Method for intelligently predicting strength of filling body by combining color and pore structure characteristics

Abstract

The invention provides a method for intelligently predicting the strength of a filling body by combining color and pore structure characteristics, and relates to the technical field of mine filling, the method comprises the following steps: firstly, selecting a filling material for mine production, making an experimental scheme on the basis of the actual proportion of mine production, and preparing and maintaining the filling body; then obtaining a color characteristic value, a pore structure characteristic value and a uniaxial compressive strength value of the filling body; and processing the color characteristic value, the pore structure characteristic value and the uniaxial compressive strength value, establishing a data set, constructing an initial strength prediction model, and performing evaluation and optimization to obtain an optimal strength prediction model.

Description

Technical Field

[0001] The invention relates to the technical field of mine filling, and in particular to an intelligent prediction method for filling body strength combining color and pore structure characteristics. Background Art

[0002] Mine backfilling is an important way to effectively recover mineral resources and achieve sustainable development of the mining industry. In recent years, in order to improve backfilling quality and reduce backfilling costs, many domestic mines have or have begun to use slag-based binders instead of cement to prepare backfills. In backfill mining, in order to ensure production safety, the backfill is required to have high stability, and strength testing is the most direct method to evaluate the stability of the backfill. However, the underground environment is complex, dangerous, and changeable, which makes on-site backfill strength testing face many difficulties, such as difficult sampling and poor sample integrity. Therefore, it is particularly important to seek a convenient, accurate, and reliable on-site slag-based backfill strength testing method.

[0003] The strength of the filling body is mainly affected by two factors: the number of hydration products and the pore structure. Generally, the more hydration products there are, the better the bonding between the particles, and the higher the strength of the filling body. In view of this, some studies predict the strength of the filling body by establishing a relationship between influencing factors and strength. The Chinese patent application CN116306300A discloses an image-based slag-based filling body strength intelligent prediction system and method. The method uses the characteristic that the more hydration products there are in the slag-based filling body, the darker the blue-green color. Through an intelligent algorithm, the color of the filling body is connected with the strength, and then the strength of the filling body is predicted. Image intelligent prediction has no requirements on the size, shape, and integrity of the sample, and is easy to operate. However, this method only considers the influence of the number of hydration products on the filling body strength, and does not consider the fact that the pore structure is also an important factor affecting the strength, which may lead to inaccurate prediction results. The Chinese patent application CN108318395A discloses a method for estimating the strength of a tailings-cemented paste backfill. The method uses a mercury injection test to measure the microscopic pore structure of the sample, and calculates and predicts the uniaxial compressive strength of the backfill based on the mathematical relationship between the uniaxial compressive strength of the backfill and the microscopic pore structure. However, the method has the following two defects: first, it only considers the influence of the pore structure, while ignoring the number of hydration products, which is a key factor affecting the strength; second, the backfill strength is low, and the mercury injection process will cause pore destruction, so the pore distribution result obtained is not a true value.

[0004] The strength of slag-based fillings is mainly affected by the amount of hydration products and the pore structure. On the one hand, the blue-green chromophores produced by slag-based fillings during the hydration process will make their color appear significantly blue-green. The more hydration products there are, the more blue-green chromophores are produced, and the higher the strength of the filling. On the other hand, during the hydration process of the filling, the hydration products will fill the pores. At the same time, under the influence of deadweight, compaction and external loads, the pore structure changes, which in turn affects the strength of the filling. Therefore, the amount of hydration products and the pore structure jointly determine the strength of slag-based fillings. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method for intelligently predicting the strength of a filling body by combining color and pore structure characteristics. By establishing a relationship between the color, pore structure characteristics and strength of a slag-based filling body, a high-precision intelligent prediction model is constructed using a neural network to predict the strength of the filling body.

[0006] The technical solution of the present invention is as follows:

[0007] An intelligent prediction method for filling body strength combining color and pore structure characteristics comprises the following steps:

[0008] Step S1, selecting the filling material produced in the mine, formulating an experimental plan based on the actual proportion of the mine production, and carrying out the preparation and maintenance of the filling body;

[0009] Step S11, selecting a filling material and obtaining the properties of the material itself;

[0010] The filling materials include tailings, slag, activator and water;

[0011] The intrinsic properties include particle size distribution, physicochemical properties, and mineral composition of the material;

[0012] Step S12, determining the influencing factors of the filling body strength, and formulating an experimental plan according to the selected influencing factors;

[0013] The influencing factors include mass concentration, binder content, activator type, activator dosage, curing temperature, self-weight stress, and drainage conditions;

[0014] Step S13, preparing a filling body sample according to the experimental plan and curing it for a set age;

[0015] Step S2, obtaining the color characteristic value, pore structure characteristic value and uniaxial compressive strength value of the filling body;

[0016] Step S21, build an image acquisition device, obtain a cross-sectional image of the filling body, and use MATLAB to extract color feature values;

[0017] The color feature values ​​include grayscale value, hue and saturation of the filling body image;

[0018] The image acquisition device includes a camera, a tripod and a studio; the camera is placed on the tripod, the filling body sample is placed in the studio, and the camera takes a picture of the filling body sample; an OLED light is installed in the studio, and the sample is placed in the same light background each time a picture is taken. During the shooting process, the position and height of the tripod, the distance between the camera and the studio, and the shooting parameters of the camera remain unchanged, and the image taken by the camera is transmitted to a computer for processing;

[0019] Step S22, collecting T2 relaxation time signals through low-field nuclear magnetic resonance experiments, converting the time distribution of the T2 relaxation time signals into pore size distribution through a conversion formula, and obtaining pore structure characteristic values ​​of the filling body;

[0020] The conversion formula is as follows:

[0021]

[0022] Where T2 is the transverse relaxation time; r is the pore radius; ρ is the surface relaxation rate of the pore corresponding to each T2 relaxation time signal; a is the geometric constant;

[0023] The pore structure characteristic values ​​include mesoporosity, macroporosity and total porosity;

[0024] Step S23, measuring the uniaxial compressive strength value of the filling body sample through a uniaxial compression test;

[0025] Step S3, processing the color characteristic values, pore structure characteristic values ​​and uniaxial compressive strength values, establishing a data set, constructing an initial strength prediction model, evaluating and optimizing to obtain an optimal strength prediction model;

[0026] Step S31, normalizing the color characteristic values, pore structure characteristic values ​​and uniaxial compressive strength values, and arranging them into a data set;

[0027] The normalization process is to scale the data to a specified range, and the formula is:

[0028]

[0029] In the formula, X is the original data, X min is the minimum value of the original data, X max is the maximum value of the original data, and x is the normalized data;

[0030] Step S32, dividing the data set into a training set and a test set;

[0031] The division is random division;

[0032] Step S33: using a neural network algorithm to establish an initial strength prediction model on the training set;

[0033] The initial strength prediction model is composed of an input layer, a hidden layer and an output layer. The data of the input layer is transmitted to the neurons of the hidden layer through weights and biases. The neurons between the hidden layers are connected through weights and biases to form a complex network structure. The output signal of the last layer of the hidden layer is transmitted to the neurons of the output layer through weights and biases. The various parts are connected and information is transmitted through weights and biases.

[0034] Step S34, using the initial strength prediction model to perform denormalization on the output data in the test set to obtain a prediction value of the neural network;

[0035] The denormalization process is to restore the normalized data to the original scale, and the formula is:

[0036] X=x·(X max -X min )+X min

[0037] Step S35: compare the predicted value with the true value, calculate the evaluation index, adjust the hyperparameters of the model, and determine the optimal strength prediction model.

[0038] The evaluation indicators include accuracy, F1 score, mean square error MSE, mean absolute error MAE, R 2 Score, mean percentage error MPE.

[0039] The hyperparameters include the learning rate, batch size, number of iterations, regularization parameters, and weight initialization of the model.

[0040] The beneficial effects of adopting the above technical solution are:

[0041] The present invention provides a method for intelligently predicting the strength of a filling body by combining color and pore structure characteristics, which has the following beneficial effects:

[0042] 1. Combining the color characteristics and pore structure characteristics of the filling body, and considering the influence of the amount of hydration products and pore structure on the strength of the filling body, the prediction results are more accurate and the application range is wider;

[0043] 2. Use low-field nuclear magnetic resonance experiment to obtain the pore structure characteristic values ​​of the filling body. Compared with the traditional mercury injection experiment, it will not cause damage to the sample, so the real pore structure characteristic values ​​can be obtained;

[0044] 3. Reducing the input parameters in the prediction model can reduce the workload of data collection, simplify the calculation process, reduce the risk of overfitting, and facilitate model debugging and optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a flow chart of the intelligent prediction method for filling body strength of the present invention;

[0046] Figure 2 A diagram of an initial intensity prediction model constructed on a training set according to a specific embodiment of the present invention;

[0047] Figure 3 A distribution diagram of the predicted value and the true value of the uniaxial compressive strength of the filling body using the optimal strength prediction model in a specific embodiment of the present invention;

[0048] Figure 4 It is a schematic diagram of the detailed steps of a specific implementation mode of the present invention. DETAILED DESCRIPTION

[0049] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0050] An intelligent prediction method for filling strength combining color and pore structure characteristics, such as Figure 1 As shown, the detailed steps in this embodiment are shown in FIG. Figure 4 As shown, the specific steps include:

[0051] Step S1, selecting the filling material produced in the mine, formulating an experimental plan based on the actual proportion of the mine production, and carrying out the preparation and maintenance of the filling body;

[0052] Step S11, selecting a filling material and obtaining the properties of the material itself;

[0053] The filling materials include tailings, slag, activator and water;

[0054] The intrinsic properties include particle size distribution, physicochemical properties, and mineral composition of the material;

[0055] Step S12, determining the influencing factors of the filling body strength, and formulating an experimental plan according to the selected influencing factors;

[0056] The influencing factors include mass concentration, binder content, activator type, activator dosage, curing temperature, self-weight stress, and drainage conditions;

[0057] Step S13, preparing a filling body sample according to the experimental plan and curing it for a set age;

[0058] In this example, tailings were selected as aggregate, slag and activator as binder, and mixing water was added to prepare the filling material. The particle size distribution of tailings and binder was measured using a laser particle size analyzer; the XRD spectrum of tailings and binder was obtained using an X-ray diffractometer; the element type and content of tailings and binder were obtained using an X-ray fluorescence spectrometer, and the mineral composition of the filling material was analyzed by combining the XRD and XRF test results.

[0059] Combined with the material's own properties and the actual production situation of the mine, mass concentration, binder content, activator concentration and curing temperature were selected as influencing factors, and the mass concentration of the filling body was determined to be 68, 70, 72, 74, 76%, the binder content was determined to be 6, 8, 10, 12, 14%, the activator concentration was determined to be 20, 25, 30, 35, 40%, and the curing temperature was determined to be 10, 20, 30, 40 ° C. Similar experiments were carried out in the laboratory to prepare filling body samples with different mass concentrations, binder contents, activator concentrations and curing temperatures, and placed in curing boxes at different temperatures for curing. The curing ages set in this embodiment are 3, 7, 14, 28, and 56 days, respectively.

[0060] Step S2, obtaining the color characteristic value, pore structure characteristic value and uniaxial compressive strength value of the filling body;

[0061] Step S21, build an image acquisition device, obtain a cross-sectional image of the filling body, and use MATLAB to extract color feature values;

[0062] The color feature values ​​include grayscale value, hue and saturation of the filling body image;

[0063] The image acquisition device includes a camera, a tripod and a studio; the camera is placed on the tripod, the filling body sample is placed in the studio, and the camera takes a picture of the filling body sample; an OLED light is installed in the studio, and the sample is placed in the same light background each time a picture is taken. During the shooting process, the position and height of the tripod, the distance between the camera and the studio, and the shooting parameters of the camera remain unchanged, and the image taken by the camera is transmitted to a computer for processing;

[0064] Step S22, collecting T2 relaxation time signals through low-field nuclear magnetic resonance experiments, converting the time distribution of the T2 relaxation time signals into pore size distribution through a conversion formula, and obtaining pore structure characteristic values ​​of the filling body;

[0065] The T2 relaxation time signal corresponds to different pore size distributions. Specifically, a shorter T2 time corresponds to a smaller pore, and a longer T2 time corresponds to a larger pore.

[0066] The conversion formula is as follows:

[0067]

[0068] Where T2 is the transverse relaxation time; r is the pore radius; ρ is the surface relaxation rate of the pore size corresponding to each T2 relaxation time signal, which is a constant and is generally taken as 12 in porous media; a is a geometric constant and is taken as 2 for cylindrical pores.

[0069] The pore structure characteristic values ​​include medium porosity, macroporosity and total porosity. Since small pores have little effect on the strength of the filling body, the present invention does not use small porosity.

[0070] Step S23, measuring the uniaxial compressive strength value of the filling body sample through a uniaxial compression test;

[0071] In order to eliminate the interference of external environmental changes on image acquisition in this embodiment, an image acquisition device is built. The device includes a digital camera, a tripod and a small studio. An OLED lamp is installed in the studio to ensure that the sample is in the same light background every time the photo is taken. The filling body sample that meets the curing age is taken out of the curing box and demoulded, placed in the arranged image acquisition device, and the sample cross-sectional image is obtained by shooting. It should be pointed out that during the sample image shooting process, the position and height of the tripod, the distance between the camera and the studio, and the focal length and aperture of the camera remain unchanged. The original image in the camera is transferred to the computer, and the image is segmented to cut out a rectangular area with a side length of 2000 pixels, which contains a complete cross-sectional image. After the pre-processed image is named, it is imported into the MATLAB software. First, the pre-processed image is read using the imread function, and then the rgb2gray function is used to convert the color image into a grayscale image, and the grayscale value is obtained by accessing the pixel matrix of the grayscale image. Use the rgb2hsv function to convert an RGB image to an HSV image. An HSV image is a three-dimensional array where each dimension corresponds to hue (H), saturation (S), and brightness (V). Extract hue and saturation by indexing this three-dimensional array. Use the hsvImage(:,:,1) function to extract hue, and use the hsvImage(:,:,2) function to extract saturation. The detailed process of extracting color features is as follows: Figure 2 shown.

[0072] The samples were placed in a convection oven at 60°C for 48 hours. During this period, each sample was weighed repeatedly until a constant weight was reached. After drying, all samples were saturated with deionized water in a pressure chamber loaded with 10 MPa for 12 hours and then evacuated. The instrument used to collect T2 relaxation time signals was the PQ001 nuclear magnetic resonance analyzer of Shanghai Newmai Electronic Technology Co., Ltd., with a spectrometer frequency of 12 MHz, a receiver bandwidth of 125 kHz, and a magnetic field strength of 0.52 T. The repetition waiting time was 1000 ms, the 90° pulse length was 7.52 μm, and the 180° pulse length was 15.04 μm. The echo time was 0.3 ms, and all measurements were performed at 32°C. The transverse relaxation spectrum was obtained by SIRT inversion, and the number of iterations and relaxation time points were set to 10,000 and 200, respectively, T 2min and T 2max They were set to 0.001ms and 10000ms respectively. Before the test, the sample was removed from the solvent and wiped with saturated tissue. Subsequently, it was wrapped with plastic film and placed in a sample tube, and the transverse relaxation spectrum was obtained after inversion within 2 minutes. The T2 time distribution was converted into pore size distribution through the conversion formula, and then the mesoporosity, macroporosity and total porosity of the filling body were obtained.

[0073] The samples were subjected to uniaxial compression tests to obtain uniaxial compressive strength. The test instrument used was the American HM-5030 loading tester. The cylindrical specimens of 5cm×5cm×10cm were tested. The loading speed of the press was set to 1mm / min, and the data recording time was 1s. The difference between the peak stress and the average stress of the same group of specimens was considered valid within 10%. Three valid results were measured for each group of specimens, and the average value of the valid results was taken as the strength test result of the group.

[0074] Step S3, processing the color characteristic values, pore structure characteristic values ​​and uniaxial compressive strength values, establishing a data set, constructing an initial strength prediction model, evaluating and optimizing to obtain an optimal strength prediction model;

[0075] Step S31, normalizing the color characteristic values, pore structure characteristic values ​​and uniaxial compressive strength values, and arranging them into a data set;

[0076] The normalization process is to scale the data to a specified range, usually [0,1], and the formula is:

[0077]

[0078] In the formula, X is the original data, X min is the minimum value of the original data, X max is the maximum value of the original data, and x is the normalized data;

[0079] Step S32, dividing the data set into a training set and a test set;

[0080] The division is random division;

[0081] Step S33: using a neural network algorithm to establish an initial strength prediction model on the training set;

[0082] The initial strength prediction model is composed of an input layer, a hidden layer and an output layer. The data of the input layer is transmitted to the neurons of the hidden layer through weights and biases. The neurons between the hidden layers are connected through weights and biases to form a complex network structure. The output signal of the last layer of the hidden layer is transmitted to the neurons of the output layer through weights and biases. The various parts are connected and information is transmitted through weights and biases.

[0083] Step S34, using the initial strength prediction model to perform denormalization on the output data in the test set to obtain a prediction value of the neural network;

[0084] The denormalization process is to restore the normalized data to the original scale, and the formula is:

[0085] X=x·(X max -X min )+X min

[0086] Step S35: compare the predicted value with the true value, calculate the evaluation index, adjust the hyperparameters of the model, and determine the optimal strength prediction model.

[0087] The evaluation indicators include accuracy, F1 score, mean square error MSE, mean absolute error MAE, R 2 Score, mean percentage error MPE.

[0088] The hyperparameters include the learning rate, batch size, number of iterations, regularization parameters, and weight initialization of the model.

[0089] This example uses the random partitioning method to divide the entire data set into a training set and a test set. The ratio of the two is determined by the convergence test. The training set accounts for 80% of the total data set, and the test set accounts for 20% of the total data set. Gray value (X0), hue (X1), saturation (X2), medium porosity (X3), macroporosity (X4), and total porosity (X5) are the six input variable parameters of the input layer, and the uniaxial compressive strength of the filling body (Y0) is the output parameter of the output layer. The number of hidden layer units is selected according to the following empirical formula:

[0090]

[0091] Where n is the number of hidden layer units, m is the number of input layer units, l is the number of output layer units, and a is a constant between 1 and 10.

[0092] The number of hidden layer units was determined to be in the range of [4, 14]. The number of hidden layer units was 4 as the starting point. Data was randomly selected from the training set for training. Finally, it was determined that when the number of hidden layer units was 9, the prediction result was the best. The initial intensity prediction model was obtained, as shown in Figure 3 shown.

[0093] The experimental data in the test set are input into the initial strength prediction model, and the output data is processed by denormalization to obtain the predicted value. The mean square error (MSE) is used as the standard for judging the prediction accuracy. The calculation formula of the mean square error (MSE) is as follows:

[0094]

[0095] Where n is the number of samples, y i is the true value of the i-th sample, is the predicted value of the ith sample. The smaller the MSE value, the more accurate the prediction.

[0096] According to the evaluation results, the model hyperparameters, such as learning rate, batch size, etc., are adjusted, and training and evaluation are performed again until the prediction accuracy meets the requirements and the optimal strength prediction model is obtained. The distribution results of the predicted value and the true value of the uniaxial compressive strength of the filling body using the optimal strength prediction model are shown in Figure 2. Figure 3 As shown in the figure, after calculation, the mean square error between the predicted value and the true value is 0.0021, which shows that the predicted value is very close to the true value. The prediction accuracy of the strength prediction model is very high and can be used for the strength prediction of slag-based filling bodies.

[0097] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) to form a technical solution.

Claims

1. An intelligent prediction method for filling body strength combining color and pore structure characteristics, characterized in that: The following steps are involved: Step S1, selecting the filling material produced in the mine, formulating an experimental plan based on the actual proportion of the mine production, and carrying out the preparation and maintenance of the filling body; Step S2, obtaining the color characteristic value, pore structure characteristic value and uniaxial compressive strength value of the filling body; Step S3: Process the color characteristic values, pore structure characteristic values ​​and uniaxial compressive strength values, establish a data set, build an initial strength prediction model, evaluate and optimize to obtain the optimal strength prediction model.

2. The method for intelligent prediction of filling body strength combining color and pore structure characteristics according to claim 1, characterized in that: The step S1 comprises the following steps: Step S11, selecting a filling material and obtaining the properties of the material itself; Step S12, determining the influencing factors of the filling body strength, and formulating an experimental plan according to the selected influencing factors; Step S13: prepare the filling body sample according to the experimental plan and cure it for a set age.

3. The method for intelligent prediction of filling body strength combining color and pore structure characteristics according to claim 2 is characterized in that: The filling materials in step S11 include tailings, slag, activator and water; the inherent properties include particle size distribution, physical and chemical properties, and mineral composition of the material; The influencing factors described in step S12 include mass concentration, binder content, activator type, activator dosage, curing temperature, self-weight stress, and drainage conditions.

4. The method for intelligent prediction of filling body strength combining color and pore structure characteristics according to claim 1 is characterized in that: The step S2 comprises the following steps: Step S21, build an image acquisition device, obtain a cross-sectional image of the filling body, and use MATLAB to extract color feature values; The color feature values ​​include grayscale value, hue and saturation of the filling body image; The image acquisition device includes a camera, a tripod and a studio; the camera is placed on the tripod, the filling body sample is placed in the studio, and the camera takes a picture of the filling body sample; an OLED light is installed in the studio, and the sample is placed in the same light background each time a picture is taken. During the shooting process, the position and height of the tripod, the distance between the camera and the studio, and the shooting parameters of the camera remain unchanged, and the image taken by the camera is transmitted to a computer for processing; Step S22, collecting T2 relaxation time signals through low-field nuclear magnetic resonance experiments, converting the time distribution of the T2 relaxation time signals into pore size distribution through a conversion formula, and obtaining pore structure characteristic values ​​of the filling body; Step S23: measuring the uniaxial compressive strength value of the filling body sample through a uniaxial compression test.

5. The method for intelligent prediction of filling body strength combining color and pore structure characteristics according to claim 4 is characterized in that: The conversion formula in step S22 is as follows: Where T2 is the transverse relaxation time; r is the pore radius; ρ is the surface relaxation rate of the pore corresponding to each T2 relaxation time signal; a is the geometric constant; The pore structure characteristic values ​​include mesoporosity, macroporosity and total porosity.

6. The method for intelligent prediction of filling body strength combining color and pore structure characteristics according to claim 1, characterized in that: The step S3 specifically comprises the following steps: Step S31, normalizing the color characteristic values, pore structure characteristic values ​​and uniaxial compressive strength values, and arranging them into a data set; The normalization process is to scale the data to a specified range, and the formula is: In the formula, X is the original data, X min is the minimum value of the original data, X max is the maximum value of the original data, and x is the normalized data; Step S32, dividing the data set into a training set and a test set; The division is random division; Step S33: using a neural network algorithm to establish an initial strength prediction model on the training set; Step S34, using the initial strength prediction model to perform denormalization on the output data in the test set to obtain a prediction value of the neural network; The denormalization process is to restore the normalized data to the original scale, and the formula is: X=x·(X max -X min )+X min Step S35, comparing the predicted value with the true value, calculating the evaluation index, adjusting the hyperparameters of the model, and determining the optimal strength prediction model; The evaluation indicators include accuracy, F1 score, mean square error MSE, mean absolute error MAE, R 2 Score, mean percentage error MPE; The hyperparameters include the learning rate, batch size, number of iterations, regularization parameters, and weight initialization of the model.

7. The method for intelligent prediction of filling body strength combining color and pore structure characteristics according to claim 6, characterized in that: The initial intensity prediction model described in step S33 is composed of an input layer, a hidden layer and an output layer. The data of the input layer is passed to the neurons of the hidden layer through weights and biases. The neurons between the hidden layers are connected through weights and biases to form a complex network structure. The output signal of the last layer of the hidden layer is passed to the neurons of the output layer through weights and biases. The various parts are connected and information is transmitted through weights and biases.

Images