A method for analyzing and predicting the stability of coal piles
By constructing a coal reactor stability analysis method based on laser scanning and neural networks, the difficulties of coal reactor stability monitoring and prediction are solved, real-time security assessment and optimization suggestions are realized, and computing and storage costs are reduced.
Patent Information
- Application Number
- CN202210338147.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-01
- Publication Date
- 2025-07-22
- Estimated Expiration
- 2042-04-01
AI Technical Summary
The prior art is difficult to effectively monitor and predict the stability changes of coal piles under interference situations, resulting in safety hazards and difficulty in optimizing layout.
By establishing simulated coal stacks, using laser scanners to obtain three-dimensional point cloud data, perform feature point marking and dimensionality reduction processing, build an artificial neural network model, monitor and predict the stability of coal stacks in real time, and provide safety and stability assessment and early warning.
It realizes rapid stability level prediction of coal yards, reduces calculation and storage costs, provides real-time safety assessment and optimization suggestions, and avoids safety accidents.
Smart Images
Figure CN114707721B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method for analyzing and predicting the stability of a coal pile, which is an engineering analysis method. It is a method for monitoring the continuously changing bulk materials during the stacking and reclaiming processes of bulk material piles, and analyzing and predicting the changes of the bulk materials under interference. Background Art
[0002] After coal is mined, it usually needs to be stored and transferred in a storage yard. In a coal storage yard, the stacking volume of coal is usually thousands of tons or even tens of thousands of tons, and the stacking height is more than ten meters or even higher. The coal storage yard is one of the objects with the most frequent dynamic updates in the industrial scenario. Mastering its mechanical stability characteristics is of great significance for applications such as automated reclaiming operations and safety stability analysis of the storage yard. Due to its frequent change characteristics, a laissez-faire attitude is usually taken towards the shape of the coal pile, which causes difficulties in optimizing the layout of the stacking yard and also poses potential safety hazards. Therefore, the stability analysis of the coal pile and the prediction of its changes are problems that need to be solved. Summary of the Invention
[0003] To overcome the problems of the prior art, the present invention proposes a method for analyzing and predicting the stability of a coal pile. The method monitors the shape of the changing material in real time through detection means, collects the three-dimensional point cloud data of the coal pile, and analyzes and predicts the stability of the coal pile through a model constructed by neural network training, providing data support for automated operations in the coal storage yard and avoiding safety accidents.
[0004] The object of the present invention is achieved as follows: A method for analyzing and predicting the stability of a coal pile includes the following steps:
[0005] The method includes a model construction process and an evaluation process:
[0006] The model construction process includes:
[0007] Step 1, establish a simulated coal pile: Stack a simulated coal pile similar to the actual shipping process at the coal pile site, and use a laser scanner to scan the simulated coal pile to construct the three-dimensional point cloud of the simulated coal pile and obtain the data characteristics of the three-dimensional point cloud data; the data characteristics include: curvature, normal vector, feature point coordinates, and the positional relationship between adjacent points.
[0008] Step 2, feature point marking: Mark feature points based on the point cloud curvature.
[0009] Step 3, streamline feature points: Use the uniform grid method to streamline non-feature points.
[0010] Step 4, PCA dimensionality reduction: Use PCA to reduce the dimensionality of the data characteristics.
[0011] Step 5, Model evaluation and training: Train an artificial neural network with the features and stability levels of the three-dimensional point cloud data after dimensionality reduction to obtain a mechanical stability analysis and prediction model for coal piles;
[0012] The analysis training process is as follows: The simulated coal pile is perturbed according to the set disturbing force. The magnitudes of the disturbing forces are divided into F 1, F 2, …… F n , a total of 1, 2, ……, n , n levels, and the perturbation increases gradually from the force with a smaller value to the force with a larger value; When under the action of the F i force, the sliding amount of the simulated coal pile reaches the preset value Q i , then mark the stability level of the current simulated coal pile as the level of the current disturbing force i ; Scan the simulated coal pile after each perturbation at least once to obtain the three-dimensional point clouds under the action of forces at each level F i , and after repeating the data processing in steps 2 - 4, input them into the artificial neural network; Establish a sample database to record the three-dimensional point cloud data samples during the training process;
[0013] The prediction training process is as follows: The prepared training samples are the point cloud features of the current simulated coal and the point cloud features of the coal pile after the action of the disturbing force corresponding to the stability level i of the current simulated coal pile F i ; During the training process, the input is the point cloud feature of the current simulated coal pile, and the output is the point cloud feature of the coal pile after the action of the disturbing force corresponding to the stability level F i of the current simulated coal pile;
[0014] The described evaluation process includes:
[0015] Step 6, Obtain the point cloud data of the coal pile: Continuously scan the coal pile using a laser scanner, monitor the changes in the coal pile, and continuously obtain the three-dimensional point cloud data of the coal pile;
[0016] Step 7, Evaluate the stability of the coal pile: Input the three-dimensional point cloud data features of the obtained coal pile into the trained mechanical stability analysis and prediction model of the coal pile to analyze the stability of the current coal pile and predict the point cloud features of the coal pile after the action of the disturbing force i corresponding to the stability level F i of the current coal pile;
[0017] Step 8, Stability Alarm: To provide a reference for the safety and stability assessment of coal piles and to perform manual intervention when the stability is poor;
[0018] Step 9, Recording: Compare the measurement data of each coal pile with the three-dimensional point cloud data samples in the database, and selectively retain the measurement data with large differences from the samples to enrich the three-dimensional point cloud data sample database.
[0019] The advantages and beneficial effects of the present invention are as follows: According to the characteristics of frequent changes in coal yards, non-contact measurement sensors are installed to scan the coal yards to form three-dimensional point cloud data. Since the amount of three-dimensional point cloud data is very large, various methods are used to simplify the three-dimensional point cloud data and reduce the dimension of features, so as to realize predicting the stability level of the rapidly changing coal yard with limited computing resources, providing prediction data in real time, and providing a basis for optimizing the coal yard and avoiding safety accidents. Brief Description of the Drawings
[0020] The present invention will be further described below in conjunction with the drawings and embodiments.
[0021] Figure 1 is a flowchart of the method described in the embodiment of the present invention;
[0022] Figure 2 is a schematic diagram of the mechanical stability assessment of the three-layer artificial neural network described in the embodiment of the present invention. Detailed Embodiment
[0023] Embodiment:
[0024] This embodiment is a method for analyzing and predicting the stability of a coal yard. The method performs a stability assessment on the commercial coal piles stacked in the coal yard.
[0025] Characteristics of the commercial coal yard: The coal piles change frequently, the stacking site environment is relatively stable, and there are not only outdoor stacks but also indoor stacks. Since the coal has been carefully screened, the coal varieties within a pile of commercial coal are the same, that is, the particle sizes are uniform and the humidity is relatively stable. The same variety makes the fluidity of the material relatively consistent. Therefore, assuming that the fluidity of the material is consistent for a specific coal pile, in order to accurately express the mechanical characteristics of the stack, a certain coefficient can be set to adjust the fluidity.
[0026] In a coal yard, various mechanical equipment is used to transport coal into and out of the stacking area, such as belt conveyors, scraper conveyors, bucket machines, raking machines, etc. Force sensors can be installed on these mechanical equipment to detect the disturbing forces generated on the coal pile. In this embodiment, as long as the disturbing forces generated by these devices can be quantitatively measured, they are all considered as the disturbing forces that cause changes in the coal pile. In this simple way, various forms of disturbing forces are simplified to adapt to the mechanical analysis and prediction of the stability of the coal pile. In other words, regardless of what kind of mechanical equipment it is, only the magnitude of the disturbing force it generates is concerned, and it doesn't matter what mechanical equipment generates the disturbing force, so as to eliminate many variable factors caused by different mechanical equipment.
[0027] Based on the above two assumptions, the complex engineering system of the coal yard is standardized into a simple system that can be controlled by a practical mathematical model. Practice has proved that the two assumptions are in line with the reality of the coal yard, and the model constructed based on these assumptions can also achieve the expected evaluation goals.
[0028] The hardware system used in the described method includes: multiple laser scanners installed in the coal yard. These laser scanners should be able to ensure real-time monitoring of each coal pile and obtain the point cloud data of each coal pile without omission. The laser scanners are connected to at least one server with the ability to store a large amount of data. Since the amount of data in engineering calculations is usually extremely large, the number of servers used may be relatively large, and even a server cluster may be required. The amount of data stored is also extremely large, and sometimes large storage devices such as disk arrays and high-end transmission networks such as SUN networks are needed. In order to reduce costs, various means are used to reduce the amount of data and the costs of calculation and storage.
[0029] Another problem that needs to be solved for realizing the monitoring of the coal yard is: the contour point cloud data of a coal pile (there are multiple coal piles in the coal yard) is extremely large, about tens of thousands or even hundreds of thousands, and as the coal pile keeps changing, these point cloud data are also constantly changing and must be processed at any time. The large amount of point cloud data acquisition requires the above-mentioned huge computing resources for processing, and the on-site computing resources cannot meet the requirement of quickly processing so much data. Of course, cloud computing can be used, but its computing cost is relatively high and it is difficult to implement. Therefore, after the point cloud data is collected, it is necessary to refine the point cloud data obtained on-site and perform dimensionality reduction processing on the features to reduce the amount of calculation. The specific way of refining the point cloud data in this embodiment is: feature point marking, uniform grid reduction, and PCA dimensionality reduction processing. For the continuous changes of the coal pile, a method similar to image compression can be adopted, that is, only the parts that are disturbed and changed are processed, and the parts that have not changed are ignored. Through these processing means, the local limited computing resources can process the huge amount of point cloud data.
[0030] It is also necessary to establish a database. On the one hand, it stores the samples during the perturbation experiment. As the experiment progresses, new data can be continuously added to the model training to optimize the model. On the other hand, it stores the evaluation results to guide the current material taking operation, such as how much disturbing force needs to be applied, etc., and can also provide data support for the current stability assessment of the coal pile.
[0031] The specific steps of the method in this embodiment are as follows (the process is as Figure 1 shown):
[0032] The method described includes a model construction process and an evaluation process: that is, first train the model, and then use the trained model to evaluate the actual coal pile.
[0033] The model construction process includes:
[0034] Step 1, establish a simulated coal pile: Stack a simulated coal pile similar to the actual shipping process at the coal pile site, and use a laser scanner to scan the simulated coal pile to construct a three-dimensional point cloud of the simulated coal pile and obtain the data characteristics of the three-dimensional point cloud data. The data characteristics of the three-dimensional point cloud data include: curvature, normal vector, characteristic point coordinates, and the positional relationship between adjacent points.
[0035] At the beginning of model training, a simulated coal pile is first established. The so-called simulated coal pile is actually a specific coal pile in the stacking yard, and a sample experiment of applying force interference to it is carried out, and the obtained samples are used to train the neural network model. At the beginning of sample acquisition, first determine the magnitude and change pattern of the disturbing force according to the characteristics of each coal pile in the current coal yard (including on-site conditions such as the relative position to the laser scanner, the stacking shape of the coal pile, and the transportation machinery used). Then enter the three-dimensional point cloud sample acquisition process, use a laser scanner to scan the coal pile to obtain the three-dimensional point cloud data of the coal pile, and perform point cloud denoising processing on the original data to obtain clear three-dimensional point cloud data.
[0036] Step 2, mark characteristic points: Mark characteristic points based on the point cloud curvature.
[0037] The process of marking the point cloud characteristic points is as follows: Define a curvature threshold, and mark the points with curvature greater than the curvature threshold as characteristic points. In theory, the non-characteristic points with curvature values less than the curvature threshold can be streamlined, but when the curvature threshold is set relatively high, there will be holes on relatively flat surfaces. Therefore, the corresponding points cannot be completely removed on relatively flat surfaces.
[0038] Step 3, streamline characteristic points: Use the uniform grid method to streamline non-characteristic points.
[0039] Uniform grid method: Divide the coordinate plane xoy into grids of the same size, then project the three-dimensional point cloud onto the xoy plane and distribute the point cloud into the corresponding grids. For all points within a grid, a z’ value is calculated as the z coordinate according to the set rules, and the x’ and y’ coordinates of the grid center point are used to replace all points assigned to this cell. Finally, all points within the grid are replaced by one point (x’, y’, z’).
[0040] Step 4, PCA dimensionality reduction: Use PCA to reduce the dimensionality of data features.
[0041] Use PCA to reduce the dimensionality of the data features of the three-dimensional point cloud. The features include curvature, normal vector, feature point coordinates, the positional relationship between adjacent points, etc. The specific reduction of dimensions also needs to be determined according to the actual data situation.
[0042] Step 5, model evaluation and training: Use the features and stability levels of the three-dimensional point cloud data after dimensionality reduction to train an artificial neural network to obtain a mechanical stability analysis and prediction model for coal piles.
[0043] The analysis training process is as follows: The simulated coal pile is disturbed according to the set disturbing forces. The magnitudes of the disturbing forces are divided into F 1,[[]]END]] F 2, …… F n , a total of 1, 2, ……, n , n levels, and the disturbance increases gradually from the force with a smaller value to the force with a larger value. When under the action of the F i force, the sliding amount of the simulated coal pile reaches the preset value Q i , then mark the stability level of the current simulated coal pile as the level of the current disturbing force i . Scan the simulated coal pile after each disturbance at least once to obtain the three-dimensional point cloud under the action of forces at each level. After repeating the data processing in steps 2 - 4, input it into the artificial neural network to establish a sample database and record the three-dimensional point cloud data samples during the training process. F i The prediction training process is as follows: The prepared training samples are the point cloud features of the current simulated coal and the point cloud features of the coal pile after the action of the disturbing force corresponding to the stability level of the current simulated coal pile
[0044] . During the training process, the input is the point cloud feature of the current simulated coal pile, and the output is the point cloud feature of the coal pile after the action of the disturbing force corresponding to the stability level of the current simulated coal pile i corresponding disturbing force F i After the action. F i After the action of the coal pile.
[0045] The evaluation and training of the mechanical stability analysis and prediction model for coal piles includes two aspects. One is stability analysis, and the other is stability prediction. The former is to input disturbances at multiple levels to generate multiple point cloud data, making the point cloud data correspond to the disturbances. The latter is to predict the changes in the point cloud data based on the disturbances, that is, the changes in the coal pile under disturbances at each level.
[0046] The slippage of the coal pile is caused by the movement behavior of the currently used mechanical equipment. The level of the disturbing force is determined by the amount of slippage of the coal pile under its action Q decided. Specifically, how much slippage is counted as one level can be determined according to the actual requirements on site.
[0047] A three-layer artificial neural network model can be used (as Figure 2 shown): At the input end of the three-layer neural network model, the dimension-reduced point feature cloud data is input, mainly including curvature, normal vector, feature point coordinates, and the positional relationship between adjacent points. Figure 2 In it, the left column represents the input features, and the right column is the output.
[0048] The operation process of the three-layer artificial neural network is as follows: Under the action of the disturbing force F i , when the amount of slippage is greater than or equal to Q i , the stability level and the point cloud data at this time are a set of samples. Through continuous experiments, multiple sets of samples are obtained to train the model. The "mechanical stability" at the output end of the model refers to the force Q i required when the slippage amount of the coal pile reaches the preset value F i of size.
[0049] Step 6, obtain the point cloud data of the coal pile: Use a laser scanner to continuously scan the coal pile, monitor the changes in the coal pile, and continuously obtain the three-dimensional point cloud data of the coal pile.
[0050] The so-called continuous scanning refers to the continuous contact with the coal pile. In fact, a certain sampling step is required during the sampling process. The sampling step can be calculated in minutes. The sampling step is related to the efficiency of the mechanical equipment at the coal pile site. The working efficiency of the mechanical equipment is usually calculated in minutes.
[0051] Step 7, evaluate the stability of the coal pile: Input the three-dimensional point cloud data features of the obtained coal pile into the trained mechanical stability analysis and prediction model of the coal pile, analyze the current stability of the coal pile, and predict the point cloud features of the coal pile after the action of the disturbing force i corresponding to the stability level F i of the coal pile.
[0052] This step of evaluating the stability of the coal pile includes two aspects: the stability analysis of the current material pile and the stability prediction. The former analyzes the stability according to the point cloud of the current pile, and the latter analyzes the changes generated in the disturbed coal piles of each level by inputting the point cloud data of the current coal pile.
[0053] The process of analysis: The input to the model is the point cloud characteristics of the current coal pile, and the output is the corresponding stability level of the current coal pile. i 。
[0054] The process of prediction: The input to the model is the point cloud characteristics of the current coal pile, and the output is the disturbing force corresponding to the stability level of the coal pile. i The corresponding disturbing force F i The point cloud characteristics of the coal pile after the action.
[0055] Step 8, stability alarm: Provide a reference for the safety and stability evaluation of the coal pile and conduct human intervention when the stability is poor. That is, input a small force F i to achieve the pre-set sliding amount Q.
[0056] Step 9, recording: Compare the evaluation data of each coal pile with the three-dimensional point cloud data samples in the database, and selectively retain the evaluation data with large differences from the samples to enrich the three-dimensional point cloud data sample database.
[0057] Finally, it should be noted that the above is only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred arrangement, those of ordinary skill in the art should understand that the technical solution of the present invention (such as the system used for calculation, the application of the neural network model, the sequence of steps, etc.) can be modified or equivalently replaced without departing from the spirit and scope of the technical solution of the present invention.
Claims
1. A method for analyzing and predicting the stability of coal piles, characterized in that, The method described above includes the following steps: The method described above includes a model construction process and an evaluation process: The model construction process includes: Step 1, establish a simulated coal pile: Stack a simulated coal pile similar to the actual shipping process at the coal pile site, and use a laser scanner to scan the simulated coal pile to construct a three-dimensional point cloud of the simulated coal pile, and obtain the data characteristics of the three-dimensional point cloud data; the data characteristics of the three-dimensional point cloud data include: curvature, normal vector, feature point coordinates, and the positional relationship between adjacent points. Step 2, feature point marking: Mark feature points based on the point cloud curvature; the process of marking point cloud feature points is as follows: Define a curvature threshold, and mark the points with curvature greater than the curvature threshold as feature points. When the curvature threshold is set relatively high, to avoid holes on relatively flat surfaces, the corresponding points cannot be completely removed on relatively flat surfaces. Step 3, streamline feature points: Use the uniform grid method to streamline non-feature points; the uniform grid method: Divide the coordinate plane xoy into grids of the same size, then project the three-dimensional point cloud onto the xoy plane, allocate the point cloud to the corresponding grids, calculate a z’ value as the z coordinate for all points in the grid according to the set rules, use the x’, y’ coordinates of the grid center point to replace all points allocated to this cell, and finally all points in the grid are replaced by one point (x’, y’, z’). Step 4, PCA dimensionality reduction: Use PCA to perform dimensionality reduction on the data characteristics. Step 5, model evaluation and training: Train an artificial neural network with the characteristics and stability levels of the dimensionality-reduced three-dimensional point cloud data to obtain a mechanical stability analysis and prediction model for coal piles. The analysis training process is: disturb the simulated coal pile according to the set disturbance force, and the magnitude of the disturbance force is divided into F 1, F 2,…… F n , a total of 1, 2, ..., n , n The disturbance increases gradually from a smaller force to a larger force. F i Under the action of force, the sliding amount of the simulated coal pile reaches the preset value Q i When the stability level of the current simulated coal pile is marked as the level of the current disturbance force i ; Scan the simulated coal pile at least once after each disturbance to obtain the force at each level. F i The three-dimensional point cloud under the action is input into the artificial neural network after repeating the data processing of steps 2-4; a sample database is established to record the three-dimensional point cloud data samples in the training process; The prediction training process is as follows: The prepared training samples are the point cloud features of the currently simulated coal and the point cloud features of the currently simulated coal pile after the corresponding disturbing force i acts on it; F i During the training process, the input is the point cloud features of the currently simulated coal pile, and the output is the point cloud features of the currently simulated coal pile after the corresponding disturbing force F i acts on it; The evaluation process described above includes: Step 6, obtain coal pile point cloud data: Continuously scan the coal pile using a laser scanner, monitor the changes in the coal pile, and continuously obtain the three-dimensional point cloud data of the coal pile. Step 7, evaluate the stability of the coal pile: Input the three-dimensional point cloud data features of the obtained coal pile into the trained mechanical stability analysis and prediction model of the coal pile, analyze the current stability of the coal pile, and predict the point cloud features of the coal pile after the action of the disturbing force corresponding to the current stability level i corresponding disturbing force F i acting on the coal pile; Step 8, stability alarm: Provide a reference for the safety and stability evaluation of the coal pile, and perform human intervention when the stability is poor. Step 9, record: Compare the evaluation data of each coal pile with the three-dimensional point cloud data samples in the database, and selectively retain the evaluation data with relatively large differences from the samples to enrich the three-dimensional point cloud data sample database.