Intelligent sanding AGV system and inventory sensing and flow self-calibration method thereof

Abstract

The application discloses an intelligent sand feeding AGV system and a stock sensing and flow self-calibration method thereof, and belongs to the technical field of rail transit. The system comprises an AGV mobile vehicle body, a sand feeding assembly, a stock sensing subsystem, a flow self-calibration subsystem and a navigation positioning subsystem. The stock sensing subsystem obtains a three-dimensional point cloud of a sand surface by using a 3D structured light, fits a funnel-shaped sand surface by using an RANSAC algorithm, separates a moving sand flow from a static mountain-shaped sand surface by using an HDBSCAN algorithm, and realizes accurate stock detection and multi-stage overflow prevention control. The flow self-calibration subsystem establishes a binary regression model based on a weighing sensor and the rotating speed of a star-shaped unloader, and dynamically calibrates the sand feeding flow in real time. The application solves the problems of inaccurate stock detection, difficulty in preventing overflow and large flow measurement error caused by irregular sand surface morphology, dust interference and sand material property changes in the prior art, and realizes full-process automatic and accurate sand feeding operation.

Description

Technical Field

[0001] This invention relates to the field of rail transit technology, and in particular to an intelligent sand-loading AGV system and its inventory sensing and flow self-calibration method. Background Technology

[0002] Increasing wheel-rail adhesion coefficient through sand application is crucial for ensuring safe operation of locomotives. Currently, sand application on locomotives mainly relies on manual labor or semi-automatic equipment, resulting in high labor intensity, low efficiency, and severe dust pollution. Automated sand application using Automated Guided Vehicles (AGVs) has become a trend, but existing solutions mostly only address sand handling, lacking effective solutions for ensuring the correct amount and accuracy of sand application. The main technical challenges are as follows: 1) Difficulty in initial inventory detection: When the locomotive is running, sand is consumed from the bottom of the sand box, naturally forming a funnel-shaped depression with a low center and high edges. Furthermore, sand blocks easily clump together on the inner wall of the sand box, further exacerbating the irregularity of the sand surface shape. Traditional single-point ranging sensors (such as ultrasonic and laser point ranging) cannot fully capture the three-dimensional contour of the funnel-shaped surface, resulting in serious inaccuracies in the calculation of sand volume.

[0003] 2) Difficulties in detecting sand levels and preventing overflow during sand adding: When adding sand, the sand falls from the top, forming a mountain-like accumulation pattern in the sand box, with a high center and low edges. At the same time, the dust concentration in the sand box increases sharply, seriously interfering with the imaging quality and signal stability of the optical sensor. The dynamic changes in the position of the mountain peaks mean that the fixed threshold triggering method cannot accurately determine the "almost full" state, easily leading to underfilling or overflow.

[0004] 3) Inaccurate sand flow rate measurement: AGVs typically use star-shaped unloaders and screw conveyors for feeding, but the bulk density and flowability of sand can change due to variations in particle size distribution and moisture content. At the same unloader speed, the actual output sand weight flow rate is not constant. Existing technologies that rely solely on weighing sensors or volumetric flow meters cannot provide real-time and accurate weight flow rate feedback, resulting in a significant error in the final total sand volume. Summary of the Invention

[0005] To address the above problems, this invention provides an intelligent sand-filling AGV system and its inventory sensing and flow self-calibration method, aiming to achieve full-process automation and precise control of the sand filling operation of the locomotive sand box, effectively solving the technical problems of inaccurate inventory detection and large errors in sand filling flow measurement in the prior art.

[0006] In a first aspect, the present invention provides an intelligent sand-loading AGV system, including an AGV mobile vehicle body, a material storage device, an unloading device, a material conveying device, a stock sensing subsystem, a flow self-calibration subsystem, and a docking and positioning module. The AGV mobile vehicle body has storage space inside; The storage device, unloading device, and conveying device are all located within the accommodating space. The discharge port of the storage device is connected to the inlet of the unloading device, the discharge port of the conveying device is connected to one end of the conveying device, and the other end of the conveying device is connected to the locomotive sand box. The docking and positioning module is installed at the end of the material conveying device and is used to achieve automatic alignment and connection with the locomotive sand box opening; The stock sensing subsystem is installed at one end of the material conveying device near the locomotive sand box. It is used to acquire the first point cloud data of the sand surface inside the locomotive sand box before sand is added, acquire the second point cloud data of the sand surface inside the locomotive sand box during sand addition, calculate the real-time sand addition volume inside the locomotive sand box based on the first point cloud data and the second point cloud data, and calculate the real-time sand addition flow rate based on the preset sand density and the real-time sand addition volume. The flow self-calibration subsystem is connected to the storage device, the unloading device, and the conveying device respectively. It is used to collect the sand output weight of the storage device, the rotation speed of the unloading device, and the sand flow rate measurement value of the conveying device in real time during the sand adding process. Based on the sand output weight, it calculates and outputs the weight change rate of the sand. Based on the weight change rate and the rotation speed of the unloading device, it calculates the actual sand adding flow rate. When the calibration conditions are met, it corrects the sand adding flow rate measurement value according to the actual sand adding flow rate. The flow self-calibration subsystem is also used to: use the real-time sand addition flow rate as a reference sand addition flow rate, compare the reference sand addition flow rate with the actual sand addition flow rate, and perform the calibration when the deviation exceeds a threshold.

[0007] Furthermore, the existing sensing subsystem is also used for auxiliary lighting and dust removal; The stock perception subsystem includes: A 3D structured light contour scanning device is installed at one end of the material conveying device near the locomotive sand box. It is used to project coded structured light into the locomotive sand box. The coded structured light forms a deformed pattern after being highly modulated by the sand surface. The deformed pattern is captured by the camera built into the 3D structured light contour scanning device to generate three-dimensional point cloud data of the sand surface. An auxiliary lighting and dust removal device is installed next to the 3D structured light contour scanning device. It includes a ring-shaped LED light source and a dust-proof blowing nozzle, which is used to remove dust from the camera lens and provide stable illumination before scanning.

[0008] By integrating a 3D structured light contour scanning device and an auxiliary lighting and dust removal device into the existing sensing subsystem, this invention can acquire high-density three-dimensional point cloud data of sand surfaces in high-dust environments: the ring LED light source provides stable illumination and enhances the imaging signal-to-noise ratio; the dust-proof blowing nozzle removes lens dust before each scan, effectively avoiding the interference of dust adhesion on the point cloud quality and providing reliable raw data for subsequent surface fitting.

[0009] Furthermore, the flow self-calibration subsystem includes: A weighing sensor group is installed between the storage device and the frame of the AGV mobile vehicle to monitor the total weight of sand in the storage device in real time. A speed sensor is installed on the drive shaft of the unloading device to monitor the speed of the unloading device in real time; A sand flow meter is installed at the outlet of the unloading device to measure the instantaneous volumetric flow rate of sand. The flow processing module is used to calculate the instantaneous mass flow rate based on the preset sand density and the instantaneous volumetric flow rate.

[0010] By using a weighing sensor to monitor the total weight of the silo in real time, a speed sensor to accurately feed back the speed of the unloader, and a sand flow meter to provide auxiliary reference for volumetric flow rate, the system not only provides the necessary input parameters for the binary regression model, but also achieves redundant backup of multi-source data, significantly improving the reliability and anti-interference capability of flow measurement.

[0011] Furthermore, the flow self-calibration subsystem is also used to dynamically adjust the rotation speed of the unloading device based on the calibrated sand flow rate measurement value.

[0012] The flow self-calibration subsystem dynamically adjusts the rotation speed of the unloading device based on the calibrated sand addition flow measurement value, forming a closed-loop control that effectively suppresses flow drift caused by changes in sand properties (such as density and flowability fluctuations), ensuring the stability and high precision of the sand addition process.

[0013] Secondly, the present invention also provides a method for stock perception and flow self-calibration of an intelligent sand-laying AGV system, comprising the following steps: Before adding sand, the first point cloud data of the sand surface inside the locomotive sand box is obtained. During the sand adding process, the second point cloud data of the sand surface inside the locomotive sand box is obtained. The real-time sand adding volume inside the locomotive sand box is calculated based on the first point cloud data and the second point cloud data. The real-time sand adding flow rate is calculated based on the preset sand density and the real-time sand adding volume. During the sand adding process, the sand weight of the storage device, the rotation speed of the unloading device, and the sand adding flow rate of the conveying device are collected in real time. The weight change rate of the sand is calculated based on the sand weight. The actual sand adding flow rate is calculated based on the weight change rate and the rotation speed of the unloading device. When the calibration conditions are met, the sand adding flow rate measurement value is corrected according to the actual sand adding flow rate. The real-time sand addition flow rate is used as the reference sand addition flow rate, and the reference sand addition flow rate is compared with the actual sand addition flow rate. When the deviation exceeds the threshold, the calibration is performed.

[0014] Furthermore, after acquiring the first point cloud data of the sand surface inside the locomotive sand box before adding sand, the volume of the first sand material is calculated based on the preprocessed three-dimensional point cloud data, including the following steps: The 3D structured light contour scanning device projects coded structured light into the locomotive sand box. The coded structured light forms a deformed pattern after being modulated by the height of the sand surface. The deformed pattern is captured by the camera built into the 3D structured light contour scanning device to generate the first point cloud data of the sand surface. The random sampling consensus algorithm is used to fit the first point cloud data to obtain a surface model that reflects the shape of the sand surface. The Boolean difference operation is performed between the surface model and the pre-stored empty sand box reference 3D model to calculate the volume of the enclosed space from the bottom of the sand surface to the bottom of the sand box, thus obtaining the first sand volume.

[0015] In the initial inventory identification before sand addition, by performing RANSAC surface fitting on the first point cloud data and Boolean subtraction with the empty sand box baseline model, the three-dimensional morphology of the funnel-shaped concave sand surface can be accurately restored and the remaining sand volume can be calculated. Compared with the traditional single-point ranging method, this method overcomes the interference of irregular sand surface, dust noise, and caking sand blocks, and reduces the relative error of the initial inventory detection.

[0016] Furthermore, after obtaining the surface model reflecting the sand surface morphology, the point cloud that deviates from the sand surface morphology is marked as an outer point to form an outer point cluster. If the number of point clouds in the outer point cluster exceeds a preset number and the spatial position is continuously distributed on the inner wall of the locomotive sand box, then the outer point cluster is a plated sand block area. The volume of the plated sand block is calculated and deducted from the first sand material volume or marked separately.

[0017] The above method not only improves the accuracy of the remaining sand quantity calculation, but also enables the issuance of maintenance prompts to operators, thus realizing preventive maintenance of the equipment.

[0018] Furthermore, the step of acquiring second point cloud data of the sand surface inside the locomotive sand box during the sand-adding process, and calculating the real-time sand-adding volume inside the locomotive sand box based on the first and second point cloud data, includes the following steps: During the sand-addition process, the 3D structured light contour scanning device is controlled by a preset cycle to scan the sand surface inside the locomotive sand box multiple times to collect second point cloud data. Before each scan, dust removal and blowing are performed through the dust-proof blowing nozzle. The second point cloud data is clustered and segmented using a hierarchical density clustering algorithm to separate the static sand surface point cloud from the moving sand flow point cloud. The static sand surface point cloud is extracted, which is the point cloud whose spatial position change is less than a preset change threshold within a preset time window. The random sampling consensus algorithm was used to fit the static sand surface point cloud to obtain a mountain-shaped surface. Based on the mountain-shaped curved surface, the differential volume method is used to calculate the volume of the second sand material in the current sand box, and the real-time sand addition volume is estimated based on the volume of the first sand material and the volume of the second sand material.

[0019] This method effectively overcomes the interference of high dust environment and moving sand flow obstruction during sand addition, and achieves accurate tracking of the shape and volume of mountain-shaped sand surface.

[0020] Furthermore, after obtaining the mountain-shaped surface through surface fitting, the following steps are also included: The coordinates of the highest point of the mountain-shaped curved surface are extracted as the current peak height of the sand pile. Based on the ratio of the peak height to the total depth of the sand box, the rotation speed of the unloading device is dynamically adjusted. When the ratio does not reach the first preset threshold, the unloading device is kept running at the rated speed. When the ratio reaches the first preset threshold, the rotation speed of the unloading device will be reduced to 50% of the rated speed; When the ratio reaches the second preset threshold, the rotation speed of the unloading device will be reduced to 25% of the rated rotation speed; When the ratio reaches the third preset threshold, unloading stops.

[0021] The aforementioned gradual speed control strategy not only avoids sand overflow but also maximizes the amount of sand added at one time, reducing the number of times sand needs to be replenished, thus significantly improving operational efficiency and safety.

[0022] Furthermore, the following conditions for stopping sand addition are determined in sequence. When any one of these conditions is met, sand addition is stopped and subsequent determinations are terminated: The peak height reaches a preset height threshold; The cumulative weight of sand added has reached the preset target weight of sand, and the cumulative weight of sand added is obtained by the actual sand addition flow rate or the weight change rate. The volume of the second sand reaches the preset target sand volume.

[0023] The above method effectively addresses the extreme case of single sensor failure, ensuring that the system can accurately stop adding sand under any fault mode and preventing sand spillage accidents.

[0024] Compared with the prior art, the beneficial effects of the present invention are: 1) Dual Precise Sensing Capabilities for Funnel-Shaped and Mountain-Shaped Sand Surfaces: This invention proposes a robust surface fitting method based on the RANSAC algorithm and a dynamic point cloud segmentation method based on the HDBSCAN algorithm for the funnel-shaped concave sand surface formed by sand spreading before sand filling in the locomotive sand box, and a mountain-shaped accumulated sand surface formed during sand filling. The former can accurately reconstruct the funnel-shaped surface from point cloud data containing dust noise and compacted sand block interference, and the relative error of remaining sand volume detection can be controlled within 2.5%; the latter can effectively separate the moving sand flow from the static sand surface, and achieve precise tracking of the peak height and real-time volume of the mountain-shaped sand surface.

[0025] 2) Intelligent Multi-Level Overflow Prevention Control: By real-time monitoring of the peak height of the mountain-shaped sand pile, combined with a three-level early warning threshold and a graded deceleration strategy, a progressive overflow prevention control from early warning to gradual stopping and then to complete shutdown is achieved. Simultaneously, a dual-redundancy verification mechanism using both volumetric and gravimetric methods is introduced to ensure reliable system operation even under extreme conditions where dust interferes with the visual sensors, thus preventing overflow accidents.

[0026] 3) Precise dynamic flow self-calibration: Based on the coupled calibration model of "weight change rate - unloader speed", it can compensate for density and flowability fluctuations caused by changes in sand moisture and particle size in real time. Its redundancy switching mechanism greatly improves the system's fault tolerance capability in the event of sensor failure. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this drawing or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this drawing. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the overall structure of the intelligent sand-laying AGV system; Figure 2 A schematic diagram of the storage capacity sensing in a funnel-shaped sand box; Figure 3 A schematic diagram illustrating the storage capacity of sand boxes shaped like mountain peaks; Figure 4 The overall workflow diagram for the intelligent sand-laying AGV system; Figure 5 Workflow diagram for funnel-shaped inventory identification; Figure 6 A flowchart illustrating the dynamic detection and overflow prevention process for mountain-shaped structures; Figure 7 Workflow diagram for the sand flow rate self-calibrator; Figure 8 shows the RANSAC fitted funnel sand surface cloud map and the triangular prism volume calculation diagram. Figure 9 shows the fitted cloud map of the mountain-shaped sand surface. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments provided by this invention without inventive effort are within the scope of protection of this invention.

[0030] This invention provides an intelligent sand-laying AGV system, such as... Figure 1 As shown, it specifically includes an AGV mobile vehicle body, a material storage device, an unloading device, a material conveying device, a stock sensing subsystem, a flow self-calibration subsystem, and a docking and positioning module.

[0031] The AGV mobile vehicle body includes a frame and an intelligent AGV chassis 7 installed below the frame. The frame is used to house the storage device, unloading device, conveying device and various sensors. The intelligent AGV chassis 7 is equipped with a navigation and positioning subsystem to guide the AGV mobile vehicle body to the working position at the locomotive sand box opening.

[0032] The storage device is an AGV sand box 4, the unloading device is a star-shaped unloader 5, and the conveying device includes a spiral sand feeding device 15, a sand feeding hose 14, and a horizontal sand-feeding auger 13. The AGV sand box 4 is used to store sand to be added. A sand level gauge 1 is installed on the AGV sand box 4 to detect whether the AGV sand box 4 is full of sand. The star-shaped unloader 5 is installed at the discharge port of the AGV sand box 4 and feeds the sand quantitatively through its rotating impeller. The inlet of the spiral sand feeding device 15 is directly connected to the outlet of the star-shaped unloader 5 to push the sand forward. The horizontal sand-feeding auger 13 and the spiral sand feeding device 15 are connected through the sand feeding hose 14 to transport the sand to the end sand inlet 12 of the horizontal sand-feeding auger 13, and finally inject it into the locomotive sand box 16.

[0033] In addition, the material handling device also includes a sand replenishment port 2, a scissor mechanism 8, and a moving module mechanism 9. The sand replenishment port 2 is located on the top of the AGV sand box 4 and is used to replenish sand into the AGV sand box 4. The scissor mechanism 8 is installed between the intelligent AGV chassis 7 and the horizontal sand auger 13 and is used to drive the horizontal sand auger 13 to rise and fall to adapt to the sand box openings of different heights. The scissor mechanism 8 is connected to the horizontal sand auger 13 through the moving module mechanism 9. The moving module mechanism 9 is used to control the horizontal movement of the horizontal sand auger 13 so that the sand inlet 12 at its end can accurately reach the working position directly above the sand box opening 17 of the locomotive.

[0034] The docking and positioning module is installed at the end of the material conveying device to achieve automatic alignment and connection with the locomotive sand box opening. Specifically, the docking and positioning module integrates a laser rangefinder sensor to detect the horizontal and vertical positional deviations between the upper sand opening 12 and the locomotive sand box opening 17 in real time. Based on the horizontal and vertical positional deviations, the control system drives the moving module mechanism 9 to make fine adjustments in the horizontal direction and drives the scissor mechanism 8 to move vertically, so that the upper sand opening 12 is accurately aligned and reaches the working position directly above the locomotive sand box opening 17.

[0035] The existing sensing subsystem includes a 3D structured light contour scanning device 11 and an auxiliary lighting and dust removal device. The 3D structured light contour scanning device 11 is installed on the side of the upper sand inlet 12 and is used to project coded structured light into the locomotive sand box 16 and to collect three-dimensional point cloud data of the sand surface through its built-in high-resolution CMOS camera. The device installed next to the 3D structured light contour scanning device 11 includes a ring LED light source and a dust-proof blowing nozzle 10, which is used to remove dust from the camera lens before scanning and provide stable illumination.

[0036] The flow self-calibration subsystem includes a weighing sensor group 3, a speed sensor, a sand flow meter 6, and a flow processing module. The weighing sensor group 3 is installed between the AGV sand box 4 and the frame of the AGV mobile vehicle to monitor the total weight of sand in the AGV sand box 4 in real time. The speed sensor is installed on the drive shaft of the star unloader 5 to monitor the speed of the star unloader 5 in real time. The sand flow meter 6 is installed at the outlet of the star unloader 5 to measure the instantaneous volumetric flow rate of the sand. The flow processing module is used to calculate the instantaneous mass flow rate based on the preset sand density and instantaneous volumetric flow rate.

[0037] The system's workflow is divided into three stages: pre-sand addition inventory identification, dynamic detection and overflow prevention during sand addition, and flow self-calibration throughout the entire process.

[0038] 1) Inventory identification stage before adding sand: Once the AGV accurately stops at the locomotive's sand box opening via the navigation and positioning subsystem, the first step is to perform inventory identification. For example... Figure 2As shown, at this point, due to sand consumption during sand spreading, the surface of the sand in the sand box exhibits a funnel-shaped depression with a low center and high edges, possibly accompanied by sand blocks caking on the inner wall. For this funnel-shaped depression, the system uses the differential volume method based on the empty chamber reference surface to calculate the inventory, such as... Figure 5 As shown, the specific steps are as follows: Step 1: After the robotic arm is positioned at a designated height directly above the locomotive's sand box opening, the system activates the dust-proof blowing nozzle to remove dust. Subsequently, the LED light source is turned on, and the 3D structured light contour scanning device projects coded structured light into the locomotive's sand box to acquire the original three-dimensional point cloud data of the sand surface. Statistical filtering algorithms are used to remove sparse outliers caused by suspended dust, and then Gaussian filtering is used to smooth the point cloud surface.

[0039] Step 2: Apply the RANSAC (Random Sample Consensus) algorithm to fit a surface to the smoothed point cloud data. RANSAC can accurately estimate mathematical model parameters from data containing a large amount of noise and outliers: it repeatedly samples the smallest dataset to estimate the model and selects the model with the most inliers as the final result.

[0040] Specifically, the system first randomly selects a set of points from the point cloud to fit an initial surface model. It then calculates the distance from all points to the model, marking points with a distance less than a threshold ε as interior points and counting them. This process is repeated N times, and the model with the most interior points is selected as the final fitting result. Finally, the model parameters are refined using the least squares method with all interior points. For regions that significantly deviate from the funnel shape (such as local protrusions formed by compacted sand blocks), the algorithm automatically marks them as exterior points and excludes them, ensuring that the fitted surface reflects the natural packing contour of the sand.

[0041] Step 3: During system initialization, the interior of the empty locomotive sand box is scanned beforehand using a 3D structured light contour scanning device to establish and store a reference 3D model of the sand box's inner wall. During inventory detection, the fitted funnel-shaped sand surface is intersected with the reference 3D model of the empty sand box using Boolean intersection to calculate the volume of the enclosed space from the bottom of the surface to the bottom of the sand box. The volume calculation uses a projection integration method based on Delaunay triangulation to ultimately obtain the volume of remaining sand inside the box (the first sand volume). .

[0042] Step 4: During the RANSAC fitting process, the system simultaneously records the point cloud clusters marked as external points. If the number of points in an external point cluster exceeds a preset threshold and its spatial location is continuously distributed near the inner wall of the sand box, it is determined to be a slab sand block area. Its volume is calculated separately and deducted from the inventory results. At the same time, a maintenance prompt of "It is recommended to clean the slabs on the inner wall" is issued.

[0043] Step 5: The system according to and the rated capacity of locomotive sand box It automatically calculates the target volume of sand that needs to be added this time. .

[0044] After completing the inventory identification and sand addition decision, the AGV immediately begins the sand addition operation. For example... Figure 3 As shown, as the sand falls from the top sand box opening, the sand surface inside the box gradually changes from a funnel shape to a mountain-shaped accumulation. At this point, the system enters the dynamic detection and overflow prevention control stage.

[0045] 2) Dynamic monitoring and overflow prevention stage during sand addition: During the sand filling process, the dust concentration inside the locomotive's sand box is extremely high, and the position of the sand pile peak changes dynamically. The system needs to track the sand surface height in real time to accurately determine the "almost full" state. The inventory sensing subsystem continues to work, but switches to the mountain-shaped dynamic detection mode, such as... Figure 6 As shown, the detection steps are as follows: Step 1: During the sand adding process, the 3D structured light contour scanning device scans the sand surface inside the sand box once at a preset cycle. Before each scan, it automatically performs dust removal and blowing to ensure that the lens is clean.

[0046] Step 2: Since the high-speed sand flow during sand addition will partially obscure the sand surface and generate dynamic noise, the system uses the HDBSCAN (Hierarchical Density Clustering) algorithm to cluster and segment the point cloud. This algorithm does not require pre-specifying the number of clusters and can adaptively separate the static sand surface point cloud from the moving sand flow point cloud, extracting the static sand surface point cloud, which is the point cloud whose spatial position change is less than a preset change threshold within a preset time window.

[0047] Step 3: For the static sand surface point cloud, the system again uses the RANSAC algorithm to fit the quadratic surface model. This model can accurately describe the mountain-like accumulation characteristics with a high center and low surrounding areas. After fitting, the coordinates of the highest point of the surface are extracted. ,in This is the current peak height of the sand pile.

[0048] Step 4: Based on the mountain-shaped fitted surface, calculate the real-time total volume of sand in the current box (second sand volume) using the differential volume method. Including the original remaining sand volume and real-time sand addition volume .

[0049] Step 5: Set three-level early warning thresholds to achieve tiered control: Level 1 (Warning Line): When the peak height is detected. When the sand box reaches 80% of its total depth, the system automatically reduces the sand feeding speed, reduces the rotation speed of the star-shaped unloader to 50% of its rated speed, and issues a "about to be full" warning.

[0050] Level 2 (Pause Line): When the peak height... When the sandbox reaches 90% of its total depth, the system further reduces the rotational speed to 25% of the rated speed.

[0051] Level 3 (Stop Line): When the peak height... When the sand box reaches 95% of its total depth, the system immediately stops adding sand.

[0052] Step 6: To overcome the potential interference of dusty environments on 3D vision, the system is based on real-time sand addition volume. Calculate real-time sand addition flow rate ( ρ (for sand density), and The cumulative weight of sand added is provided by the flow self-calibration subsystem. Real-time comparison is performed. If the deviation between the two exceeds a preset weight threshold, it is determined that the data acquired by the 3D structured light contour scanning device may be severely distorted, and the system automatically switches to the weight-based priority mode. As the primary basis for decision-making, that is, based on Has the target sand weight been achieved? Determine whether to stop adding sand to ensure the sand adding process is safe and reliable.

[0053] 3) Flow self-calibration phase throughout the entire operation: Because the bulk density and flowability of sand vary with factors such as particle size distribution and moisture content, precise weight flow rate control cannot be achieved solely by relying on the rotational speed of the rotary valve. Therefore, this invention proposes a dynamic calibration method for sand flow rate based on the increment of weight change and the rotational speed of the rotary valve, such as... Figure 7 As shown, the specific steps are as follows: The system pre-establishes a binary regression model through calibration experiments: ; in, This refers to the actual sand addition flow rate after calibration. n is the real-time rotational speed of the star-shaped unloader; This represents the rate of change in sand weight. W represents the weight of the sand produced; t represents time; These are the coefficients determined through calibration experiments.

[0054] The specific procedure for dynamically calibrating the readings of the sand flow meter is as follows: After the AGV starts adding sand, the control system synchronously collects the rotational speed n of the star unloader, the sand weight W of the AGV sand box, and the original reading of the sand flow meter every 100ms. Calculate the rate of change of weight within the current time window. ; Let n and Substituting into the above binary regression model, the actual sand addition flow rate at the current moment is calculated. ; Real-time sand injection flow rate As a reference sand addition flow rate, and comparing the reference sand addition flow rate with the actual sand addition flow rate. A comparison is performed, and when the deviation exceeds a threshold, based on... The original reading of the sand flow meter is corrected in real time, and... Integrate over time to obtain the accurate cumulative weight of sand added. This ensures that the displayed instantaneous flow rate and cumulative sand addition are highly consistent with the actual weight changes, and dynamically adjusts the rotation speed of the unloading device based on the calibrated sand flow meter readings.

[0055] The mass flow rate is obtained by converting the sand flow meter reading to density. Q sensor When the system detects that the mass flow rate is... When the deviation continuously exceeds the preset flow threshold, it is determined that the sand flow meter may be obstructed by dust or malfunction, triggering a sensor maintenance alarm, and the system automatically switches to a method based on the rate of change of weight. Integral mode: for the rate of change of weight Integrate over time to calculate the cumulative weight of sand added. When the accumulated weight of the added sand reaches the target weight, the system issues a stop sand adding command to complete the precise sand adding process.

[0056] The system workflow of the present invention will be described in detail below with reference to specific embodiments.

[0057] This embodiment uses a sand box for a certain type of electric locomotive (internal dimensions: 600mm×400mm×500mm, rated capacity 100L, approximately 160kg of quartz sand, bulk density 1.6g / cm³). 3 (Calculation) is the target object, and the following methods are used: Figure 1 The intelligent sand-filling AGV system shown performs a complete automated sand-filling operation, with the dust concentration in the working environment controlled at 150 mg / m³. 3 Simulates harsh working conditions.

[0058] like Figure 4As shown, the AGV system receives work tasks from the sand-loading management platform via a wireless network. Task information includes the target locomotive location and sand box number. The AGV precisely stops at the sand box opening using the navigation and positioning subsystem (positioning accuracy ±10mm). The scissor lift mechanism 8 and the moving module mechanism 9 move the horizontal sand-loading auger 13 and its end-mounted 3D structured light contour scanning device 11 to a designated height directly above the sand box opening. At this time, due to sand consumption during normal locomotive operation, the sand surface inside the sand box has a funnel-shaped depression with a low center and high edges, and approximately 2kg of hardened sand lumps adhere to one inner wall.

[0059] The existing sensing subsystem starts working, the dust blowing nozzle 10 sprays high-pressure airflow to remove dust from the lens, the ring LED light source is turned on to supplement light, and the 3D structured light contour scanning device 11 projects coded structured light into the box to obtain approximately 950,000 spatial coordinate points of the original three-dimensional point cloud data (each point contains x, y, and z coordinates, with the z axis pointing vertically upward from the bottom of the sand box as the origin). Table 1 gives five examples of the original three-dimensional point cloud data.

[0060] Table 1 Original 3D point cloud data

[0061] Statistical filtering was used to remove outliers. Specifically, the average distance from each point to its 50 nearest neighbors was calculated. The mean of the average distances to all points was set to 2.3 mm, and the standard deviation to 1.0 mm. Therefore, the distance threshold for statistical filtering was 3.3 mm (mean + 1.0 × standard deviation). After statistical filtering, approximately 65,000 points were removed (such as points located below the bottom of the sandbox and points far from the sandbox area). Then, a Gaussian kernel with a standard deviation σ = 1.5 was used to smooth the point cloud. The z-coordinate of each point was updated using the weighted average position of its neighbors. After processing, the z-coordinate of point 1 was adjusted from 185.2 mm to 184.9 mm (a change of approximately 0.16%).

[0062] The parameters of the RANSAC algorithm were set as follows: the distance threshold for RANSAC fitting was ε = 6.0 mm, the maximum number of iterations was 5000, and the target inlier ratio was 90%. After 523 iterations, 815,000 inliers were obtained (92.1% ratio), and the optimal fitted quadratic surface equation was: .

[0063] As shown in Figure 8(a), the surface exhibits a funnel-shaped feature with a low center and high edges. Meanwhile, of the 70,000 points marked as outer points, approximately 6,000 points are concentrated on one side of the inner wall of the sand box (x ranges from 570 to 600 mm). The average z-value of these outer points is 135 mm, which differs from the average z-value (118 mm) of the surrounding inner points by 17 mm. The system determines this part to be a hardened sand block area and calculates the volume of the sand blocks in this area.

[0064] As shown in Figure 8(b), the fitted surface and the pre-stored empty sand box reference 3D model were intersected using Boolean method. The volume of the enclosed space from the bottom of the surface to the bottom of the sand box was calculated using the projection integration method based on Delaunay triangulation (mesh size of 5mm×5mm, generating approximately 9600 triangles). The result was 24.68L, which is equivalent to a weight of 39.5kg. After deducting the volume of the compacted sand blocks, the remaining sand volume was 38.2kg. Manual weighing verified that the actual remaining sand was 38.8kg, with a relative error of 1.6%, which meets the design accuracy requirement of less than or equal to 5%.

[0065] Based on the remaining sand volume of 38.2 kg (24.68 L) and the rated capacity of 160 kg (100 L), the system calculates the target sand addition amount of 120.5 kg (75.32 L) and issues a maintenance prompt that "it is recommended to clean the inner wall caking".

[0066] After inventory identification is completed, the AGV begins adding sand. The star-shaped unloader starts at its rated speed of 800 rpm, and sand falls into the sand box from the top sand inlet (coordinates (300, 200, 500)) at a rate of approximately 40 kg / min. The sand surface gradually changes from a funnel shape to a mountain-shaped accumulation. The system then enters dynamic detection and overflow prevention control mode, with a 3D structured light camera periodically scanning every 5 seconds. The scanning results are shown in Figure 9.

[0067] First scan (5 seconds after sand addition begins): As shown in Figure 9(a), approximately 3.3 kg of sand has been added at this point. The original point cloud contains approximately 980,000 points. After segmentation using the HDBSCAN algorithm, the moving sand flow point cloud contains approximately 180,000 points (low density, surrounding the sand addition port), and the static sand surface point cloud contains approximately 800,000 points. RANSAC quadratic surface fitting was performed on the static sand surface point cloud, yielding a peak height of 125.4 mm and a current total volume of 30.2 L (an increase of 5.52 L from the initial volume). The peak height represents 25.1% of the sand box depth of 500 mm, and no warning was triggered; normal sand addition continues.

[0068] Third scan (15 seconds after sand addition begins): As shown in Figure 9(b), approximately 10.0 kg of sand has been added, the peak height is 285.6 mm, the total volume is 52.8 L, the peak height accounts for 57.1%, and no warning has been triggered.

[0069] Fifth scan (25 seconds after sand addition begins): As shown in Figure 9(c), approximately 16.7 kg of sand has been added, the peak height is 408.6 mm, the total volume is 78.5 L, and the peak height percentage is 81.7%, triggering the first-level warning threshold (80%). The system automatically reduces the rotation speed of the star-shaped unloader to 400 rpm (50% of the rated speed), the sand addition rate is reduced to approximately 20 kg / min, and the HMI displays a "Almost full" warning.

[0070] Sixth scan (31 seconds after sand addition begins): As shown in Figure 9(d), the peak height is 453.1 mm, the total volume is 92.8 L, and the peak height percentage is 90.6%, triggering the second-stage slow-stop line (90%). The system further reduces the rotation speed to 200 rpm (25% of the rated speed), and the sand addition rate is reduced to approximately 10 kg / min.

[0071] Seventh scan (34 seconds after sand addition begins): As shown in Figure 9(e), the peak height is 476.8 mm, the total volume is 97.5 L, and the peak height percentage is 95.4%, triggering the third-level stop line (95%). The system immediately stops the star-shaped unloader, the spiral sand feeding device, and the horizontal sand auger, and sand addition ends.

[0072] During the sand addition process, the flow self-calibration subsystem synchronously accumulates the weight of sand already added. The volumetric method estimated the real-time sand addition volume at 72.82L (97.5L - 24.68L), which translates to a weight of 116.5kg. The deviation was 2.8%, less than the 15% threshold, and did not trigger redundancy switching, indicating reliable data. After the operation, the weighbridge measured the actual sand addition at 120.2kg, with an error of only 0.33% compared to the system's detected value of 119.8kg, meeting high precision requirements.

[0073] During the sand addition process, the sand flow rate self-calibration subsystem operates in real time, dynamically correcting flow rate measurement deviations caused by changes in sand density.

[0074] First, the system underwent calibration experiments before being put into use: it was run at 400rpm, 600rpm, and 800rpm for 60 seconds respectively, and the weight loss of the weighing sensor and the reading of the sand flow meter were recorded to obtain the calibration data table in Table 2 below.

[0075] Table 2 Calibration Data Table

[0076] The least squares method was used to perform bivariate linear regression, resulting in the calibrated model: ; At a certain moment during the sand adding process, the system collected the following data: rotation speed n = 600 rpm, hopper weight W(t) = 685.3 kg, W(t-100ms) = 685.36 kg, and the average mass flow rate of the sand flow meter after density conversion was 0.68 kg / s; the calculated weight change rate dW / dt = (685.36-685.3) / 0.1 = 0.60 kg / s; substituting these values ​​into the model yielded... .

[0077] The average mass flow rate of the sand flow meter, converted from density to 0.68 kg / s, deviated by 35.4% from the calibrated value of 1.053 kg / s, exceeding the preset threshold by 15%. The system determined that the sand flow meter might be obstructed by dust and automatically switched to pure weight change rate (dW / dt) integral mode to continue measurement, displaying a maintenance prompt: "Sand flow meter malfunction, switched to gravimetric method." After switching, the system relies solely on the weighing sensor data for cumulative measurement, ensuring that sand addition accuracy is not affected.

[0078] After sand addition stops, the mobile module mechanism 9 retracts, and the sand box cover closes automatically. The AGV uploads the operation data (initial stock 38.2kg, actual sand addition 119.8kg, information on caking sand blocks, and early warning records for each stage) to the sand loading management platform via wireless network, and then automatically returns to the standby area. The entire operation process achieves fully unmanned closed-loop control from stock identification and dynamic overflow prevention to flow self-calibration, significantly improving the intelligence level and reliability of sand loading operations.

[0079] It should be noted that the present invention is not limited to the above-described embodiments. The above embodiments are merely examples, and any embodiments that have the same structure and perform the same effects as the technical concept within the scope of the present invention are included within the scope of the present invention. Furthermore, various modifications that can be conceived by those skilled in the art to the embodiments, and other ways of constructing by combining some of the constituent elements of the embodiments, without departing from the spirit of the present invention, are also included within the scope of the present invention.

Claims

1. An intelligent sand-laying AGV system, characterized in that, It includes the AGV mobile vehicle body, storage device, unloading device, conveying device, inventory sensing subsystem, flow self-calibration subsystem, and docking and positioning module; The AGV mobile vehicle body has storage space inside; The storage device, unloading device, and conveying device are all located within the accommodating space. The discharge port of the storage device is connected to the inlet of the unloading device, the discharge port of the conveying device is connected to one end of the conveying device, and the other end of the conveying device is connected to the locomotive sand box. The docking and positioning module is installed at the end of the material conveying device and is used to achieve automatic alignment and connection with the locomotive sand box opening; The stock sensing subsystem is installed at one end of the material conveying device near the locomotive sand box. It is used to acquire the first point cloud data of the sand surface inside the locomotive sand box before sand is added, acquire the second point cloud data of the sand surface inside the locomotive sand box during sand addition, calculate the real-time sand addition volume inside the locomotive sand box based on the first point cloud data and the second point cloud data, and calculate the real-time sand addition flow rate based on the preset sand density and the real-time sand addition volume. The flow self-calibration subsystem is connected to the storage device, the unloading device, and the conveying device respectively. It is used to collect the sand output weight of the storage device, the rotation speed of the unloading device, and the sand flow rate measurement value of the conveying device in real time during the sand adding process. Based on the sand output weight, it calculates and outputs the weight change rate of the sand. Based on the weight change rate and the rotation speed of the unloading device, it calculates the actual sand adding flow rate. When the calibration conditions are met, it corrects the sand adding flow rate measurement value according to the actual sand adding flow rate. The flow self-calibration subsystem is also used to: use the real-time sand addition flow rate as a reference sand addition flow rate, compare the reference sand addition flow rate with the actual sand addition flow rate, and perform the calibration when the deviation exceeds a threshold.

2. The intelligent sand-laying AGV system according to claim 1, characterized in that, The existing sensing subsystem is also used for auxiliary lighting and dust removal; The stock perception subsystem includes: A 3D structured light contour scanning device is installed at one end of the material conveying device near the locomotive sand box. It is used to project coded structured light into the locomotive sand box. The coded structured light forms a deformed pattern after being highly modulated by the sand surface. The deformed pattern is captured by the camera built into the 3D structured light contour scanning device to generate three-dimensional point cloud data of the sand surface. An auxiliary lighting and dust removal device is installed next to the 3D structured light contour scanning device. It includes a ring-shaped LED light source and a dust-proof blowing nozzle, which is used to remove dust from the camera lens and provide stable illumination before scanning.

3. The intelligent sand-laying AGV system according to claim 1, characterized in that, The flow self-calibration subsystem includes: A weighing sensor group is installed between the storage device and the frame of the AGV mobile vehicle to monitor the total weight of sand in the storage device in real time. A speed sensor is installed on the drive shaft of the unloading device to monitor the speed of the unloading device in real time; A sand flow meter is installed at the outlet of the unloading device to measure the instantaneous volumetric flow rate of sand. The flow processing module is used to calculate the instantaneous mass flow rate based on the preset sand density and the instantaneous volumetric flow rate.

4. The intelligent sand-laying AGV system according to claim 1, characterized in that, The flow self-calibration subsystem is also used to dynamically adjust the rotation speed of the unloading device based on the calibrated sand addition flow measurement value.

5. A method for inventory sensing and flow self-calibration in the intelligent sand-laying AGV system according to any one of claims 1-4, characterized in that, Includes the following steps: Before adding sand, the first point cloud data of the sand surface inside the locomotive sand box is obtained. During the sand adding process, the second point cloud data of the sand surface inside the locomotive sand box is obtained. The real-time sand adding volume inside the locomotive sand box is calculated based on the first point cloud data and the second point cloud data. The real-time sand adding flow rate is calculated based on the preset sand density and the real-time sand adding volume. During the sand adding process, the sand weight of the storage device, the rotation speed of the unloading device, and the sand adding flow rate of the conveying device are collected in real time. The weight change rate of the sand is calculated based on the sand weight. The actual sand adding flow rate is calculated based on the weight change rate and the rotation speed of the unloading device. When the calibration conditions are met, the sand adding flow rate measurement value is corrected according to the actual sand adding flow rate. The real-time sand addition flow rate is used as the reference sand addition flow rate, and the reference sand addition flow rate is compared with the actual sand addition flow rate. When the deviation exceeds the threshold, the calibration is performed.

6. The stock sensing and flow self-calibration method according to claim 5, characterized in that, After acquiring the first point cloud data of the sand surface inside the locomotive sand box before adding sand, the volume of the first sand material is calculated based on the preprocessed three-dimensional point cloud data, including the following steps: The 3D structured light contour scanning device projects coded structured light into the locomotive sand box. The coded structured light forms a deformed pattern after being modulated by the height of the sand surface. The deformed pattern is captured by the camera built into the 3D structured light contour scanning device to generate the first point cloud data of the sand surface. The random sampling consensus algorithm is used to fit the first point cloud data to obtain a surface model that reflects the shape of the sand surface. The Boolean difference operation is performed between the surface model and the pre-stored empty sand box reference 3D model to calculate the volume of the enclosed space from the bottom of the sand surface to the bottom of the sand box, thus obtaining the first sand volume.

7. The stock sensing and flow self-calibration method according to claim 6, characterized in that, After obtaining the surface model reflecting the sand surface morphology, the point cloud that deviates from the sand surface morphology is marked as an outer point to form an outer point cluster. If the number of point clouds in the outer point cluster exceeds the preset number and the spatial position is continuously distributed on the inner wall of the locomotive sand box, then the outer point cluster is a plated sand block area. The volume of the plated sand block is calculated and deducted from the first sand material volume or marked separately.

8. The stock sensing and flow self-calibration method according to claim 6, characterized in that, The process of acquiring second point cloud data of the sand surface inside the locomotive sand box during sand addition, and calculating the real-time sand volume inside the locomotive sand box based on the first and second point cloud data, includes the following steps: During the sand-addition process, the 3D structured light contour scanning device is controlled by a preset cycle to scan the sand surface inside the locomotive sand box multiple times to collect second point cloud data. Before each scan, dust removal and blowing are performed through the dust-proof blowing nozzle. The second point cloud data is clustered and segmented using a hierarchical density clustering algorithm to separate the static sand surface point cloud from the moving sand flow point cloud. The static sand surface point cloud is extracted, which is the point cloud whose spatial position change is less than a preset change threshold within a preset time window. The random sampling consensus algorithm was used to fit the static sand surface point cloud to obtain a mountain-shaped surface. Based on the mountain-shaped curved surface, the differential volume method is used to calculate the volume of the second sand material in the current sand box, and the real-time sand addition volume is estimated based on the volume of the first sand material and the volume of the second sand material.

9. The stock sensing and flow self-calibration method according to claim 8, characterized in that, After obtaining the mountain-shaped surface through surface fitting, the following steps are also included: The coordinates of the highest point of the mountain-shaped curved surface are extracted as the current peak height of the sand pile. Based on the ratio of the peak height to the total depth of the sand box, the rotation speed of the unloading device is dynamically adjusted. When the ratio does not reach the first preset threshold, the unloading device is kept running at the rated speed. When the ratio reaches the first preset threshold, the rotation speed of the unloading device will be reduced to 50% of the rated speed; When the ratio reaches the second preset threshold, the rotation speed of the unloading device will be reduced to 25% of the rated rotation speed; When the ratio reaches the third preset threshold, unloading stops.

10. The stock sensing and flow self-calibration method according to claim 9, characterized in that, The following conditions for stopping sand addition will be determined in sequence. Sand addition will be stopped and subsequent determinations will be terminated when any one of the conditions is met: The peak height reaches a preset height threshold; The cumulative weight of sand added has reached the preset target weight of sand, and the cumulative weight of sand added is obtained by the actual sand addition flow rate or the weight change rate. The volume of the second sand reaches the preset target sand volume.

Images