Coal bunker residual coal quantity detection and bunker clearance method based on AI vision

Abstract

The invention discloses a coal bunker residual coal quantity detection and bunker cleaning method based on AI vision, relates to the technical field of coal bunker residual coal quantity detection and bunker cleaning, and aims to solve the technical problem that adjustment cannot be carried out in time due to coal deposit form change at present. S1, operation is started, an intelligent bunker cleaning system is powered on, equipment self-inspection is carried out, and it is ensured that the state of each unit is normal; s2, the system controls various carried sensors to carry out all-directional scanning on the interior of the coal bunker, and original data are collected; and S3, performing fusion analysis on the acquired multi-modal data by using a core AI algorithm, accurately identifying a coal accumulation area by using a deep learning model, estimating the residual coal quantity, generating a high-precision three-dimensional coal quantity distribution map, and autonomously planning an optimal cleaning path and equipment parameters according to the high-precision three-dimensional coal quantity distribution map. Real-time feedback and dynamic optimization in the cleaning process are achieved, it is ensured that the final cleaning effect comprehensively reaches the standard, the ineffective operation cost and time are greatly reduced, and the problem that adjustment cannot be conducted in time due to coal deposit form change at present is solved.

Description

Technical Field

[0001] This invention relates to the field of coal bunker residual coal quantity detection and cleaning technology, and more specifically, to a coal bunker residual coal quantity detection and cleaning method based on AI vision. Background Technology

[0002] As a core storage facility in the coal production, storage, and transportation process, coal bunkers play a crucial role in temporary coal storage, buffering supply and demand, and ensuring production continuity. However, during storage, coal particles are prone to accumulating in areas such as the bunker walls, corners, and bottom of chutes due to factors like particle adhesion, the bunker's internal structure, and unloading methods. If this residual coal cannot be detected and thoroughly cleaned in a timely and accurate manner, it will not only lead to a gradual reduction in the effective storage capacity of the coal bunker year by year, but also affect coal turnover efficiency.

[0003] Traditional coal bunker cleaning operations are one-off processes; the cleaning machine considers the operation complete once it completes the preset path. This lack of real-time monitoring and effect analysis during the cleaning process means that if changes in coal accumulation or initial detection errors lead to excessive residual coal levels in some areas, the system cannot detect and adjust in time, ultimately resulting in substandard cleaning and requiring additional manpower and resources for secondary cleaning, further increasing operating costs and time. Therefore, we propose an AI-based vision-based method for detecting and cleaning residual coal in coal bunkers. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology, adapt to the needs of reality, and provide a method for detecting and clearing residual coal in coal bunkers based on AI vision, so as to solve the current technical problem that it is impossible to adjust in time due to changes in the form of accumulated coal.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a method for detecting and clearing residual coal in a coal bunker based on AI vision, comprising the following steps:

[0006] S1. Operation begins. The intelligent clearing system is powered on and performs a self-check to ensure that each unit is in normal condition.

[0007] S2. The system control system is equipped with multiple sensors to perform a comprehensive scan of the inside of the coal bunker and collect raw data;

[0008] S3, the core AI algorithm fuses and analyzes the collected multimodal data, uses a deep learning model to accurately identify coal accumulation areas and estimate the amount of residual coal, generates a high-precision three-dimensional coal distribution map, and autonomously plans the optimal cleaning path and equipment parameters based on this.

[0009] S4. The control system receives instructions generated by the AI ​​core and drives the cleaning machine to perform automated cleaning operations strictly according to the planned path and parameters.

[0010] S5. Continuously scan and monitor the cleaned area, and send real-time data back to the AI ​​analysis core to determine whether the cleaning effect of the current area has reached the preset standard.

[0011] S6. If the standard is not met, the AI ​​core will immediately generate a new supplementary cleanup instruction for the area that did not meet the standard, and send it to the executing agency again;

[0012] S7. The system automatically generates a detailed report for this task and archives all data for subsequent analysis and model optimization.

[0013] Preferably, the coverage calculation of the multimodal sensor combination in step S2 satisfies the formula: In the formula, For sensors field of view, For resolution, This represents the maximum detection range.

[0014] Preferably, the AI ​​intelligent analysis in step S3 includes the following sub-steps:

[0015] S301. Denoise, correct, and perform multimodal spatiotemporal registration and fusion on the original data to generate a unified information base;

[0016] S302. Input the fused data into the deep learning model and perform semantic segmentation and thickness estimation;

[0017] S303. Combine point cloud data to generate a three-dimensional coal distribution map, quantifying the location, volume and shape of coal accumulation;

[0018] S304. Based on the three-dimensional distribution map and preset strategies, the optimal path is planned through reinforcement learning algorithm, and the robot arm speed and hydraulic pressure parameters are adaptively set.

[0019] Preferably, the thickness estimation model in S302 satisfies the loss function: In the formula, This is the total variational regularization term, ensuring smooth thickness.

[0020] Preferably, the path planning in S304 adopts... The algorithm, with its reward function defined as: In the formula: To address changes in coal quantity, For energy consumption, This refers to the time allotted for the assignment.

[0021] Preferably, the dynamic adjustment mechanism in step S6 includes: real-time detection of residual thickness. ,when When the (preset threshold) is reached, the S3 command is triggered to regenerate.

[0022] Preferably, in step S7, the data archiving is used for online model optimization, and the update rule is as follows: In the formula, For learning rate, This is the loss function.

[0023] Compared with the prior art, the beneficial effects of the present invention are:

[0024] 1. This invention breaks away from the traditional model where cleaning ends once the preset path is completed by dynamically adjusting the judgment function to compare the residual thickness of the cleaning area with a preset threshold in real time. Even if the residual amount exceeds the standard due to changes in the coal accumulation pattern or initial detection errors, if the residual coal amount in a certain area exceeds the standard, the AI ​​core is immediately triggered to re-execute the path planning and parameter settings, generating a supplementary cleaning instruction for that area. This breaks the limitations of traditional one-time operations, achieving real-time feedback and dynamic optimization during the cleaning process, ensuring that the final cleaning effect fully meets the standards, eliminating the need for rework, significantly reducing the cost and time of ineffective operations, and solving the current problem of not being able to adjust in time due to changes in the coal accumulation pattern.

[0025] 2. This invention also leverages the real-time triggering of the dynamic adjustment judgment function, the accurate identification of the thickness estimation loss function, and the efficient planning of the Q-learning reward function. When the dynamic adjustment judgment function triggers the adjustment requirement, the Q-learning reward function quickly replans the cleaning path and equipment parameters. By balancing the changes in the amount of coal cleaned, energy consumption, and time, it generates the optimal supplementary cleaning plan for the substandard area. This eliminates the need to invest additional manpower and resources in developing adjustment strategies, avoiding the passive situation of traditional cleaning methods that cannot detect and adjust in a timely manner, and ensuring that the cleaning effect meets the standard in one go. Attached Figure Description

[0026] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0027] Example: Figure 1 As shown, the present invention relates to a method for detecting and clearing residual coal in a coal bunker based on AI vision, comprising the following steps:

[0028] S1. Operation begins. The intelligent clearing system is powered on and performs a self-check to ensure that each unit is in normal condition.

[0029] S2. The system control unit uses various sensors (high-definition camera, 3D LiDAR / depth camera, multispectral camera, etc.) to perform a comprehensive scan of the inside of the coal bunker and collect raw data.

[0030] The coverage area of ​​data acquisition can be quantified using a sensor model. Let the sensor set be... Coverage area volume Determined by the field of view (FOV) and resolution: ,in, For sensors field of view, For resolution, This is the maximum detection range;

[0031] S3, the core AI algorithm fuses and analyzes the collected multimodal data, uses a deep learning model to accurately identify coal accumulation areas and estimate the amount of residual coal, generates a high-precision three-dimensional coal distribution map, and autonomously plans the optimal cleaning path and equipment parameters based on this.

[0032] In step S3, the raw data needs to be transformed into decision instructions, which also includes the following steps:

[0033] S301. Data Preprocessing and Fusion: The acquired raw images, point clouds, and other data undergo denoising, correction, and registration. Data from different modalities are then aligned and fused spatiotemporally to form a rich information base. Algorithm formula verification: A weighted feature fusion model is used, assuming image data... Point cloud data Spectral data fusion features for: ,in: For adaptive weights (based on environmental parameters such as dust concentration) ,and PCA is a multi-principal component analysis for noise reduction, and ICP is an iterative nearest point algorithm.

[0034] The noise removal rate is quantified using a noise removal rate formula to ensure the quality of the preprocessed data. Image noise removal rate:

[0035] Among them, signal-to-noise ratio The mean of the image signal. (for noise standard deviation), requirements Point cloud noise removal rate:

[0036]

[0037] in, The average distance error between the point cloud and the 3D model of the coal bunker , For point cloud coordinates, Given the coordinates of the corresponding location in the coal bunker model, the following is required: .

[0038] S302, Deep Learning Model Inference: The processed data is input into a pre-trained deep learning model for semantic segmentation and thickness estimation. Algorithm Formula Evidence: Thickness estimation is performed using a U-Net architecture semantic segmentation model and a regression network.

[0039] Semantic segmentation: Output probability: ,in It is a convolutional neural network. These are the model parameters. The loss function used is cross-entropy: Optimized average intersection-union ratio ;

[0040] Thickness estimation: A residual regression network is used to... To input and output coal accumulation thickness, a total variational regularization is introduced into the loss function. Ensure a smooth thickness distribution: Symbol definition: The coal accumulation thickness predicted by the model; The actual thickness was measured manually. for norm , ( (adjacent pixels) (Balance deviation and smoothness).

[0041] S303. Generate 3D distribution map content:

[0042] By combining model output and 3D point cloud data, a high-precision, visualized 3D distribution map of residual coal in the coal bunker is generated, intuitively displaying the location, volume, and shape of the accumulated coal. Algorithm formula verification:

[0043] Point cloud reconstruction formula. Assume the segmentation result... Point cloud coordinates Coal accumulation volume in 3D diagram ,in This is an indicator function (1 if the condition is met, 0 otherwise). Voxel size.

[0044] S304, Intelligent Decision-Making and Planning Content:

[0045] Based on a 3D distribution map and preset cleaning strategies, the system automatically plans the optimal cleaning path for the unmanned cleaning machine. Simultaneously, it adaptively sets the robotic arm's rotation speed, pressure, and other operating parameters according to the coal's shape and thickness. Algorithm formula verification:

[0046] Reinforcement learning Algorithm. Define the state. (Coal quantity distribution), actions (Robotic arm movement), rewards : ,in: To address changes in coal quantity (maximizing the objective). For energy consumption, The optimal path is determined by the task time. The decision is made by leveraging the real-time triggering of the dynamically adjusted judgment function, the accurate identification of the thickness estimation loss function, and... Efficient planning of the reward function, when the dynamic adjustment judgment function triggers an adjustment requirement, The reward function quickly replans the cleaning path and equipment parameters. By balancing changes in the amount of coal cleaned, energy consumption, and time, it generates the optimal supplementary cleaning plan for areas exceeding the standard. There is no need to invest additional manpower and resources to develop adjustment strategies, avoiding the passive situation of traditional cleaning methods that cannot detect and adjust in time, and ensuring that the cleaning effect meets the standard on the first try.

[0047] S4. The control system receives instructions generated by the AI ​​core and drives the cleaning machine to perform automated cleaning operations strictly according to the planned path and parameters.

[0048] The control system drives the cleaning machine (hydraulic pump station, robotic arm) to strictly execute the planned path. Algorithm formula evidence:

[0049] The control model uses PID control: ,in For positional error, This is the PID gain.

[0050] S5. Continuously scan and monitor the cleaned area, and send real-time data back to the AI ​​analysis core to determine whether the cleaning effect of the current area has reached the preset standard.

[0051] S6. If the standard is not met, the AI ​​core will immediately generate a new supplementary cleanup instruction for the area that did not meet the standard, and send it to the executing agency again;

[0052] Real-time scanning after cleaning (S5), AI judges the effect; if it does not meet the standard, the path and parameters are dynamically adjusted (S6). Algorithm formula evidence:

[0053] Validate the threshold model. Set the cleanup criteria. (If the residual thickness is <5cm), the judgment function is:

[0054]

[0055] The adjustment instructions are regenerated using the S304 algorithm.

[0056] By dynamically adjusting the judgment function to compare the residual thickness of the cleaned area with the preset threshold in real time, this method breaks the traditional model where the cleanup ends once the preset path is completed. Even if the residual amount exceeds the standard due to changes in the coal accumulation pattern or initial detection errors, if the residual coal amount in a certain area exceeds the standard, the AI ​​core is immediately triggered to re-execute the path planning and parameter settings, generating supplementary cleanup instructions for that area. This breaks the limitations of traditional one-time operations, enabling real-time feedback and dynamic optimization during the cleanup process. This ensures that the final cleanup effect fully meets the standards, eliminating the need for rework, significantly reducing the cost and time of ineffective operations, and solving the current problem of not being able to adjust in time due to changes in the coal accumulation pattern.

[0057] S7. Once the system determines that the cleaning effect of the entire coal bunker has met the standards, the control equipment returns to its original position, the system automatically generates a detailed report of this operation, and archives all data for subsequent analysis and model optimization.

[0058] Generate a cleanup report (including cleanup volume, time consumption, etc.), and archive the data for continuous model optimization. Algorithm formula evidence:

[0059] Online learning model parameter update formula: ,in This is the learning rate.

[0060] The embodiments disclosed in this invention are preferred embodiments, but are not limited thereto. Those skilled in the art can easily understand the spirit of this invention based on the above embodiments and make different extensions and variations, but as long as they do not depart from the spirit of this invention, they are all within the protection scope of this invention.

Claims

1. An AI vision-based coal bunker residual coal quantity detection and cleaning method, characterized in that, The method comprises the following steps: S1, starting the work, powering on the intelligent clearing system, performing self-checking of the equipment, and ensuring that the state of each unit is normal; S2, the system controls the multiple sensors carried to perform omnidirectional scanning on the inside of the coal bunker and collect original data; S3, the core AI algorithm fuses and analyzes the collected multi-modal data, uses a deep learning model to accurately identify the accumulated coal area and estimate the residual coal quantity, generates a high-precision three-dimensional coal quantity distribution map, and autonomously plans an optimal cleaning path and equipment parameters according to the three-dimensional coal quantity distribution map; S4, the control system receives the instructions generated by the AI core, drives the coal bunker to strictly perform automatic cleaning work according to the planned path and parameters; S5, the cleaned area is continuously scanned and monitored, and real-time data is sent back to the AI analysis core to determine whether the cleaning effect of the current area has reached the preset standard; S6, if the standard is not met, the AI core will immediately generate new supplementary cleaning instructions for the area that does not meet the standard and send them to the execution mechanism again; S7, the system automatically generates a detailed report of this work and archives all data for subsequent analysis and model optimization.

2. The AI vision-based residual coal quantity detection and cleaning method for a coal bunker according to claim 1, characterized in that, The coverage range calculation of the multi-modal sensor combination in the step S2 satisfies the formula: , wherein, is a field of view angle of the sensor , is a resolution, is a maximum detection distance.

3. The AI vision-based residual coal quantity detection and cleaning method of coal bunker according to claim 1, characterized in that, The AI intelligent analysis in the step S3 comprises the following sub-steps: S301, denoising, correcting, and multi-modal space-time registration and fusion of the original data to generate a unified information base; S302, inputting the fused data into a deep learning model to perform semantic segmentation and thickness estimation; S303, generating a three-dimensional coal quantity distribution map in combination with point cloud data to quantify the accumulated coal position, volume, and shape; S304, planning an optimal path through reinforcement learning algorithm based on the three-dimensional distribution map and preset strategy, and adaptively setting the mechanical arm rotating speed and hydraulic pressure parameters.

4. The AI vision-based residual coal quantity detection and cleaning method of a coal bunker according to claim 3, characterized in that, The thickness estimation model in S302 satisfies a loss function: , wherein is a total variation regularization term, ensuring thickness smoothness.

5. The AI vision-based residual coal quantity detection and cleaning method of coal bunker according to claim 3, characterized in that, The path planning in the S304 adopts An algorithm, the reward function is defined as: , wherein: is the change of the amount of cleaned coal, is the energy consumption, is the operation time.

6. The AI vision-based residual coal quantity detection and cleaning method of coal bunker according to claim 1, characterized in that, The dynamic adjustment mechanism in step S6 includes: detecting the residual thickness in real time When , wherein is a preset threshold, triggering S3 to regenerate the instruction.

7. The AI vision-based residual coal quantity detection and cleaning method of coal bunker according to claim 1, characterized in that, The data archiving in the step S7 is used for online model optimization, and the update rule is: , wherein, is a learning rate, is a loss function.

Images