Coal-rock synchronous loading and CT scanning method based on deep learning adaptive control

Abstract

The present invention belongs to the field of coal-rock synchronous loading and CT scanning test and CT image processing, and specifically relates to a coal-rock synchronous loading and CT scanning method based on deep learning adaptive control; including constructing a coal-rock synchronous loading and CT scanning system and an AI edge computing unit; constructing a coal-rock synchronous loading and CT scanning adaptive control model: the model includes several feature extraction modules based on a deep convolutional neural network and a time series information capture module based on a long short-term memory network connected one-to-one with each feature extraction module, the time series information capture module is further connected to a state prediction module, and the state prediction module is used to predict the fracture state; the model is trained and embedded in the AI ​​edge computing unit in the form of software, and the corresponding loading rate adjustment parameters are set when each fracture state occurs. The present invention can accurately grasp the timing of loading rate adjustment, obtain sufficient images of the coal-rock test fracture development process on the basis of ensuring the overall test efficiency, and facilitate subsequent test analysis.

Description

Technical Field

[0001] The present invention belongs to the field of coal-rock synchronous loading and CT scanning test and CT image processing, and specifically relates to a coal-rock synchronous loading and CT scanning method based on deep learning adaptive control. Background Art

[0002] When conducting simultaneous coal-rock loading and CT scanning tests, the loading rate of the loading equipment and the interval between CT scans are generally preset before the test. However, since the deformation rate of coal-rock specimens after cracks are generated is relatively fast, a faster loading rate will result in fewer images of the crack development process captured by the CT scanning equipment, affecting subsequent test analysis. A slower loading rate, on the other hand, will result in a longer loading time before cracks are generated in the coal-rock specimens, affecting overall test efficiency. Artificially adopting different loading rates at different times during the loading of coal-rock specimens based on experience makes it difficult to accurately grasp the timing of loading rate adjustment. Furthermore, manually analyzing the crack development after obtaining CT scan images and then adjusting the loading rate is also time-consuming, making it difficult to adjust the loading rate in a timely manner. Furthermore, the crack development rates of coal-rock specimens vary at different stages of crack development, and the corresponding loading rates required are also different, posing a greater challenge to adjusting the loading rate. Summary of the Invention

[0003] To solve the above problems, the present invention proposes a coal-rock synchronous loading and CT scanning method based on deep learning adaptive control, which includes the following steps:

[0004] S1. Construction of coal-rock synchronous loading and CT scanning system

[0005] The coal-rock synchronous loading and CT scanning system includes a loading system, a CT scanning system, and an AI edge computing unit; the AI ​​edge computing unit can acquire CT images scanned and reconstructed by the CT scanning system in real time, and can also provide loading rate parameters to the loading system in real time;

[0006] S2. Constructing a model for adaptive control of coal-rock synchronous loading and CT scanning

[0007] The model includes several feature extraction modules based on deep convolutional neural networks and time series information capture modules based on long short-term memory networks, which are connected one-to-one with each feature extraction module. The output ends of all time series information capture modules are connected to the state prediction module, and adjacent time series information capture modules are connected. The state prediction module is a fully connected layer that outputs predicted crack states, including crack initiation state, crack expansion state, and crack penetration state.

[0008] S3. Train the model and embed it into the AI ​​edge computing unit as software, and set the loading rate adjustment parameters corresponding to each crack state;

[0009] S4. Perform adaptively controlled synchronous loading of coal and rock and CT scanning.

[0010] Preferably, in step S1, the loading system includes a fixed base, a fixed gantry is provided on the fixed base, a loading motor is provided on the fixed gantry, a pressure rod is connected to the lower portion of the loading motor, a pressure head is rotatably connected to the lower portion of the pressure rod, and a rotating base is rotatably connected to the fixed base; and also includes a loading control unit, which is at least used to control the loading rate and loading pressure of the loading system.

[0011] Preferably, in step S1, the CT scanning system includes an irradiator and a ray source respectively located on both sides of the loading system, and also includes a scanning control unit, and the scanning control unit is at least used to control the time interval of a scan.

[0012] Preferably, in step S2, the feature extraction module is implemented by convolution and pooling operations based on the ResNet-18 model, the activation function uses the nonlinear activation function ReLU, and the pooling operation type is L p Pooling: The input of each feature extraction module is a single CT image, and the last layer of the ResNet-18 network model is removed.

[0013] Preferably, in step S2, the crack state is divided based on the crack length change rate and the crack width change rate.

[0014] Preferably, in step S3, CT images are obtained after a plurality of coal rock samples are subjected to synchronous loading and CT scanning tests; for each coal rock sample, it is assumed that it has a total of M CT images, starting from the first N CT images are used as sample features, based on the N +1 CT image and N The crack state corresponding to the crack length change rate and width change rate calculated from the CT image is used as the sample label, 2≤ N +1≤ M ; Create several samples and bring them into the model for training.

[0015] Preferably, in step S4, a coal-rock synchronous loading and CT scanning system is used for testing. The CT scanning system acquires CT images in real time and inputs the CT images into the trained model. Before scanning the next CT image, the model predicts the crack changes and their corresponding crack states between the next CT image and the most recently acquired CT image, and adjusts the loading rate when different crack states appear.

[0016] Beneficial Technical Effects: Based on the fusion modeling of two deep learning networks, a convolutional neural network and a long short-term memory network, the present invention can predict the fracture state corresponding to the fracture changes between the next CT image and the most recently acquired CT image based on the acquired CT image before scanning the next CT image, and promptly adjust the loading rate when different fracture states appear. The present invention can accurately grasp the timing of loading rate adjustment, obtain sufficient images of the fracture development process in coal and rock tests while ensuring overall test efficiency, and facilitate subsequent test result analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of the loading device and CT scanning device of the present invention;

[0018] Figure 2 This is the coal-rock synchronous loading and CT scanning adaptive control model of the present invention;

[0019] In the figure: 1-loading motor; 2-pressure rod; 3-pressure head; 4-coal and rock sample; 5-rotating base; 6-fixed base; 7-fixed gantry; 8-radiator (radiation source); 9-radiation source. DETAILED DESCRIPTION

[0020] In the detailed description section, the technical solution of the present invention will be described in detail with reference to the accompanying drawings.

[0021] like Figure 1-2 As shown, the present invention proposes a method for synchronous coal-rock loading and CT scanning based on deep learning adaptive control, which includes the following steps:

[0022] S1. Construction of coal-rock synchronous loading and CT scanning system

[0023] like Figure 1 As shown, the coal-rock synchronous loading and CT scanning system includes a loading system, a CT scanning system, and an AI edge computing unit; the loading system includes a fixed base 6, a fixed gantry 7 is provided on the fixed base 6, a loading motor 1 is provided at the center of the fixed gantry 7, a pressure rod 2 is connected to the lower part of the loading motor 1, a pressure head 3 is rotatably connected to the lower part of the pressure rod 2, a rotating base 5 is rotatably connected to the fixed base 6, the rotating base 5 is located directly below the pressure head 3, and a coal-rock sample 4 for testing is provided between the rotating base 5 and the pressure head 3; the loading system also includes a loading control unit, which is used to at least control the loading rate and loading pressure of the loading system;

[0024] The CT scanning system includes a radiator (radiation source) 8 and a radiation source 9, which are arranged at the same height as the coal and rock sample 4 and are located on both sides of the loading system, such as on the left and right sides or on the front and back sides; the CT scanning system also includes a scanning control unit, which is used to control the time interval between scans at least once;

[0025] The AI ​​edge computing unit is connected to the CT scanning system to obtain CT images scanned and reconstructed by the CT scanning system in real time. The AI ​​edge computing unit is also connected to the loading system to provide loading rate parameters to the loading control unit in real time. The AI ​​edge computing unit uses NVIDIA Jetson AGX Xavier as hardware;

[0026] S2. Constructing a model for adaptive control of coal-rock synchronous loading and CT scanning

[0027] like Figure 2 As shown in the figure, the model is modeled by integrating two deep learning networks, convolutional neural network (CNN) and long short-term memory network (LSTM), including several feature extraction modules based on deep convolutional neural network and a corresponding number of time series information capture modules based on long short-term memory network. The feature extraction module is implemented by convolution and pooling operations based on ResNet-18 model. The activation function uses nonlinear activation function ReLU, and the pooling operation type is L p Pooling; the input of each feature extraction module is a single CT image. In order to integrate the ResNet-18 model with the LSTM model, the last layer of the ResNet-18 network model is removed to output a three-dimensional tensor, and the dimension values ​​of the three dimensions are 1, 1, and 4096 respectively; the temporal information capture module is mainly constructed based on the long short-term memory network (LSTM model). The long short-term memory network contains three control gates: input, output, and forgetting. Its structure is well known in the art and will not be described here; the input of each temporal information capture module is the output of the feature extraction module, and the output ends of all temporal information capture modules are connected to the state prediction module, and adjacent temporal information capture modules are connected; the state prediction module is a fully connected layer with an output dimension of 3, corresponding to three predicted crack states, including crack initiation state, crack expansion state, and crack penetration state;

[0028] A convolutional neural network (CNN) is used to extract crack features from CT images, including crack length and width. A long short-term memory network (LSTM) is used to predict crack evolution trends and determine crack status based on historical data.

[0029] Among them, the conditions corresponding to the crack initiation state are: the crack length change rate is less than 0.05mm / min, and the crack width change rate is less than 0.02mm / min; the conditions corresponding to the crack expansion state are: the crack length change rate is between 0.05 and 0.2mm / min, or the crack width change rate is between 0.02 and 0.1mm / min; the crack penetration state is: the crack length change rate is greater than 0.2mm / min, or the crack width change rate is greater than 0.1mm / min; the crack length change rate is calculated by: using the ratio of the difference between the maximum length of the crack on the next CT image and the maximum length of the crack on the previous CT image to the difference in scanning time between the two CT images; the crack width change rate is calculated by: using the ratio of the difference between the maximum width of the crack on the next CT image and the maximum width of the crack on the previous CT image to the difference in scanning time between the two CT images;

[0030] S3. Train the model and embed it into the AI ​​edge computing unit in the form of software

[0031] Obtain CT images of several coal and rock samples after synchronous loading and CT scanning tests; for each coal and rock sample, it is assumed that it has a total of M CT images, which can be continuous from the first image N CT images are used as sample features, based on the N +1 CT image and N The crack state corresponding to the crack length change rate and width change rate calculated from the CT image is used as the sample label, 2≤ N +1≤ M ; Create several samples and bring them into the model for training;

[0032] The trained model is embedded in the AI ​​edge computing unit in the form of software, and it is set that when a crack initiation state occurs, the AI ​​edge computing unit gives an instruction to the loading control unit: adjust the loading rate to 50% of the initial set loading rate; when a crack initiation state occurs, the AI ​​edge computing unit gives an instruction to the loading control unit: adjust the loading rate to 20% of the initial set loading rate; when a crack initiation state occurs, the AI ​​edge computing unit gives an instruction to the loading control unit: adjust the loading rate to 10% of the initial set loading rate.

[0033] S4. Adaptive Control of Coal-Rock Synchronous Loading and CT Scanning

[0034] The experiment was conducted using a coal-rock synchronous loading and CT scanning system. The CT scanning system acquired CT images in real time and input the CT images into the trained model. The model can predict the crack changes and corresponding crack states between the next CT image and the most recently acquired CT image before scanning the next CT image, and adjust the loading rate when different crack states appear.

[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Within the technical scope disclosed by the present invention, any changes or alternatives that can be easily thought of by any person skilled in the art should be included in the scope of protection of the present invention.

Claims

1. A method for synchronous coal-rock loading and CT scanning based on deep learning adaptive control, characterized in that: include: S1. Constructing a coal-rock synchronous loading and CT scanning system: The coal-rock synchronous loading and CT scanning system includes a loading system, a CT scanning system, and an AI edge computing unit; the AI ​​edge computing unit can acquire CT images scanned and reconstructed by the CT scanning system in real time and can also provide loading rate parameters to the loading system in real time; S2. Construct an adaptive control model for simultaneous coal-rock loading and CT scanning: The model includes several feature extraction modules based on deep convolutional neural networks and a time series information capture module based on a long short-term memory network, each of which is connected to the feature extraction module in a one-to-one correspondence. The outputs of all time series information capture modules are connected to a state prediction module, and adjacent time series information capture modules are connected to each other. The state prediction module is a fully connected layer that outputs predicted fracture states, including fracture initiation, fracture expansion, and fracture penetration. S3. Train the model and embed it into the AI ​​edge computing unit as software, and set the loading rate adjustment parameters corresponding to each crack state; S4. Adaptively control synchronous loading and CT scanning of coal and rock: Experiments are conducted using a synchronous loading and CT scanning system for coal and rock. The CT scanning system acquires CT images in real time and inputs the CT images into the trained model. Before scanning the next CT image, the model predicts the crack changes and corresponding crack states between the next CT image and the most recently acquired CT image, and adjusts the loading rate when different crack states occur.

2. The method for simultaneous coal and rock loading and CT scanning according to claim 1, characterized in that: In step S1, the loading system includes a fixed base, a fixed gantry is provided on the fixed base, a loading motor is provided on the fixed gantry, a pressure rod is connected to the lower part of the loading motor, a pressure head is rotatably connected to the lower part of the pressure rod, and a rotating base is rotatably connected to the fixed base; and also includes a loading control unit, which is at least used to control the loading rate and loading pressure of the loading system.

3. The method for simultaneous coal and rock loading and CT scanning according to claim 1, characterized in that: In step S2, the feature extraction module is implemented by convolution and pooling operations based on the ResNet-18 model, the activation function uses the nonlinear activation function ReLU, and the pooling operation type is L p Pooling: The input of each feature extraction module is a single CT image, and the last layer of the ResNet-18 network model is removed.

4. The method for simultaneous coal and rock loading and CT scanning according to claim 3, characterized in that: In step S2, the crack state is divided based on the crack length change rate and the crack width change rate.

5. The method for simultaneous coal and rock loading and CT scanning according to claim 4, characterized in that: In step S3, CT images are obtained after a number of coal and rock samples are subjected to synchronous loading and CT scanning tests; for each coal and rock sample, it is assumed that it has a total of M CT images, starting from the first N CT images are used as sample features, based on the N +1 CT image and N The crack state corresponding to the crack length change rate and width change rate calculated from the CT image is used as the sample label, 2≤ N +1≤ M ; Create several samples and bring them into the model for training.

Images