CN108256258A - Cemented fill mechanical response characteristic Forecasting Methodology based on SEM image - Google Patents

Abstract

The invention discloses a method for predicting the mechanical response characteristics of a cemented filling body based on an SEM image, which comprises the steps of: 1. making a SEM scanning electron microscope sample; 2. scanning to form a SEM scanning electron microscope scanning image and storing it in a computer; The scanned image is processed by Gaussian filtering; 4. Obtain multiple cluster images of cemented filling bodies; 5. Determine the microscopic pore map of the cemented filling body, and obtain the binary image of the microscopic pores of the cemented filling body; 6. Scan the SEM electron microscope after Gaussian filtering The image is merged with the microscopic pore binary image of the cemented filling body to obtain the test sample image; 7. Normalize the test sample image; 8. Input the normalized test sample image into the pre-built Tensorflow deep learning mechanical response prediction network to obtain Uniaxial mechanical response prediction results. The invention has high prediction efficiency, consumes less manpower and material resources, and has great significance for studying the strength and stability of cemented filling bodies.

Description

Prediction method of mechanical response characteristics of cemented filling body based on SEM image

technical field

The invention belongs to the technical field of cemented filling mining, and in particular relates to a method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images.

Background technique

With the development of national science and technology, the requirements for energy saving and environmental protection technology are getting higher and higher. Traditional cemented backfill mining uses cement as the cementitious material, and the cost of cement is as high as 75% of the total backfill cost. Through research and development, tailings contain active silica and alumina, and using tailings instead of part of cement as cementing material can not only reduce the discharge of tailings, effectively reduce the cost of filling mining, but also improve the strength of the filling body and reduce the ground The collapsed area also plays an active role in promoting the protection of the environment. Therefore, the tailings discharged from the concentrator gradually become the main aggregate of cemented filling in mines. The cemented filling body is the core content of the cemented filling mining method, which involves mine safety and mine economic benefits. Filling body mechanics, with the content of "filling body mechanism, filling body strength, reasonable matching of filling body and mechanical response characteristics of filling body", has been highly valued by the mining industry for several years. In recent years, eight international academic conferences on filling have been held, and great progress has been made in many aspects of filling body mechanics. The academic circles generally believe that the mechanical properties of filling body are the key factors that seriously affect and restrict the cemented filling mining method.

The use of tailings to achieve different water-cement ratios and different curing ages have a direct impact on the mechanical properties of the cemented backfill. As one of the most commonly used filling aggregates for filling gobs, tailings not only solve the shortage of filling aggregates, but also have a low dilution rate and a large loss rate when mining pillars of extremely thick ore bodies, and the "three down" resource mining is safe It provides an effective way to solve the problems of low ground pressure and difficult ground pressure control in deep rock mass. Many researchers have done in-depth research on the composition ratio, stabilization process and mechanical strength of tailings paste filling. For example, Kesimal A et al. studied the relationship between deslimed copper-lead-zinc tailings and paste strength, and found that the particle size distribution of tailings has a great influence on the strength of cemented filling bodies; The effect of desliming bysedimentation on paste backfill performance was published on "Minerals Engineering"; Fall et al. studied the effect of curing temperature on the strength of tailings cemented backfill In 2010, the journal "Engineering Geology" (Engineering Geology), Issue 4, Volume 10, published an article A Contribution to understanding the effects of curing temperature on the mechanical properties of minecemented tailings backfill. The contribution of mechanical properties); G Xiu et al. used different proportions of tailings cementation strength and different concentrations to conduct experiments in the laboratory to reveal the chemical reaction mechanism of tailings in the microscopic aspect and the influence of the macroscopic size on the stability of the filling body. ; Published the article Microstructure Test and Macro Size Effect on the Stability of Cemented Tailings Backfill (Microstructure Test and Macro Size Effect on the Stability of Cemented Tailings Backfill) in the journal "International Journal of Digital Content Technology & Its Applications" (Digital Content Technology and Its Applications) in the 6th Volume 14 of 2012 The influence of macroscopic size on the stability of cement-tailed sand filling); Li Wenchen et al. studied the influence of sulfate on the relationship between uniaxial compressive strength and elastic modulus of cemented filling through mortar experiments; in the first issue of 2016 The 42nd volume of the journal "China Coal" published an article "Research on the Effect of Sulfate on the Relationship between Uniaxial Compressive Strength and Elastic Modulus of Cemented Filling Body". However, in the existing technology, the prediction of the mechanical response characteristics of cemented filling bodies is mostly based on experimental testing methods. The test period is long, the efficiency is low, and the cost of manpower and material resources is high, which affects the rapid promotion and application of new cemented filling bodies. Delay in construction period.

Contents of the invention

The technical problem to be solved by the present invention is to provide a method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images, which has simple steps, novel and reasonable design, convenient and fast implementation, and high prediction efficiency. The cycle is short, and it consumes less manpower and material resources. It is of great significance for the study of the strength and stability of the cemented filling body. It has strong practicability, a wide range of applications, and high promotion and application value.

In order to solve the above-mentioned technical problems, the technical solution adopted in the present invention is: a method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images, characterized in that the method includes the following steps:

Step 1, taking a part of the cemented filling body sample to make a SEM scanning electron microscope sample;

Step 2, using SEM scanning electron microscope to scan the SEM scanning electron microscope sample, forming a SEM scanning electron microscope scanning image and storing it in the computer;

Step 3, the computer calls the Gaussian filter processing module to perform Gaussian filter processing on the SEM electron microscope scanned image, and obtains the SEM electron microscope scanned image after the Gaussian filter process;

Step 4, the computer invokes the FCM fuzzy clustering processing module to extract the pore image from the Gaussian filter-processed SEM electron microscope scanning image, and obtain a plurality of cemented filling body clustering images equal to the number of clustering centers;

Step 5. The computer determines the cluster image of the cemented filling body with the smallest gray value as the microscopic pore map of the cemented filling body, and performs binary processing on the microscopic pore map of the cemented filling body to obtain the microscopic pore size of the cemented filling body. value map;

Step 6, the computer merges the Gaussian filter-processed SEM scanning image obtained in step 3 with the binary image of the microscopic pores of the cemented filling body obtained in step 5 to obtain a test sample image;

Step 7, the computer normalizes the test sample image to form a normalized test sample image with pixels of 960×960;

Step 8: The computer inputs the normalized test sample image obtained in step 7 into the pre-built Tensorflow deep learning mechanical response prediction network to obtain a uniaxial mechanical response prediction result.

The above method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images is characterized in that: the length, width and height of the SEM scanning electron microscope sample in step 1 are all 10 mm.

The above method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images is characterized in that: the computer described in step 3 calls the Gaussian filter processing module to perform Gaussian filter processing on the SEM electron microscope scanning image. The formula used is L(x, y)= I(x,y)*G(x,y), where I(x,y) represents the SEM scanning image, G(x,y) is the Gaussian filter function, and L(x,y) is the Gaussian filter The SEM electron microscope scanning image, x is the abscissa of the image, and y is the ordinate of the image.

The above method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images is characterized in that: the computer calls the FCM fuzzy clustering processing module in step 4 to extract the pore images from the SEM electron microscope scanning images after the Gaussian filter processing, and obtain the same as The specific process of clustering images of multiple cemented fillings with the same number of cluster centers is as follows:

Step 401, define the FCM fuzzy clustering algorithm based on sample weighting, and the objective function is The constraint condition for satisfying the extremum value is Among them, U is a fuzzy matrix and U=[u ₁₁ ,u ₂₂ ,…,u _cn ], u _ik is an element of matrix U and u _ik represents the membership degree of the kth sample point belonging to the i-th class, and n is the sample point The total number, c is the number of cluster centers; V={v ₁ ,v ₂ ,...v _c } is the cluster centers of c classes, w _k is the weight of sample point x _k , d _ik is the sample point x The Euclidean distance from _k to the center point v _i , v _i is the element of V, x _k is the kth sample point of the sample set X and X={x ₁ ,x ₂ ,...x _n }, m is the degree of membership The weight index of u _ik and m>1;

Step 402, setting the value of the number of cluster centers c, the value of the weight index m of the degree of membership u _ik and the value of the minimum iteration error ε;

Step 403, use the formula Update the weight w _k of the sample point x _k ; u _τj is an element of the matrix U and u _τj represents the membership degree of the jth sample point belonging to the τth class, 1≤τ≤c, 1≤j≤n; v _τj is The element of V; u _ij is the element of matrix U and u _ij represents the membership degree of the j-th sample point belonging to the i-th class;

Step 404, use the formula Update u _ik ; among them, d _rk is the Euclidean distance from the sample point x _k to the center point v _r , 1≤r≤c;

Step 405, use the formula update v _i ;

Step 406. Judging whether ||J ^(t+1) -J(t)||<ε is satisfied, and when ||J ^(t+1) -J(t)||<ε is satisfied, the clustering stops and extraction Obtain multiple cluster images of cemented filling bodies equal to the number of cluster centers; otherwise, return to step 403; wherein, t is time.

The above method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images is characterized in that: in step 402, the value of the number c of cluster centers is set to 4, the value of the weight index m of the degree of membership u _ik is set to 2, and the minimum iteration The value of the error ε is 0.3.

The above method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images is characterized in that: the construction method of the Tensorflow deep learning mechanical response characteristic prediction network described in step 8 is:

Step 801, take a part of each cemented filling body sample after multiple numbers to make a SEM scanning electron microscope sample, and the remaining part is used as a uniaxial compressive strength test sample; and for multiple SEM scanning electron microscope samples and multiple compressive strength samples The test samples are numbered one by one;

Step 802: Use the uniaxial compressive strength test device for the cemented filling body to perform uniaxial compressive strength tests on multiple compressive strength test samples respectively, and take the measured uniaxial compressive strength of the multiple compressive strength test samples as The average value is obtained to obtain the uniaxial compressive strength of the cemented filling body sample;

Step 803, obtaining the training sample image of the Tensorflow deep learning mechanical response characteristic prediction network, the specific process is:

Step 8031, using the SEM scanning electron microscope to scan multiple SEM scanning electron microscope samples respectively to form multiple SEM scanning electron microscope scanning images and store them in the computer; the number of the SEM scanning electron microscope scanning images is at least 150;

Step 8032, the computer invokes the Gaussian filter processing module to perform Gaussian filter processing on a plurality of SEM electron microscope scanned images respectively, to obtain a plurality of Gaussian filtered SEM electron microscope scanned images;

Step 8033, the computer invokes the FCM fuzzy clustering processing module to extract pore images from the Gaussian-filtered SEM electron microscope scanning images to obtain multiple groups of cemented filling cluster images, each group of cemented filling cluster images The number of cluster images of medium cemented filling body is equal to the number of cluster centers;

Step 8034, the computer determines the cemented backfill cluster image with the smallest gray value in each group of cemented backfill cluster images as the microscopic pore map of the cemented backfill, and double-checks the multiple cemented backfill microscopic pore maps Value-based processing to obtain multiple microscopic pore binary maps of cemented filling bodies;

Step 8035, the computer combines the multiple Gaussian-filtered SEM scanning images obtained in step 8032 with the multiple binary images of microscopic pores of the cemented filling body obtained in step 8034 according to the corresponding numbers to obtain multiple training samples image;

Step 804, the computer performs normalization processing on a plurality of training sample images respectively to form a plurality of normalized training sample images with a pixel size of 960×960;

Step 805, the computer constructs a Tensorflow deep learning network with five layers of convolutional network cores, an input layer of normalized training sample images, and an output layer of uniaxial compressive strength corresponding to the normalized training sample images. A plurality of normalized training sample images stored in it are used as training samples to train the Tensorflow deep learning network to obtain the Tensorflow deep learning mechanical response characteristic prediction network; the Tensorflow deep learning mechanical characteristic response prediction network is the size of the five-layer convolutional network kernel From the first floor to the fifth floor are 3x3, 2x2, 3x3, 2x2, 2x2 respectively.

The above method for predicting mechanical response characteristics of cemented filling body based on SEM images is characterized in that: the uniaxial compressive strength test device for cemented filling body described in step 802 includes a seat cushion and a plurality of tie rods fixedly connected to the top of the seat cushion, and An axial pressure force transmission mechanism for applying axial pressure to the cemented filling body sample and an axial pressure power system for providing power to the axial pressure force transmission mechanism; the bottom of the seat cushion is fixedly connected with A plurality of bases, the top of the seat cushion is provided with a sample placement groove for placing the cemented filling body sample, and the seat cushion is provided with a drain valve at the center of the sample placement groove; the plurality of pull rods The middle part is provided with a fixed frame for fixing multiple pull rods, and the upper part of the multiple pull rods is fixedly connected with a top loading plate; the axial pressure force transmission mechanism includes a cylinder installed on the top loading plate, and the piston rod of the cylinder is directed toward the The bottom of the piston rod of the cylinder is connected with a pressure transmission plate; the axial pressurization power system includes a compressed air source and a pressurization controller, and a device with one end connected to the compressed air source and the other end connected to the cylinder Gas delivery pipe; the gas delivery pipe is sequentially provided with a pneumatic triple piece, a pressure sensor and a cylinder control solenoid valve from the position connecting the compressed air source to the connection cylinder, and the pressure sensor is connected to the input end of the pressurization controller, The cylinder control solenoid valve is connected with the output end of the pressurization controller, and the pressurization controller is connected with the computer through the communication module.

The method for predicting the mechanical response characteristics of the cemented filling body based on the SEM image is characterized in that: the sample placement groove is provided with an O-ring set on the bottom of the cemented filling body sample, and the seat cushion is provided with a The sample is placed in a porous stone around the groove.

The above method for predicting mechanical response characteristics of cemented filling bodies based on SEM images is characterized in that: the pressurization controller is a programmable logic controller, and the communication module is an RS-485 communication module.

The above method for predicting mechanical response characteristics of cemented filling body based on SEM images is characterized in that: in step 802, the uniaxial compressive strength test device for cemented filling body is used to test the uniaxial compressive strength of multiple compressive strength test samples respectively. Test, wherein the specific process of uniaxial compressive strength test for each compressive strength test sample is:

Step 8021. After putting the O-ring into the sample placement groove, put the cemented filling body sample into the sample placement groove, so that the center of the cemented filling body sample is in contact with the piston rod and pressure of the cylinder. Corresponding to the center of the transfer plate; and put porous stones located around the sample placement groove on the seat cushion;

Step 8022, turn on the compressed air source, adjust the air pressure of the compressed air output by the compressed air source by adjusting the pneumatic triple unit, the pressurization controller controls the reversing of the solenoid valve by controlling the cylinder, and controls the piston rod of the cylinder to move downward or upward, Apply pressure or unload pressure to the cemented filling body sample, record the pressure value detected by the pressure sensor collected by the pressure controller when the cemented filling body sample ruptures as F, and the pressure controller transmits the pressure value F to the computer, and the computer According to the formula Calculate the uniaxial compressive strength P of the compressive strength test sample; where, S is the top surface area of the compressive strength test sample; when the piston rod of the cylinder moves downward, it drives the pressure transmission plate to move downward, through the pressure transmission The plate exerts pressure on the cemented filling body sample. When the piston rod of the cylinder moves upward, it drives the pressure transmission plate to move upward, and the pressure transmission plate leaves the upper surface of the cemented filling body sample to unload the pressure.

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

1. The present invention uses scanning electron microscope (Scanning Electronic Microscopy, SEM) to scan and collect sample images, uses FCM fuzzy clustering processing method to extract the microscopic pore map of cemented filling body, and then uses Tensorflow deep learning network to establish the relationship between the image and the mechanical response characteristics The end-to-end prediction model is used to predict the mechanical response characteristics of the cemented filling body. The method has simple steps, novel and reasonable design, convenient and quick implementation, high prediction efficiency, short cycle time, and less manpower and material resources.

2. Before using the FCM fuzzy clustering processing method to extract the microscopic pore image of the cemented filling body, the present invention also uses the Gaussian filter method to perform Gaussian filter processing on the SEM electron microscope scanning image, which helps to obtain more accurate prediction results.

3. The present invention adopts the FCM fuzzy clustering processing method to extract the microscopic pore image of the cemented filling body, which can avoid different influences of different sample vectors in the sample space on the clustering results.

4. The present invention adopts the method of merging the Gaussian filter-processed SEM electron microscope scanning image with the cemented filling body microscopic pore binary image to obtain the test sample image as the input of the Tensorflow deep learning mechanical response prediction network, which can avoid the original SEM electron microscope scanning image The impact of redundant information in the medium improves the prediction accuracy of mechanical response characteristics.

5. The present invention uses the Tensorflow deep learning network to construct the relationship between the SEM electron microscope scanning image and the mechanical response characteristics of the cemented filling body, and constructs the Tensorflow deep learning mechanical response prediction network once, which can be used conveniently and quickly for many times, so that the cemented filling can be carried out When predicting the mechanical response characteristics of the body, there is no need to do experimental tests many times, just collect the scanning images of the SEM electron microscope into the computer, and the whole process of predicting the mechanical response characteristics can be automatically completed, which is convenient and fast.

6. The uniaxial compressive strength is an important parameter reflecting the mechanical properties of the filling body, and it can reflect the strength and stability of the filling body to a certain extent; when the present invention constructs the Tensorflow deep learning mechanical response prediction network, the output is the uniaxial compressive strength It is of great significance to study the strength and stability of cemented filling bodies by using the uniaxial compressive strength as the prediction result of uniaxial mechanical response.

7. The present invention adopts the self-developed and manufactured uniaxial compressive strength test device for the cemented filling body to test the compressive strength test sample. The uniaxial compressive strength test device for the cemented filling body has a simple structure and is easy to implement and operate , and can measure the accurate uniaxial compressive strength.

8. The mechanical response characteristics studied by the present invention are important characteristics of the cemented backfill, and the cemented backfill is the core content of the cemented backfill mining method. Therefore, the method of the present invention can not only contribute to the study of new cemented backfills, but also contribute to the reduction of Discharge of tailings, reduce the cost of filling mining, protect the environment, increase the ore recovery rate, alleviate the high temperature of deep wells, optimize the environment of the mining area, and control the surface subsidence; the present invention has strong practicability, wide application range, and high application value .

In summary, the method of the present invention has simple steps, novel and reasonable design, convenient and quick implementation, high prediction efficiency, short cycle, and less manpower and material resources. The application range is wide, and the promotion and application value is high.

The technical solutions of the present invention will be described in further detail below with reference to the accompanying drawings and embodiments.

Description of drawings

Fig. 1 is a flow chart of the method of the present invention.

Fig. 2 is a SEM electron microscope scanning image after Gaussian filter processing in a specific embodiment of the present invention.

Fig. 3A is a bright cemented filling body clustering image obtained by performing FCM fuzzy clustering processing on the SEM electron microscope scanning image after Gaussian filtering processing in a specific embodiment of the present invention.

Fig. 3B is a brighter cemented filling cluster image obtained by performing FCM fuzzy clustering on the Gaussian-filtered SEM scanning image in a specific embodiment of the present invention.

Fig. 3C is a cluster image of dark cemented filling bodies obtained by performing FCM fuzzy clustering on the SEM scanning image after Gaussian filtering in a specific embodiment of the present invention.

Fig. 3D is the darkest cemented filling body clustering image obtained by performing FCM fuzzy clustering processing on the SEM electron microscope scanning image after Gaussian filtering processing in a specific embodiment of the present invention.

Fig. 4 is a binary map of microscopic pores of a cemented filling body in a specific embodiment of the present invention.

Fig. 5 is a test sample image in a specific embodiment of the present invention.

Fig. 6 is a schematic structural view of the uniaxial compressive strength testing device of the cemented filling body of the present invention.

Explanation of reference signs:

1—gas delivery pipe; 2—cylinder; 3—pressure transmission plate;

4—compressed air source; 5—drain valve; 6—O-ring;

7—porous stone; 8—tie rod; 9—top loading plate;

10—seat cushion; 11—sample placement slot; 12—pneumatic triple piece;

13—pressure sensor; 14—cylinder control solenoid valve; 15—base;

16—communication module; 17—computer; 18—pressurization controller;

19—Consolidated filling body sample.

Detailed ways

As shown in Figure 1, the method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images of the present invention includes the following steps:

Step 1, taking a part from the cemented filling body sample 19 to make a SEM scanning electron microscope sample;

During specific implementation, the SEM scanning electron microscope sample was also subjected to multiple carbon spraying treatments.

In this embodiment, the length, width and height of the SEM scanning electron microscope sample in step 1 are all 10 mm.

Step 2, using SEM scanning electron microscope to scan the SEM scanning electron microscope sample, forming a SEM scanning electron microscope scanning image and storing it in the computer 17;

Step 3, the computer 17 invokes the Gaussian filter processing module to perform Gaussian filter processing on the SEM electron microscope scanned image, and obtains the SEM electron microscope scanned image after the Gaussian filter process;

In this embodiment, the computer 17 in step 3 invokes the Gaussian filter processing module to perform Gaussian filter processing on the SEM electron microscope scanning image using the formula L(x, y)=I(x, y)*G(x, y) , wherein, I(x, y) represents the SEM electron microscope scanning image, G(x, y) is the Gaussian filter function, L(x, y) is the SEM electron microscope scanning image after the Gaussian filter processing, and x is the abscissa of the image, y is the vertical coordinate of the image.

In this embodiment, the SEM electron microscope scanning image after Gaussian filter processing is shown in FIG. 2 .

Step 4, the computer 17 invokes the FCM fuzzy clustering processing module to extract pore images from the Gaussian filter-processed SEM electron microscope scanning image, and obtain a plurality of cemented filling body cluster images equal to the number of cluster centers;

In this embodiment, the computer 17 in step 4 invokes the FCM fuzzy clustering processing module to extract pore images from the Gaussian filter-processed SEM electron microscope scanning image, and obtain multiple cemented filling body clusters equal to the number of cluster centers The specific process of the image is:

Step 401, define the FCM fuzzy clustering algorithm based on sample weighting, and the objective function is The constraint condition for satisfying the extremum value is Among them, U is a fuzzy matrix and U=[u ₁₁ ,u ₂₂ ,…,u _cn ], u _ik is an element of matrix U and u _ik represents the membership degree of the kth sample point belonging to the i-th class, and n is the sample point The total number (the gray value corresponding to each coordinate point in the image), c is the number of cluster centers (classified according to the brightness of the image in the image); V={v ₁ ,v ₂ ,...v _c } is c The cluster center of each class, w _k is the weight of the sample point x _k , d _ik is the Euclidean distance from the sample point x _k to the center point v _i , v _i is the element of V, x _k is the kth of the sample set X sample points and X={x ₁ ,x ₂ ,...x _n }, m is the weight index of the membership degree u _ik and m>1; the FCM fuzzy clustering algorithm based on sample weighting can avoid the The different influences of different sample vectors on the clustering results;

Step 402, setting the value of the number of cluster centers c, the value of the weight index m of the degree of membership u _ik and the value of the minimum iteration error ε;

In this embodiment, in step 402, the value of the number c of cluster centers is set to 4, the value of the weight index m of the degree of membership u _ik is set to 2, and the value of the minimum iteration error ε is set to 0.3.

Step 403, use the formula Update the weight w _k of the sample point x _k ; u _τj is an element of the matrix U and u _τj represents the membership degree of the jth sample point belonging to the τth class, 1≤τ≤c, 1≤j≤n; v _τj is The element of V; u _ij is the element of matrix U and u _ij represents the membership degree of the j-th sample point belonging to the i-th class;

Step 404, use the formula Update u _ik ; among them, d _rk is the Euclidean distance from the sample point x _k to the center point v _r , 1≤r≤c;

Step 405, use the formula update v _i ;

Step 406. Judging whether ||J ^(t+1) -J(t)||<ε is satisfied, and when ||J ^(t+1) -J(t)||<ε is satisfied, the clustering stops and extraction Obtain multiple cluster images of cemented filling bodies equal to the number of cluster centers; otherwise, return to step 403; wherein, t is time.

In this embodiment, the value of the number c of cluster centers is 4, and four clustered images of cemented backfills are obtained, as shown in Fig. 3A to Fig. 3D, which are bright, brighter, dark and darkest four cemented backfills respectively Cluster images.

Step 5, the computer 17 determines the cemented filling body cluster image with the smallest gray value (the darkest cemented filling body cluster image) as the cemented filling body microscopic pore map, and performs the microscopic pore map of the cemented filling body Binary processing to obtain the binary map of the microscopic pores of the cemented filling body;

In this embodiment, the cluster image of the cemented filling body with the smallest gray value is shown in Figure 3D, that is, Figure 3D is the microscopic pore map of the cemented filling body, and the obtained binary image of the microscopic pores of the cemented filling body is shown in Figure 4 .

Step 6, the computer 17 merges the Gaussian filter processed SEM scanning image obtained in step 3 with the binary image of the microscopic pores of the cemented filling body obtained in step 5 to obtain a test sample image; since the original SEM scanning image There is a lot of redundant information in , that is, it is generally considered that the black area is a pore part. Due to the influence of different factors during scanning, there may be situations where the degree of lightness and darkness of the scan are different. Therefore, the present invention processes the Gaussian filter obtained in step 3 The final SEM electron microscope scanning image is merged with the microscopic pore binary image of the cemented filling body obtained in step 5 to obtain a test sample image, which can avoid the influence of redundant information and improve the prediction accuracy of uniaxial mechanical response characteristics.

For example, as shown in FIG. 5 , it is the test sample image obtained in this embodiment.

Step 7, the computer 17 normalizes the test sample image to form a normalized test sample image with pixels of 960×960;

Step 8. The computer 17 inputs the normalized test sample image obtained in step 7 into the pre-built Tensorflow deep learning mechanical response prediction network to obtain a uniaxial mechanical response prediction result.

In this embodiment, the construction method of the Tensorflow deep learning mechanical response characteristic prediction network described in step eight is:

Step 801, take a part of each cemented filling body sample 19 after multiple numbers to make a SEM scanning electron microscope sample, and the remaining part is used as a uniaxial compressive strength test sample; and multiple SEM scanning electron microscope samples and multiple compression The strength test samples are numbered one by one; for example, the numbers of multiple cemented filling body samples 19 are 1, 2, ..., N, and the numbers of multiple SEM scanning electron microscope samples are A1, A2, ..., AN, and multiple The numbers of the uniaxial compressive strength test samples are B1, B2, ..., BN;

During specific implementation, the SEM scanning electron microscope sample was also subjected to multiple carbon spraying treatments.

Step 802: Use the uniaxial compressive strength test device for the cemented filling body to perform uniaxial compressive strength tests on multiple compressive strength test samples respectively, and take the measured uniaxial compressive strength of the multiple compressive strength test samples as The mean value is obtained to obtain the uniaxial compressive strength of the cemented filling body sample 19;

In this embodiment, as shown in FIG. 6, the uniaxial compressive strength test device for the cemented filling body described in step 802 includes a seat cushion 10 and a plurality of pull rods 8 fixedly connected to the top of the seat cushion 10, and is used to give the cemented filling The body sample 19 applies an axial pressure force transmission mechanism for axial pressure and an axial pressure power system for providing power to the axial pressure force transmission mechanism; the bottom of the seat cushion 10 is fixedly connected with a plurality of bases 15. The top of the seat cushion 10 is provided with a sample placement groove for placing the cemented filling body sample 19, and the seat cushion 10 is provided with a drain valve 5 at the center of the sample placement groove; The middle part of the pull rods 8 is provided with a fixed frame 11 for fixing multiple pull rods 8, and the upper part of the multiple pull rods 8 is fixedly connected with a top loading plate 9; the axial pressure transmission mechanism includes a cylinder installed on the top loading plate 9 2. The piston rod of the cylinder 2 is set downwards, and the bottom of the piston rod of the cylinder 2 is connected with a pressure transmission plate 3; the axial pressurization power system includes a compressed air source 4 and a pressurization controller 18, and One end is connected to the compressed air source 4, and the other end is connected to the gas cylinder 2; the gas delivery pipe 1 is sequentially provided with a pneumatic triple piece 12, a pressure The sensor 13 and the cylinder control solenoid valve 14, the pressure sensor 13 is connected to the input end of the pressurization controller 18, the cylinder control solenoid valve 14 is connected to the output end of the pressurization controller 18, and the pressurization controller 18 It is connected with the computer 17 through the communication module 16 . During specific implementation, the pressure transmission plate 3 is made of rubber. The pressure transmission plate 3 is made of rubber. On the one hand, it can distribute the pressure transmitted by the piston rod of the cylinder 2, so that the pressure is more evenly applied to the top of the cemented filling body sample 19; on the other hand, the pressure transmission plate 3 transmits pressure to the cemented filling body. When the filling body sample 19 is put on, the top surface of the cemented filling body sample 19 will not be damaged.

In this embodiment, the O-shaped sealing ring 6 set on the bottom of the cemented filling body sample 19 is arranged in the sample placement groove, and the porous stone 7 located around the sample placement groove is arranged on the seat cushion 10. . By providing the O-ring 6 , it is possible to prevent damage to the cemented filling body sample 19 caused by hard contact between the cemented filling body sample 19 and the seat cushion 10 when axial pressure is applied to the cemented filling body sample 19 . By providing the porous stone 7, the water seeped from the cemented filling body sample 19 can be absorbed.

In this embodiment, the pressurization controller 18 is a programmable logic controller, and the communication module 16 is an RS-485 communication module.

In this embodiment, as described in step 802, the uniaxial compressive strength test device for the cemented filling body is used to perform the uniaxial compressive strength test on a plurality of compressive strength test samples respectively, wherein the uniaxial compressive strength test is performed on each compressive strength test sample. The specific process of compressive strength test is as follows:

Step 8021, after putting the O-ring 6 into the sample placement groove, put the cemented filling body sample 19 into the sample placement groove, so that the center of the cemented filling body sample 19 is aligned with the center of the cylinder 2 The piston rod corresponds to the center of the pressure transmission plate 3; and the porous stone 7 positioned around the sample placement groove is placed on the seat cushion 10;

Step 8022, open the compressed air source 4, adjust the air pressure of the compressed air output by the compressed air source 4 by adjusting the pneumatic triple unit 12, and the pressurization controller 18 controls the reversing of the solenoid valve 14 by controlling the cylinder to control the piston rod of the cylinder 2 Move downward or upward, apply pressure to the cemented filling body sample 19 or unload the pressure, record the pressure value detected by the pressure sensor 13 collected by the pressure controller 18 when the cemented filling body sample 19 ruptures as F, pressurization control The device 18 transmits the pressure value F to the computer 17, and the computer 17 according to the formula Calculate the uniaxial compressive strength P of the compressive strength test sample; where, S is the top surface area of the compressive strength test sample; when the piston rod of the cylinder 2 moves downward, it drives the pressure transmission plate 3 to move downward, through The pressure transmission plate 3 applies pressure to the cemented filling body sample 19. When the piston rod of the cylinder 2 moves upward, it drives the pressure transmission plate 3 to move upward, and the pressure transmission plate 3 leaves the upper surface of the cemented filling body sample 19 to unload the pressure.

Step 803, obtaining the training sample image of the Tensorflow deep learning mechanical response characteristic prediction network, the specific process is:

Step 8031, using the SEM scanning electron microscope to scan multiple SEM scanning electron microscope samples respectively, forming multiple SEM scanning electron microscope scanning images and storing them in the computer 17; the number of the SEM scanning electron microscope scanning images is at least 150;

Step 8032, the computer 17 invokes the Gaussian filter processing module to perform Gaussian filter processing on a plurality of SEM electron microscope scanned images respectively, to obtain a plurality of Gaussian filtered SEM electron microscope scanned images;

In this embodiment, the computer 17 calls the Gaussian filter processing module to perform Gaussian filter processing on a plurality of SEM electron microscope scanning images respectively. The formula adopted is L(x, y)=I(x, y)*G(x, y) , wherein, I(x, y) represents the SEM electron microscope scanning image, G(x, y) is the Gaussian filter function, L(x, y) is the SEM electron microscope scanning image after the Gaussian filter processing, and x is the abscissa of the image, y is the vertical coordinate of the image.

Step 8033, the computer 17 invokes the FCM fuzzy clustering processing module to extract pore images from a plurality of SEM electron microscope scanning images after Gaussian filter processing, and obtain multiple groups of cemented filling cluster images, each group of cemented filling clusters The number of cluster images of cemented filling bodies in the image is equal to the number of cluster centers;

In this embodiment, the specific process of the computer 17 invoking the FCM fuzzy clustering processing module to extract the pore image from the Gaussian filtered SEM image is the same as step 4.

Step 8034, the computer 17 determines the cemented backfill cluster image with the smallest gray value (the darkest cemented backfill cluster image) in each group of cemented backfill cluster images as the cemented backfill microscopic pore map, And carry out binarization processing on the microscopic pore maps of multiple cemented filling bodies, and obtain the microscopic pore binary maps of multiple cemented filling bodies;

Step 8035, the computer 17 merges the multiple Gaussian-filtered SEM scanning images obtained in step 8032 with the multiple binary images of the microscopic pores of the cemented filling body obtained in step 8034 according to the corresponding numbers to obtain multiple training sample image;

Step 804, the computer 17 respectively performs normalization processing on a plurality of training sample images to form a plurality of normalized training sample images with a pixel size of 960×960;

During specific implementation, when the training sample image is larger than the pixel 960×960 to be obtained by normalization, the image is scaled down to obtain a normalized training sample image; when the training sample image is smaller than the pixel 960×960 to be obtained by normalization, The edge of the image is expanded with white, and the normalized training sample image is obtained.

Step 805, the computer 17 constructs a Tensorflow deep learning network with five layers of convolutional network core layers, an input layer of normalized training sample images, and an output layer of uniaxial compressive strength corresponding to the normalized training sample images, A plurality of normalized training sample images stored in it are used as training samples, and the Tensorflow deep learning network is trained to obtain a Tensorflow deep learning mechanical response characteristic prediction network; the Tensorflow deep learning mechanical characteristic response prediction network has five layers of convolutional network cores The sizes from the first floor to the fifth floor are 3x3, 2x2, 3x3, 2x2, 2x2 respectively.

In summary, the present invention uses a scanning electron microscope to scan and collect sample images, uses a Gaussian filter method to perform Gaussian filter processing on the SEM electron microscope scanned image, uses FCM fuzzy clustering processing method to extract the microscopic pore map of the cemented filling body, and performs Gaussian filter processing. The test sample image is obtained by merging the SEM electron microscope scanning image and the microscopic pore binary image of the cemented filling body, which is used as the input of the Tensorflow deep learning mechanical response prediction network, and then the Tensorflow deep learning network is used to establish the end-to-end relationship between the image and the mechanical response characteristics. The end prediction model is used to predict the mechanical response characteristics of the cemented filling body. The method has simple steps, novel and reasonable design, convenient and quick implementation, high prediction efficiency, short cycle time, and less manpower and material resources.

The above are only preferred embodiments of the present invention, and do not limit the present invention in any way. All simple modifications, changes and equivalent structural changes made to the above embodiments according to the technical essence of the present invention still belong to the technical aspects of the present invention. within the scope of protection of the scheme.

Claims (10)

1. A method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images, characterized in that the method may further comprise the steps: Step 1, taking a part from the cemented filling body sample (19) to make a SEM scanning electron microscope sample; Step 2, using SEM scanning electron microscope to scan the SEM scanning electron microscope sample, forming a SEM scanning electron microscope scanning image and storing it in the computer (17); Step 3, the computer (17) calls the Gaussian filter processing module to carry out Gaussian filter processing to the SEM electron microscope scanned image, and obtains the SEM electron microscope scanned image after the Gaussian filter process; Step 4, the computer (17) invokes the FCM fuzzy clustering processing module to extract the pore image from the Gaussian filter-processed SEM electron microscope scanning image, and obtain a plurality of cemented filling body cluster images equal to the number of cluster centers; Step 5. The computer (17) determines the cluster image of the cemented filling body with the smallest gray value as the microscopic pore map of the cemented filling body, and performs binary processing on the microscopic pore map of the cemented filling body to obtain the cemented filling body Microscopic pore binary map; Step 6, the computer (17) merges the Gaussian filter processed SEM scanning image obtained in step 3 with the binary image of the microscopic pores of the cemented filling body obtained in step 5 to obtain a test sample image; Step 7, the computer (17) normalizes the test sample image to form a normalized test sample image with pixels of 960×960; Step 8. The computer (17) inputs the normalized test sample image obtained in step 7 into the pre-built Tensorflow deep learning mechanical response prediction network to obtain a uniaxial mechanical response prediction result. 2. The method for predicting mechanical response characteristics of cemented filling bodies based on SEM images according to claim 1, wherein the length, width and height of the SEM scanning electron microscope sample in step 1 are all 10 mm. 3. according to the method for predicting the mechanical response characteristic of cemented filling body based on SEM image according to claim 1, it is characterized in that: described in step 3, computer (17) transfers Gaussian filter processing module to carry out Gaussian filter process to SEM electron microscope scanning image using The formula is L(x,y)=I(x,y)*G(x,y), wherein, I(x,y) represents the SEM electron microscope scanning image, G(x,y) is the Gaussian filter function, L (x, y) is the SEM scanning image after Gaussian filtering, x is the abscissa of the image, and y is the ordinate of the image. 4. according to the method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images according to claim 1, it is characterized in that: the computer (17) described in step 4 calls the FCM fuzzy clustering processing module to carry out the SEM after Gaussian filter processing The specific process of extracting the pore image from the scanning electron microscope image to obtain multiple cemented filling body cluster images equal to the number of cluster centers is as follows: Step 401, define the FCM fuzzy clustering algorithm based on sample weighting, and the objective function is The constraint condition for satisfying the extremum value is Among them, U is a fuzzy matrix and U=[u ₁₁ ,u ₂₂ ,…,u _cn ], u _ik is an element of matrix U and u _ik represents the membership degree of the kth sample point belonging to the i-th class, and n is the sample point The total number, c is the number of cluster centers; V={v ₁ ,v ₂ ,...v _c } is the cluster centers of c classes, w _k is the weight of sample point x _k , d _ik is the sample point x The Euclidean distance from _k to the center point v _i , v _i is the element of V, x _k is the kth sample point of the sample set X and X={x ₁ ,x ₂ ,...x _n }, m is the degree of membership The weight index of u _ik and m>1; Step 402, setting the value of the number of cluster centers c, the value of the weight index m of the degree of membership u _ik and the value of the minimum iteration error ε; Step 403, use the formula Update the weight w _k of the sample point x _k ; u _τj is an element of the matrix U and u _τj represents the membership degree of the jth sample point belonging to the τth class, 1≤τ≤c, 1≤j≤n; v _τj is The element of V; u _ij is the element of matrix U and u _ij represents the membership degree of the j-th sample point belonging to the i-th class; Step 404, use the formula Update u _ik ; among them, d _rk is the Euclidean distance from the sample point x _k to the center point v _r , 1≤r≤c; Step 405, use the formula update v _i ; Step 406. Judging whether ||J ^(t+1) -J(t)||<ε is satisfied, and when ||J ^(t+1) -J(t)||<ε is satisfied, the clustering stops and extraction Obtain multiple cluster images of cemented filling bodies equal to the number of cluster centers; otherwise, return to step 403; wherein, t is time. 5. according to the method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images according to claim 4, it is characterized in that: in step 402, the value of the number c of cluster centers is set to 4, and the weight index m of the degree of membership u _ik is set The value is 2, and the value of the minimum iteration error ε is set to 0.3. 6. according to the method for predicting mechanical response characteristics of cemented filling body based on SEM image according to claim 1, it is characterized in that: the construction method of Tensorflow deep learning mechanical response characteristic prediction network described in step 8 is: Step 801, take a part from each cemented filling body sample (19) after multiple numbers to make a SEM scanning electron microscope sample, and the remaining part is used as a uniaxial compressive strength test sample; and multiple SEM scanning electron microscope samples and multiple The compressive strength test samples correspond to the numbers one by one; Step 802: Use the uniaxial compressive strength test device for the cemented filling body to perform uniaxial compressive strength tests on multiple compressive strength test samples respectively, and take the measured uniaxial compressive strength of the multiple compressive strength test samples as The mean value is obtained to obtain the uniaxial compressive strength of the cemented filling body sample (19); Step 803, obtaining the training sample image of the Tensorflow deep learning mechanical response characteristic prediction network, the specific process is: Step 8031, using the SEM scanning electron microscope to scan multiple SEM scanning electron microscope samples respectively, forming multiple SEM scanning electron microscope scanning images and storing them in the computer (17); the number of the SEM scanning electron microscope scanning images is at least 150; Step 8032, the computer (17) calls the Gaussian filter processing module to perform Gaussian filter processing on a plurality of SEM electron microscope scanning images respectively, and obtains a plurality of SEM electron microscope scanning images after Gaussian filter processing; Step 8033, the computer (17) invokes the FCM fuzzy clustering processing module to extract pore images from a plurality of Gaussian-filtered SEM electron microscope scanning images to obtain clustered images of multiple groups of cemented filling bodies, each group of cemented filling bodies The number of cemented filling cluster images in the cluster image is equal to the number of cluster centers; Step 8034, the computer (17) determines the cemented filling body cluster image with the smallest gray value in each group of cemented filling body cluster images as the cemented filling body microscopic pore map, and compares the microscopic pores of multiple cemented filling bodies Binary processing is performed on the map to obtain binary maps of the microscopic pores of multiple cemented filling bodies; Step 8035, the computer (17) combines the multiple Gaussian-filtered SEM electron microscope scanning images obtained in step 8032 with the multiple binary images of cemented filling bodies obtained in step 8034 according to the corresponding numbers to obtain multiple a training sample image; Step 804, the computer (17) performs normalization processing on a plurality of training sample images respectively, forming a plurality of normalized training sample images whose pixels are 960×960; Step 805, the computer (17) constructs a convolutional network with five layers of layers, the input layer is a normalized training sample image, and the output layer is the Tensorflow deep learning of the uniaxial compressive strength corresponding to the normalized training sample image The network uses a plurality of normalized training sample images stored in it as training samples to train the Tensorflow deep learning network to obtain a Tensorflow deep learning mechanical response characteristic prediction network; the Tensorflow deep learning mechanical characteristic response prediction network is a five-layer convolutional network The kernel sizes are 3x3, 2x2, 3x3, 2x2, 2x2 from the first layer to the fifth layer. 7. The method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images according to claim 6, characterized in that: the uniaxial compressive strength test device for cemented filling bodies described in step 802 includes seat cushions (10) and fixed connections A plurality of pull rods (8) on the top of the seat cushion (10), and an axial pressure force transmission mechanism for applying axial pressure to the cemented filling body sample (19) and a force transmission mechanism for axial pressure An axial pressurization power system that provides power; the bottom of the seat cushion (10) is fixedly connected with a plurality of bases (15), and the top of the seat cushion (10) is provided with a cemented filling body sample (19) The sample placement groove of the seat cushion (10) is provided with a drain valve (5) at the center of the sample placement groove; the middle part of the plurality of pull rods (8) is provided with a plurality of pull rods (8) ) fixed frame (11), the top of a plurality of pull rods (8) is fixedly connected with a top loading plate (9); the axial pressurization force transmission mechanism includes a cylinder (2) mounted on the top loading plate (9) , the piston rod of the cylinder (2) is set downwards, and the bottom of the piston rod of the cylinder (2) is connected with a pressure transmission plate (3); the axial pressurized power system includes a compressed air source (4) and Pressurization controller (18), and the gas delivery pipe (1) that one end is connected with compressed air source (4), and the other end is connected with cylinder (2); The position connecting the source (4) to the cylinder (2) is provided with a pneumatic triple piece (12), a pressure sensor (13) and a cylinder control solenoid valve (14) in sequence, and the pressure sensor (13) is connected with the pressurization controller (18 ), the cylinder control solenoid valve (14) is connected with the output end of the pressurization controller (18), and the pressurization controller (18) is connected with the computer (17) through the communication module (16). 8. The method for predicting the mechanical response characteristics of cemented filling body based on SEM images according to claim 7, characterized in that: an O-shaped seal set at the bottom of the cemented filling body sample (19) is arranged in the said sample placement groove ring (6), the seat cushion (10) is provided with porous stones (7) located around the sample placement groove. 9. The method for predicting mechanical response characteristics of cemented filling bodies based on SEM images according to claim 8, characterized in that: the pressurization controller (18) is a programmable logic controller, and the communication module (16) is RS-485 communication module. 10. The method for predicting the mechanical response characteristics of cemented filling bodies based on SEM images according to claim 8, characterized in that: in step 802, the uniaxial compressive strength testing device for cemented filling bodies is used to test the compressive strength of multiple The sample is subjected to a uniaxial compressive strength test, wherein the specific process of performing a uniaxial compressive strength test on each compressive strength test sample is: Step 8021, after putting the O-ring (6) into the sample placement groove, put the cemented filling body sample (19) into the sample placement groove, so that the cemented filling body sample (19) Corresponding to the center of the piston rod of the cylinder (2) and the center of the pressure transmission plate (3); and on the seat cushion (10), put the porous stone (7) that is positioned around the sample placement groove; Step 8022, open the compressed air source (4), adjust the air pressure of the compressed air output by the compressed air source (4) by adjusting the pneumatic triple unit (12), and the pressurization controller (18) controls the solenoid valve (14) by controlling the cylinder ) reversing, control the downward or upward movement of the piston rod of the cylinder (2), apply pressure to the cemented filling body sample (19) or unload the pressure, and pressurize the controller (18) when the cemented filling body sample (19) ruptures ) The pressure value detected by the pressure sensor (13) collected is denoted as F, and the pressurization controller (18) transmits the pressure value F to the computer (17), and the computer (17) according to the formula Calculate the uniaxial compressive strength P of the compressive strength test sample; wherein, S is the top surface area of the compressive strength test sample; when the piston rod of the cylinder (2) moves downward, it drives the pressure transmission plate (3) to The downward movement applies pressure to the cemented filling body sample (19) through the pressure transmission plate (3). When the piston rod of the cylinder (2) moves upward, it drives the pressure transmission plate (3) to move upward, and the pressure transmission plate (3) Leave the upper surface of the cemented filling body sample (19) and unload the pressure.