the standardized feature vector [0.82, 0.67, 0.43]; Call the XGBoost model optimized by the gradient boosting decision tree for inference, and output the density correction coefficient; Calculate the corrected density: reference density × density correction coefficient × moisture attenuation factor; Step 8, weight calculation and report generation: Final weight = corrected density × volume, and the automated output module (105) outputs a PDF report.

Images