Intelligent infrared radiation identification method for the damage and failure state of coal-bearing rock

Abstract

The present invention discloses an intelligent infrared radiation recognition method for the damage and failure state of coal and rock under load, belonging to the field of water-preserved mining and strata control. Using the infrared radiation monitoring technology of coal and rock, from the perspective of computer vision, first obtain the stress-strain and infrared radiation time series data during the damage and failure process of coal and rock under load, and establish an infrared radiation grayscale image classification data set for different progressive damage and failure stages based on the crack strain model method; build a convolutional neural network CNN classifier, input the prepared data set into the CNN model for training; finally, use the trained model for the intelligent recognition of the damage and failure stage of coal and rock under load. It realizes non-destructive and real-time intelligent monitoring and judgment of the damage and failure stage of coal and rock under load, and has certain practical significance for infrared radiation damage monitoring and early warning in mining engineering and geotechnical engineering.

Description

Technical Field

[0001] The present invention relates to an intelligent infrared radiation identification method for the damage and failure state of bearing coal and rock, belonging to the fields of water-preserved mining and strata control. Background Art

[0002] The identification and judgment of the damage stage of bearing coal and rock can be used to evaluate the stability of surrounding rocks, which is of great significance for mining engineering, geotechnical engineering, etc. The existing methods for judging the damage stage of bearing coal and rock mainly rely on the magnitude of characteristic strength. The main characteristic strength judgment methods include stress-strain curve method, volume strain method, crack volume strain method, acoustic emission method, moving point regression method, etc. Although the above methods can indirectly judge the damage stage of coal and rock through the measurement of characteristic strength, they are complex in operation, have low data utilization rate, and are all contact measurements, with low identification efficiency, resulting in a large amount of information waste and unable to monitor online in real time. In addition, in engineering practice, what is most concerned about is to judge which damage stage the bearing coal and rock is in. Therefore, it is necessary to study an efficient identification method for the damage stage of bearing coal and rock.

[0003] Research shows that infrared thermal radiation is accompanied during the damage and failure process of coal and rock, which has theoretical significance for the direct judgment of the damage stage of coal and rock. With the development of computer hardware and the level of infrared thermal imagers, it has high theoretical significance and application value to establish an identification method for the damage stage of coal and rock based on machine vision through infrared radiation monitoring technology. Summary of the Invention

[0004] Aiming at the defects of the existing methods for judging the damage and failure stage of bearing coal and rock, an intelligent infrared radiation identification method for the damage and failure state of bearing coal and rock is proposed, which can realize non-destructive, real-time, efficient and intelligent monitoring, judgment and early warning of the damage and failure stage of bearing coal and rock.

[0005] To achieve the above object, the present invention discloses an intelligent infrared radiation identification method for the damage and failure state of bearing coal and rock. According to experiments, a large amount of stress-strain and infrared radiation time series data during the damage and failure process of bearing coal and rock are obtained. Based on the crack strain model method, an infrared radiation grayscale image classification data set for different progressive damage and failure stages is established; the convolutional neural network model is trained by using the infrared radiation grayscale image classification data set, and the trained convolutional neural network model is used for the identification of the damage and failure state of bearing coal and rock;

[0006] The specific steps are as follows:

[0007] Step 1: First, collect samples for the combined monitoring of infrared radiation and stress-strain of the bearing coal rock. While obtaining the stress-strain time series data, obtain the infrared radiation monitoring data; use the crack strain model method to determine the characteristic strengths of all samples, including the closure strength, initiation strength, damage strength, and peak strength, and record the time nodes corresponding to each characteristic strength.

[0008] Step 2: Preprocess the original infrared data sequence, including regional division, denoising, and secondary reconstruction, and form a new infrared grayscale image sequence from the infrared data.

[0009] Step 3: According to the time nodes of the occurrence of each characteristic strength obtained in Step 1, divide the infrared grayscale image sequence into five stages: the initial compaction stage, the elastic deformation stage, the stable crack development stage, the unstable crack development stage, and the post-peak deformation stage.

[0010] Step 4: Divide the thermal images of the infrared grayscale image sequences of the same stage of all samples to form five types of infrared thermal image sequences: the initial compaction stage, the elastic deformation stage, the stable crack development stage, the unstable crack development stage, and the post-peak deformation stage; combine the infrared thermal images of the coal rock in the same stage to obtain an infrared thermal image dataset for the identification of the coal rock damage stage; make corresponding labels for each type of infrared thermal image.

[0011] Step 5: Build a convolutional neural network (CNN) model, input the infrared thermal image dataset into the CNN model for training and verification, and use the trained CNN model for the intelligent identification of the damage and failure stages of the bearing coal rock, realizing the non-destructive and real-time intelligent monitoring of the damage and failure stages of the bearing coal rock.

[0012] Furthermore, the steps for making the infrared thermal image classification dataset for each stage of the damage and failure of the bearing coal rock are as follows:

[0013] Step 4.1: Conduct the combined observation experiment of infrared radiation and stress-strain of the bearing coal rock: Place a reference sample beside the loaded sample so that the reference sample is of the same height and flush with the loaded sample; paste transverse and radial strain gauges on the side and back of the loaded coal rock sample, and place the infrared radiation observation device in front of the loaded sample for observation; synchronously conduct stress, strain, and infrared radiation monitoring experiments during the observation to obtain the stress and strain time series data of the loaded sample and the infrared radiation time series data of the loaded sample and the reference sample.

[0014] Step 4.2: Use the crack strain model method to calculate the stress-strain data of all coal rock samples to obtain the closure strength, initiation strength, damage strength, and peak strength of each coal rock sample, and record the time nodes of the changes.

[0015] During uniaxial loading, let the axial strain of the loaded coal rock sample be εaxial , the radial strain is ε radial , then the actual volume strain of the coal-rock sample during loading is: ε vol = ε axial + 2ε radial , where the elastic strain is: Then. The crack volume strain is: where υ and E are the Poisson's ratio and elastic modulus of the sample respectively, and ε axial and ε radial are the average values of the respective axial and radial strains;

[0016] The crack closure strength σ cc is the stress level at the start of the nearly horizontal segment in the curve, and the crack initiation strength σ ci is the stress level at the end of the nearly horizontal segment in the curve, and the crack damage strength σ cd is ε vol - ε axial the stress level corresponding to the inflection point or maximum point of the curve, and the peak stress level is σ f , record the times corresponding to σ cc , σ ci , σ cd , σ f as: t cc , t ci , t cd , t f ;

[0017] Step 4.3, denoise the infrared radiation thermal image data of the coal-rock sample recorded by the infrared radiation observation equipment;

[0018] Step 4.4, reconstruct the denoised infrared thermal image to generate a sequence of coal-rock thermal images;

[0019] Step 4.5, divide the time period of the reconstructed sequence of coal-rock thermal images to form sequences of infrared thermal images in the initial compaction stage, elastic deformation stage, stable crack development stage, unstable crack development stage, and post-peak deformation stage.

[0020] Further, the method for denoising the infrared radiation thermal image data of the coal-rock sample is:

[0021] Let the infrared temperature matrix sequence of the loaded coal-rock sample detected by the infrared radiation observation equipment be ITM l , and the infrared temperature matrix sequence of the reference sample be ITM r , subtract the infrared temperature matrix of each frame of the loaded coal-rock sample from the infrared temperature matrix of the reference sample to obtain the denoised infrared temperature matrix ITM ad, the calculation method is: ITM ad = ITM l - ITM r ;; Then, the multiplicative noise is removed by the wavelet soft threshold denoising method.

[0022] Furthermore, the method for reconstructing the denoised infrared thermal image is as follows:

[0023] Arbitrarily select an infrared temperature matrix sequence after denoising of a coal and rock sample as ITM ad , traverse all the temperature data in the infrared temperature matrix sequence to obtain the maximum temperature data value T max and the minimum temperature data value T min . Scale all the infrared temperature matrices of this coal and rock sample to the range of 0-255 to form an infrared thermal image gray matrix. The calculation formula is:

[0024]

[0025] where GMIT is the infrared thermal image gray matrix, x and y are the row index and column index of the matrix respectively. Then save GMIT as an infrared gray image in picture format and name it in the way of "sample number - frame serial number" to form a coal and rock thermal image sequence.

[0026] Furthermore, the specific method for dividing the time period of the reconstructed coal and rock thermal image sequence is as follows:

[0027] Since the recorded moments corresponding to σ cc , σ ci , σ cd , σ f are t cc , t ci , t cd , t f respectively, then the time period from 0 to t cc is defined as the initial compaction stage, t cc ~t ci is the elastic deformation stage, t ci ~t cd is the stable crack development stage, t cd ~t f is the unstable crack development stage, and t f ~the end of loading is the post-peak failure stage;

[0028] According to these five time intervals, the infrared thermal image sequence is correspondingly divided into five stages, and each stage is stored separately, namely: the infrared thermal image sequence in the initial compaction stage, the infrared thermal image sequence in the elastic deformation stage, the infrared thermal image sequence in the stable crack development stage, the infrared thermal image sequence in the unstable crack development stage, and the infrared thermal image sequence in the post-peak failure stage. Labels of corresponding categories are made for each frame of infrared thermal image, and the labeling format is "the absolute path of each frame of picture + category". The category names are 0, 1, 2, 3, 4 in sequence, where "0" represents the infrared thermal image in the initial compaction stage, "1" represents the infrared thermal image in the elastic deformation stage, "2" represents the infrared thermal image in the stable crack development stage, "3" represents the infrared thermal image in the unstable crack development stage, and "4" represents the infrared thermal image in the post-peak failure stage;

[0029] All the labels corresponding to the same category are placed in the folder with the name of "category name + Labels"; finally, all the data are packaged to form an infrared thermal image dataset for coal and rock damage stage recognition, and are divided into a training set, a test set and a validation set according to a certain proportion.

[0030] Furthermore, the convolutional neural network CNN includes four layers of structures. The first layer is a convolutional layer and a pooling layer, the second layer is a convolutional layer and a pooling layer, the third layer is a convolutional layer and a pooling layer, and the fourth layer is three fully connected layers.

[0031] Furthermore, the convolutional layer of the first layer includes 32 3*3 convolutional kernels, and the pooling layer includes 32 2*2 pooling kernels; the convolutional layer of the second layer includes 10 3*3 convolutional kernels, and the pooling layer includes 10 2*2 pooling kernels; the convolutional layer of the third layer includes 5 3*3 convolutional kernels, and the pooling layer includes 5 2*2 pooling kernels. The pooling layers of the first layer, the second layer and the third layer all use the average pooling method; the last fully connected layer uses the Softmax activation function, and the first three layers of structures and the first two fully connected layers of the fourth layer all use the Relu activation function. The cross-entropy loss function is used as the cost function, and the error backpropagation training is carried out by using the stochastic gradient descent method, and the learning rate uses a dynamic update strategy.

[0032] Beneficial effects:

[0033] The present invention proposes a method for establishing an infrared thermal image dataset for bearing coal and rock damage stage recognition, constructs a convolutional neural network model suitable for infrared radiation coal and rock damage stage recognition, realizes non-destructive, real-time, efficient and intelligent monitoring judgment and failure warning of the bearing coal and rock damage and failure stage, and has certain practical significance for infrared radiation damage monitoring and its warning in mining engineering and geotechnical engineering. Description of the drawings

[0034] Figure 1Schematic flow chart of the intelligent infrared radiation identification method for the damage and failure state of the bearing coal and rock in the embodiments of the present invention;

[0035] Figure 2 Schematic flow chart of the method for constructing the intelligent infrared thermal image identification data set of the coal and rock damage state in the embodiments of the present invention;

[0036] Figure 3 Schematic diagram of the crack strain model method and the classification of thermal image categories in the embodiments of the present invention;

[0037] Figure 4 Schematic diagram of the data set in the initial compaction stage in the embodiments of the present invention;

[0038] Figure 5 Schematic diagram of the data set in the elastic deformation stage in the embodiments of the present invention;

[0039] Figure 6 Schematic diagram of the data set in the stage of stable crack development in the embodiments of the present invention;

[0040] Figure 7 Schematic diagram of the data set in the stage of unstable crack development in the embodiments of the present invention;

[0041] Figure 8 Schematic diagram of the data set in the post-peak deformation stage in the embodiments of the present invention;

[0042] Figure 9 Schematic diagram of the data set structure in the embodiments of the present invention;

[0043] Figure 10 Intelligent infrared radiation identification model diagram for the damage and failure state of the bearing coal and rock in the embodiments of the present invention. Detailed implementation manners

[0044] The following further describes the implementation of the present invention with reference to the accompanying drawings:

[0045] As Figure 1 shown, first, the stress-strain and infrared radiation time series data during the damage and failure process of the bearing coal and rock are obtained, and an infrared radiation grayscale image classification data set for different progressive damage and failure stages is established based on the crack strain model method. A convolutional neural network (CNN) classifier is built, the prepared data set is input into the CNN model for training, and the trained model is used for the intelligent identification of the damage and failure stages of the bearing coal and rock.

[0046] Specifically, it includes the following steps:

[0047] Step 1: Conduct an infrared radiation observation experiment on the uniaxial compression of the bearing coal and rock to obtain the stress, strain, and infrared radiation time series data during the damage and failure process of the bearing coal and rock. The strain data includes axial strain and lateral strain.

[0048] Step 2: Produce a classification data set of infrared thermal images at each stage of coal and rock damage and failure. As Figure 2 shown, the specific method is as follows:

[0049] (1) Conduct a combined observation experiment on infrared radiation, stress, and strain of coal and rock under load. Place a reference specimen (at a distance of 10 cm) beside the loaded specimen so that the reference specimen is at the same height and level as the loaded specimen. Paste transverse and radial strain gauges on the side and back of the loaded coal and rock specimen, and use the front of the loaded specimen for infrared radiation observation. Simultaneously conduct stress, strain, and infrared radiation monitoring experiments to obtain the stress and strain time series data of the loaded specimen and the infrared radiation time series data of the loaded specimen and the reference specimen; among the five categories: the initial compaction stage, the elastic deformation stage, the stable crack development stage, the unstable crack development stage, and the post-peak deformation stage. Since the data of the above five categories are from one loading and failure cycle, the thermal image sequence within this cycle is divided into five category sequences according to the crack strain model method: store each category separately, then connect all categories in a folder, and finally put these five sequences together, which is the complete data set for identifying the rock damage state.

[0050] (2) Use the crack strain model method to calculate the stress and strain data of all coal and rock specimens to obtain the closure strength, initiation strength, damage strength, and peak strength of each coal and rock specimen, and record the time points.

[0051] As Figure 3 shown, taking the uniaxial loading method as an example, assume that the axial strain of the loaded specimen is ε axial , and the radial strain is ε radial , then the actual volume strain of the specimen during loading is: ε vol = ε axial + 2ε radial , where the elastic body strain is: Then, the crack volume strain is: where υ and E are the Poisson's ratio and elastic modulus of the specimen respectively, and ε axial and ε radial are the average values of each axial strain and radial strain respectively.

[0052] The crack closure strength σ cc is the stress level at the start of the nearly horizontal segment in the curve. The crack initiation strength σ ci is the stress level at the end of the nearly horizontal segment in the curve. The crack damage strength σ cd is the stress level corresponding to the inflection point or maximum point of the ε vol - ε axial curve. The peak stress level is σ f , record σcc , σ ci , σ cd , σ f The corresponding moments are: t cc , t ci , t cd , t f .

[0053] (3) Denoise the infrared radiation thermal image data of coal and rock. The denoising method is as follows: Let the infrared temperature matrix sequence of the loaded specimen be ITM l , and the infrared temperature matrix sequence of the reference specimen be ITM r . Subtract the infrared temperature matrix of each frame of the loaded specimen from the infrared temperature matrix of the reference specimen to obtain the denoised infrared temperature matrix ITM ad , and the calculation method is: ITM ad = ITM l - ITM r . Then, use the wavelet soft threshold denoising method to remove the multiplicative noise.

[0054] (4) Reconstruct the denoised infrared thermal image. Taking a certain specimen as an example, the infrared temperature matrix sequence after denoising of this specimen is ITM ad . Traverse all the temperature data in the sequence to obtain the maximum temperature data value T max and the minimum temperature data value T min . Scale all the infrared temperature matrices of this specimen to the range of 0 - 255 to form the infrared thermal image gray matrix. The calculation formula is:

[0055]

[0056] where GMIT is the infrared thermal image gray matrix, and x and y are the row index and column index of the matrix respectively. Then save GMIT as a jpg - format infrared gray image, named "specimen number - frame sequence number" to form a sequence of coal and rock thermal images.

[0057] (5) Divide the reconstructed sequence of coal and rock thermal images into time periods to form sequences of infrared thermal images in the initial compaction stage, elastic deformation stage, stable crack development stage, unstable crack development stage, and post - peak deformation stage.

[0058] Record the moments corresponding to σ cc , σ ci , σ cd , σ f as: t cc , t ci , t cd , t f . Then the time period from 0 to t cc is the initial compaction stage, tcc ~t ci is the elastic deformation stage, t ci ~t cd is the stage of stable crack development, t cd ~t f is the stage of unstable crack development, t f ~to the end of loading is the post-peak failure stage. According to these five time intervals, the infrared thermal image sequence is correspondingly divided into five stages. Respectively: the thermal image of the initial compaction stage ( Figure 4 ), the thermal image of the elastic deformation stage ( Figure 5 ), the thermal image of the stage of stable crack development, such as Figure 6 ; the thermal image of the stage of unstable crack development, such as Figure 7 , and the thermal image of the post-peak failure stage, such as Figure 8 . Make labels of corresponding categories for each frame of thermal image and save them as txt text. The label format is "the absolute path of each frame of picture + category". The category names are 0, 1, 2, 3, 4 in turn, where "0" represents the thermal image of the initial compaction stage, "1" represents the thermal image of the elastic deformation stage, "2" represents the thermal image of the stage of stable crack development, "3" represents the thermal image of the stage of unstable crack development, and "4" represents the thermal image of the post-peak failure stage. Put all the labels corresponding to the same category into the folder of the corresponding category, and the folder is named "category name + Labels". Finally, all the data are packed to form an infrared thermal image dataset for coal and rock damage stage identification, as shown in Figure 9 . And divide it into a training set, a test set and a validation set according to a certain proportion.

[0059] Step 3, establish a CNN-based identification model for the damage and failure stages of coal-bearing rock. It mainly includes four layers of structures, as shown in Figure 10 . The first layer is a convolutional layer and a pooling layer, the second layer is a convolutional layer and a pooling layer, the third layer is a convolutional layer and a pooling layer, and the fourth layer is three fully connected layers. The convolutional layer of the first layer includes 32 3*3 convolutional kernels, and the pooling layer includes 32 2*2 pooling kernels. The convolutional layer of the second layer includes 10 3*3 convolutional kernels, and the pooling layer includes 10 2*2 pooling kernels. The convolutional layer of the third layer includes 5 3*3 convolutional kernels, and the pooling layer includes 5 2*2 pooling kernels. The pooling layers of the first, second, and third layers all use the average pooling method. The last fully connected layer uses the Softmax activation function, and the first three layers of structures and the first two fully connected layers of the fourth layer all use the Relu activation function.

[0060] Step 4: Input the classification dataset obtained in Step 2 into the CNN model in Step 3 for model training and parameter optimization. During the training process, the cross-entropy loss function is used as the cost function, and the stochastic gradient descent method is used for error backpropagation training. The learning rate adopts a dynamic adjustment strategy. The optimized model is used for the intelligent classification and recognition of the damage and failure stages of the bearing coal and rock mass.

[0061] Based on the identification of the characteristic strength of the bearing coal and rock mass by the crack strain model method, the present invention establishes a classification dataset of infrared radiation thermal images in the damage stage of the bearing coal and rock mass, and divides the infrared radiation thermal images into five categories, namely: thermal images in the initial compaction stage, thermal images in the elastic deformation stage, thermal images in the stable crack development stage, thermal images in the unstable crack development stage, and thermal images in the post-peak failure stage. Then, from the perspective of machine learning, a convolutional neural network classification model for the damage stage of the bearing coal and rock mass is constructed, the model parameters are trained and optimized, realizing non-destructive, real-time, efficient intelligent monitoring and judgment and failure warning in the damage and failure stages of the bearing coal and rock mass, which has certain practical significance for infrared radiation damage monitoring and its warning in mining engineering and geotechnical engineering.

Claims

1. An intelligent infrared radiation identification method for the damage and failure state of coal and rock under load, characterized in that: Based on a large amount of stress-strain and infrared radiation time series data obtained during the damage and failure process of coal and rock under load through experiments, an infrared radiation grayscale image classification data set for different progressive damage and failure stages is established using the crack strain model method; the trained convolutional neural network model is used to identify the damage and failure state of coal and rock under load; The specific steps are as follows: Step 1: First, collect specimens for joint monitoring of infrared radiation and stress-strain of coal and rock under load. While obtaining the stress-strain time series data, obtain the infrared radiation time series data; use the crack strain model method to determine the characteristic strengths of all specimens, including the closure strength, initiation strength, damage strength, and peak strength, and record the time nodes when each characteristic strength occurs; Step 2: Preprocess the original infrared radiation time series data, including region division, denoising, and secondary reconstruction, and form a new infrared grayscale image sequence from the infrared radiation time series data; Step 3: According to the time nodes when each characteristic strength occurs obtained in Step 1, divide the infrared grayscale image sequence into five stages: initial compaction stage, elastic deformation stage, stable crack development stage, unstable crack development stage, and post-peak deformation stage; Step 4: Divide the thermal images of the infrared grayscale image sequences of the same stage of all specimens to form five types of infrared thermal image sequences: initial compaction stage, elastic deformation stage, stable crack development stage, unstable crack development stage, and post-peak deformation stage; combine the infrared thermal images of coal and rock in the same stage to obtain an infrared thermal image data set for identifying the damage stage of coal and rock; make corresponding labels for each type of infrared thermal image; Step 5: Build a convolutional neural network CNN model, input the infrared thermal image data set into the CNN model for training and verification, and use the trained CNN model for intelligent identification of the damage and failure stage of coal and rock under load to achieve non-destructive and real-time intelligent monitoring of the damage and failure stage of coal and rock under load.

2. The intelligent infrared radiation identification method for the damage and failure state of coal and rock under load according to claim 1, characterized in that, The steps for making the infrared thermal image classification data set for each stage of the damage and failure of coal and rock under load are as follows: Step 4.1: Conduct a joint observation experiment on infrared radiation and stress-strain of coal and rock under load: Place a reference specimen beside the loaded specimen so that the reference specimen is of the same height and level as the loaded specimen; paste transverse and radial strain gauges on the side and back of the loaded coal and rock specimen, and place the infrared radiation observation equipment in front of the loaded specimen for observation; during the observation, conduct stress, strain, and infrared radiation monitoring experiments synchronously to obtain the stress-strain time series data of the loaded specimen and the infrared radiation time series data of the loaded specimen and the reference specimen; Step 4.2: Use the crack strain model method to calculate the stress-strain data of all coal and rock specimens to obtain the closure strength, initiation strength, damage strength, and peak strength of each coal and rock specimen, and record the time nodes when changes occur; When uniaxially loaded, assume that the axial strain of the loaded coal and rock specimen is ε axial , and the radial strain is ε radial . Then the actual volume strain of the coal and rock specimen during loading is: ε vol = ε axial + 2ε radial , where the elastic body strain is: Then, the crack volume strain is: where υ and E are the Poisson's ratio and elastic modulus of the specimen respectively, and ε axial and ε radial are the average values of the respective axial strain and radial strain; The crack closure strength σ cc is the stress level at the start of the nearly horizontal segment in the curve, and the crack initiation strength σ ci is the stress level at the end of the nearly horizontal segment in the curve, and the crack damage strength σ cd is ε vol -ε axial the stress level corresponding to the inflection point or the maximum point of the ε f -ε cc curve, and the peak stress level is σ ci . Record that the moments corresponding to σ cd , σ f , σ cc , and σ ci are respectively: t cd , t f ; Step 4.3, denoise the infrared radiation thermal image data of the coal and rock specimens recorded by the infrared radiation observation equipment; Step 4.4, reconstruct the denoised infrared thermal image to generate a sequence of coal and rock thermal images; Step 4.5, divide the time period of the reconstructed sequence of coal and rock thermal images to form sequences of infrared thermal images in the initial compaction stage, elastic deformation stage, stable crack development stage, unstable crack development stage, and post-peak deformation stage.

3. The infrared radiation intelligent identification method for the damage and failure state of the bearing coal and rock according to claim 2, characterized in that, the method for denoising the infrared radiation thermal image data of the coal and rock specimens is: Let the infrared temperature matrix sequence of the loaded coal and rock specimen detected by the infrared radiation observation device be ITM l , and the infrared temperature matrix sequence of the reference specimen be ITM r . Subtract the infrared temperature matrix of each frame of the loaded coal and rock specimen from the infrared temperature matrix of the reference specimen to obtain the denoised infrared temperature matrix ITM ad . The calculation method is: ITM ad = ITM l - ITM r ; then perform wavelet soft threshold denoising method to remove multiplicative noise on it.

4. The infrared radiation intelligent identification method for the damage and failure state of the bearing coal and rock according to claim 2, characterized in that, the method for reconstructing the denoised infrared thermal image is: The infrared temperature matrix sequence after noise reduction of any coal-rock sample is ITM ad , traverse all temperature data in the infrared temperature matrix sequence to obtain the maximum temperature data value T max and the minimum temperature data value T min , scale all infrared temperature matrices of the coal-rock sample to the range of 0-255 to form an infrared thermal image grayscale matrix. The calculation formula is as follows: where GMIT is the grayscale matrix of the infrared thermal image, x and y are the row index and column index of the matrix respectively, and then GMIT is saved as an infrared grayscale image in picture format, named in the way of "specimen number - frame serial number" to form a sequence of coal and rock thermal images.

5. The infrared radiation intelligent identification method for the damage and failure state of the bearing coal and rock according to claim 2, characterized in that, the specific method for dividing the time period of the reconstructed sequence of coal and rock thermal images is: Due to the record σ cc , σ ci , σ cd , σ f , the corresponding moments are respectively: t cc , t ci , t cd , t f . Then, it is defined that the time period from 0 to t cc is the initial consolidation stage, the time period from t cc to t ci is the elastic deformation stage, the time period from t ci to t cd is the stable crack growth stage, the time period from t cd to t f is the unstable crack growth stage, and the time period from t f to the end of loading is the post-peak failure stage; According to these five time intervals, the sequence of infrared thermal images is correspondingly divided into five stages, and each stage is stored separately, namely: the sequence of thermal images in the initial compaction stage, the sequence of thermal images in the elastic deformation stage, the sequence of thermal images in the stable crack development stage, the sequence of thermal images in the unstable crack development stage, and the sequence of thermal images in the post-peak failure stage. Make labels of corresponding categories for each frame of thermal image for annotation, and the annotation format is "the absolute path of each frame of picture + category". The category names are 0, 1, 2, 3, 4 in turn, where "0" represents the thermal image in the initial compaction stage, "1" represents the thermal image in the elastic deformation stage, "2" represents the thermal image in the stable crack development stage, "3" represents the thermal image in the unstable crack development stage, and "4" represents the thermal image in the post-peak failure stage; Put all the labels corresponding to the same category into the folder corresponding to the category, and the folder is named "category name + Labels"; finally, all the data are packaged to form an infrared thermal image dataset for identifying the damage stage of coal and rock, and are divided into a training set, a test set, and a validation set according to a ratio.

6. The infrared radiation intelligent identification method for the damage and failure state of the bearing coal and rock according to claim 1, characterized in that, the convolutional neural network CNN includes four-layer structure. The first layer is a convolutional layer and a pooling layer, the second layer is a convolutional layer and a pooling layer, the third layer is a convolutional layer and a pooling layer, and the fourth layer is three fully connected layers.

7. The infrared radiation intelligent identification method for the damage and failure state of the bearing coal and rock according to claim 6, characterized in that The first convolutional layer consists of 32 3*3 convolutional kernels, and the pooling layer consists of 32 2*2 pooling kernels; the second convolutional layer consists of 10 3*3 convolutional kernels, and the pooling layer consists of 10 2*2 pooling kernels; the third convolutional layer consists of 5 3*3 convolutional kernels, and the pooling layer consists of 5 2*2 pooling kernels. The pooling layers of the first, second, and third layers all use average pooling; the last fully connected layer uses the Softmax activation function, and the first three layers and the first two fully connected layers of the fourth layer all use the Relu activation function; the cross-entropy loss function is used as the cost function, and the error backpropagation training is carried out using the stochastic gradient descent method, and the learning rate uses a dynamic update strategy.

Images