A tunnel secondary lining material distribution system based on neural network

Abstract

The present invention relates to the field of engineering digital technology, and in particular to a tunnel secondary lining material distribution system based on a neural network, which generates a height difference matrix by acquiring concrete height data at different points in the material distribution process in real time. Multiple image capture devices are configured to obtain concrete lining state images in real time. After preprocessing and annotation, these images are input into a convolutional neural network for training and processing to determine the characterization state of concrete, including the degree of surface bubbling, the degree of slurry overflow, and the location of defects. The material distribution operation window is selected according to the height difference, and the automatic material distribution machine is controlled to perform material distribution operations. The concrete characterization state is generated based on the convolutional neural network to obtain a continuous state set. The height difference and the vibration parameters are combined to perform reinforcement learning and control the operation of the vibration equipment. By recording the duration of the concrete lining stop and issuing an abnormal alarm. The system realizes intelligent analysis and optimization control of the concrete state, ensures the uniformity and density of the material distribution, and improves construction efficiency and quality.

Description

Technical Field

[0001] The invention relates to the technical field of engineering digitization, and in particular to a tunnel secondary lining material distribution system based on a neural network. Background Art

[0002] In tunnel construction, secondary lining (referred to as second lining) is an important link to ensure the safety and durability of tunnel structure. The secondary lining spreading process involves the spraying, vibration and curing of concrete, and the uniformity and density of concrete must be ensured to avoid problems such as bubbles, slurry and settlement. The traditional secondary lining spreading system mainly relies on manual operation and experience judgment, and has many problems such as low efficiency, poor accuracy and safety hazards.

[0003] In current technology, mechanical equipment (such as concrete placing machines) is used for concrete spraying. Although mechanical equipment improves the efficiency of concrete placing, it still requires manual operation and monitoring. Due to the lack of intelligent control, the thickness and uniformity of concrete are difficult to control accurately. Multiple people are required to observe the operation window of the concrete placing trolley in real time with the naked eye and feedback to the control placing machine. Due to the reliance on manual operation and experience judgment, the thickness and uniformity of concrete placing are difficult to guarantee. During the construction of the secondary lining of the tunnel, the qualified standard for concrete placing is that there are no bubbles, no obvious settlement, and no surface slurry after vibration. However, the existing mechanical placing system cannot automatically meet these standards and requires a lot of manual monitoring and adjustment, which further reduces construction efficiency and increases the risk of error. Under the premise of being based on a large number of manual multiple judgments and requiring information synchronization for judgment and operation, mechanical placing still requires a lot of manual intervention, low efficiency, and long construction period. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a tunnel secondary lining material distribution system based on neural network, which replaces the existing material distribution process that requires a lot of manual participation in monitoring and control and has poor accuracy by combining convolutional neural network with reinforcement learning.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] A tunnel secondary lining material distribution system based on a neural network comprises: a distance sensing module, an image acquisition module, an intelligent control module and an abnormal alarm module, wherein the distance sensing module, the image acquisition module and the abnormal alarm module are respectively connected to the intelligent control module;

[0007] The distance sensing module is used to obtain concrete height data at several points during the laying process, and includes several laser radar sensors arranged at the laying operation window of the lining trolley;

[0008] The image acquisition module is used to acquire the concrete lining status image in real time, and includes a plurality of image capture devices arranged at the material distribution operation window of the lining trolley;

[0009] The intelligent control module is used to perform the following steps:

[0010] S1. Select the material placing operation window according to the height difference of concrete at different points and control the placing machine to carry out the material placing operation;

[0011] S2, judging the concrete representation state according to the concrete lining state image based on the convolutional neural network;

[0012] S3, generating a continuous state set according to the concrete representation state, and performing reinforcement learning based on the continuous state set, the concrete height difference and the vibration parameters to obtain a reinforcement learning model;

[0013] S4, based on the reinforcement learning model, real-time generation of vibration parameters to control the vibration equipment;

[0014] The abnormal alarm module is used to record the duration of the concrete lining stoppage and issue an abnormal alarm.

[0015] Furthermore, the S1 comprises the following steps:

[0016] Calculate the concrete height difference at different points based on the concrete height data at several points during the placement process;

[0017] Generate a height difference matrix based on the concrete height difference at different points;

[0018] Traversing the matrix to obtain the insufficient thickness area and generate a material distribution strategy, wherein the material distribution strategy includes determining the current material distribution operation window, planning the material distribution path and adjusting the material distribution speed;

[0019] The material placing machine is controlled to perform material placing operations based on the material placing strategy.

[0020] Further, the S2 comprises the following steps:

[0021] The image acquisition module is used to collect the concrete lining state image during the laying process, and the concrete characterization state is annotated on the concrete lining state image;

[0022] The convolutional neural network is trained with the concrete lining state image as input and the concrete state annotation result as output;

[0023] Based on the trained convolutional neural network, the real-time collected concrete lining status images are processed to extract the real-time concrete representation status.

[0024] Furthermore, the concrete characterization state includes the degree of surface bubbling, the degree of slurry bleeding and the location of defects.

[0025] Furthermore, S3 includes: generating a continuous state set according to the concrete representation state, defining the state parameters of the reinforcement learning model according to the continuous state set and the height difference matrix, using vibration positioning, vibration time and vibration intensity as action parameters, generating a reward value according to the concrete state after vibration and the height difference matrix, and training the reinforcement learning model.

[0026] Furthermore, the formula of the reinforcement learning is as follows:

[0027] Q(s t , a t )=Q(s t , a t )+α[r t +γ(∑ s′ P(s t+1 =s′|s t , a t )max a′ Q(s′, a′))-Q(s t , a t )];

[0028] Among them, s t is the state parameter at time step t; a t is the state parameter set s t The action parameters under the condition include vibration positioning, vibration time and vibration intensity; Q(s t , a t ) is the state parameter s t Execute action parameter a t The quality value of α is the learning rate, which is used to control the learning speed; r t is the reward value at time step t; γ is the discount factor; P(s t+1 =s′|s t , a t ) is in state s t Next, perform action a t Then transfer to the next state s t+1 = probability of s′; max a′ Q(s′, a′) is the maximum quality value of selecting the optimal action a′ in the state s′ at the next step t+1.

[0029] Furthermore, the reward value r t The formula is as follows:

[0030] r t =w 1 ·f b (s t+1 )+w 2 ·f p (st+1 )+w 3 ·f s (s t+1 );

[0031] Among them, f b (s t+1 ) is the state parameter s of concrete at time step t+1 t+1 The degree of bubbling on the lower surface; f p (s t+1 ) is the state parameter s of concrete at time step t+1 t+1 The value of the flooding degree under f s (s t+1 ) is the state parameter s of concrete at time step t+1 t+1 The value of the settlement degree under 1 、w 2 and w 3 is the weight parameter, where w 1 is a negative number, w 2 is a positive number, w 3 Is a negative number.

[0032] Furthermore, the settlement degree value is determined according to the difference between the height difference matrix values ​​at time step t+1 and time step t.

[0033] Furthermore, the abnormal alarm module is used to perform the following steps:

[0034] Preset initial setting time of concrete;

[0035] Obtain the duration of concrete lining stoppage by monitoring the intelligent control module;

[0036] If the duration of the concrete lining stoppage is greater than or equal to the initial setting time of the concrete, an abnormal alarm will be issued;

[0037] If the duration of the concrete lining stoppage is less than the initial setting time of the concrete, no abnormal alarm will be issued.

[0038] The beneficial effects of the present invention are as follows: the scheme adopts several laser radar sensors through the distance perception module to obtain the concrete height data of several points in the material distribution process in real time, so as to accurately calculate the concrete height difference at different points and generate a height difference matrix. The image acquisition module is equipped with multiple image capture devices to obtain the concrete lining state image in real time. After preprocessing and annotation, these images are input into the convolutional neural network for training and processing, so as to judge the characterization state of the concrete, including the degree of surface bubbling, the degree of slurry overflow and the defect location. The intelligent control module selects the material distribution operation window according to the concrete height difference at different points, and controls the automatic material distribution machine to perform the material distribution operation. Subsequently, the concrete lining state image is analyzed based on the convolutional neural network to judge the characterization state of the concrete. Then, a continuous state set is generated according to the concrete characterization state, and reinforcement learning is performed in combination with the concrete height difference and the vibrating parameter to obtain an optimized reinforcement learning model. Finally, the vibrating parameter is generated in real time based on the reinforcement learning model to control the vibrating equipment to ensure the compactness and uniformity of the concrete and meet the construction requirements. The abnormal alarm module records the duration of the concrete lining stoppage and triggers an alarm when the preset initial setting time is exceeded, ensuring the continuity and quality of the construction process. Through the above solution, the system uses lidar sensors and image capture devices to achieve real-time monitoring of concrete height and status, eliminating the subjectivity and uncertainty of manual observation. The introduction of convolutional neural networks and reinforcement learning algorithms enables the system to intelligently analyze and optimize the state of concrete, thereby ensuring the uniformity and density of the material distribution. The intelligent control module reduces manual intervention and improves the efficiency of material distribution and vibration; sensors and neural network technology realize real-time monitoring and adjustment, significantly improving the construction quality; the abnormal alarm module ensures the safety and continuity of the construction process. The intelligent and automated control of the tunnel secondary lining material distribution process is realized, significantly improving the construction efficiency and quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a structural schematic diagram of a tunnel secondary lining material distribution system based on a neural network in the present invention.

[0040] Figure 2 It is a flow chart of the execution steps of the intelligent control module in the present invention. DETAILED DESCRIPTION

[0041] See also Figure 1-2 As shown, the present invention relates to a tunnel secondary lining material distribution system based on a neural network, comprising: a distance sensing module, an image acquisition module, an intelligent control module and an abnormal alarm module, wherein the distance sensing module, the image acquisition module and the abnormal alarm module are respectively connected to the intelligent control module;

[0042] The distance sensing module is used to obtain concrete height data at several points during the laying process, and includes several laser radar sensors arranged at the laying operation window of the lining trolley;

[0043] The image acquisition module is used to acquire the concrete lining status image in real time, and includes a plurality of image capture devices arranged at the material distribution operation window of the lining trolley;

[0044] The intelligent control module is used to perform the following steps:

[0045] S1. Select the material placing operation window according to the height difference of concrete at different points and control the placing machine to carry out the material placing operation;

[0046] S2, judging the concrete representation state according to the concrete lining state image based on the convolutional neural network;

[0047] S3, generating a continuous state set according to the concrete representation state, and performing reinforcement learning based on the continuous state set, the concrete height difference and the vibration parameters to obtain a reinforcement learning model;

[0048] S4, based on the reinforcement learning model, real-time generation of vibration parameters to control the vibration equipment;

[0049] The abnormal alarm module is used to record the duration of the concrete lining stoppage and issue an abnormal alarm.

[0050] In some embodiments, the distance sensing module is used to obtain the concrete height data at several points during the material distribution process in real time. Specifically, the module includes several laser radar sensors, which are evenly distributed and installed in each material distribution operation window of the lining trolley and tilted downward toward the window, that is, when the operation window is closed, it is isolated and stops working. The laser radar sensor measures the height data of the concrete surface by emitting and receiving laser beams. This height data is specifically determined by the window height and the current relative distance between the concrete and the window. These sensors can collect concrete height information in real time and accurately, and transmit the data to the intelligent control module for processing. The image acquisition module is composed of multiple high-resolution image capture devices, which are also installed around the material distribution operation window of the lining trolley. Each image capture device can be an Internet of Things camera, which is set on one side of the distance sensor to capture the concrete lining state image under different lighting conditions. After the captured image is initially processed, it is transmitted to the intelligent control module for subsequent convolutional neural network analysis. The abnormal alarm module is used to monitor abnormal conditions during the concrete lining process. This module is installed in the intelligent control module and is equipped with a high-sensitivity sensor and a data processing unit. The abnormal alarm module records the duration of the concrete lining stop in real time. When it detects that the stop time exceeds the preset initial setting time, it immediately issues an alarm signal. The alarm information includes the abnormal point, time and specific cause, helping construction personnel to quickly take corrective measures to ensure the continuity and quality of the construction process.

[0051] Furthermore, the S1 comprises the following steps:

[0052] Calculate the concrete height difference at different points based on the concrete height data at several points during the placement process;

[0053] Generate a height difference matrix based on the concrete height difference at different points;

[0054] Traversing the matrix to obtain the insufficient thickness area and generate a material distribution strategy, wherein the material distribution strategy includes determining the current material distribution operation window, planning the material distribution path and adjusting the material distribution speed;

[0055] The material placing machine is controlled to perform material placing operations based on the material placing strategy.

[0056] In some embodiments, the distance sensing module collects the concrete height data of several points, and inputs these data into the intelligent control module for processing. First, the system calculates the concrete height difference of each point, that is, the difference between the actual height of each point and the reference height. Through this calculation, the system can generate a height difference matrix, which records the height difference values ​​of each point in the distribution area. Next, the system traverses the height difference matrix to identify the area where the concrete thickness is insufficient. The traversal algorithm compares the height difference of each point with the preset thickness threshold and marks the points with insufficient thickness. After marking these areas, the system generates the corresponding distribution strategy. The distribution strategy includes determining the current distribution operation window, planning the distribution path, and adjusting the distribution speed. When determining the current distribution operation window, the system dynamically adjusts the opening and closing state of the distribution operation window according to the position of the insufficient thickness area to ensure that the operation window is aligned with the area where the material needs to be replenished. Through the optimization algorithm, the system can ensure that the distribution efficiency is maximized. When planning the distribution path, the system uses a path planning algorithm, such as the A* algorithm or the Dijkstra algorithm, to ensure the uniform distribution of concrete and the optimal movement trajectory of the distribution equipment. The path planning takes into account the actual situation of the distribution area and designs the shortest path to cover all areas with insufficient thickness. The distribution speed is adjusted based on the real-time feedback of concrete thickness data. The system dynamically adjusts the speed of the distribution boom through the control algorithm to ensure the uniformity and density of the distribution process. The control algorithm comprehensively considers the error between the target thickness and the actual thickness, and ensures the accuracy of the concrete distribution process by adjusting the operating parameters of the distribution boom. Finally, based on the above distribution strategy, the intelligent control module controls the distribution boom to carry out the distribution operation. During the whole process, the intelligent control module monitors the distribution situation in real time, and continuously adjusts the distribution parameters through the closed-loop feedback control system to ensure the uniformity and density of the concrete. Through this intelligent and precise distribution control process, the system not only improves the efficiency and quality of the distribution process, but also significantly reduces the necessity of manual intervention, and improves the overall safety and reliability of tunnel construction.

[0057] Further, the S2 comprises the following steps:

[0058] The image acquisition module is used to collect the concrete lining state image during the laying process, and the concrete characterization state is annotated on the concrete lining state image;

[0059] The convolutional neural network is trained with the concrete lining state image as input and the concrete state annotation result as output;

[0060] Based on the trained convolutional neural network, the real-time collected concrete lining status images are processed to extract the real-time concrete representation status.

[0061] In some embodiments, the collected images are transmitted to the intelligent control module for preprocessing. In the preprocessing stage, the system performs denoising, contrast enhancement and other processing on the images to improve the quality and clarity of the images. Then, the system annotates the preprocessed images for concrete characterization status. This process combines expert knowledge and historical data to annotate the key features of concrete in the image, including the degree of surface bubbling, the degree of slurry overflow and the location of defects. The annotated results are used as a standard data set for training convolutional neural networks. In the training stage of the convolutional neural network, the system uses the annotated concrete lining state image as input and the concrete characterization state annotation results as output for supervised learning training. The convolutional neural network consists of multiple convolutional layers, pooling layers and fully connected layers. Through repeated iterative training, the network parameters are continuously adjusted to optimize the performance of the model. During the training process, the system uses the gradient descent method and the back propagation algorithm to minimize the loss function and improve the accuracy and robustness of the model. After sufficient training, the convolutional neural network has the ability to automatically identify and analyze the state of the concrete lining. In practical applications, the system uses the trained convolutional neural network to process the real-time collected concrete lining state images. Specifically, after the real-time image is input into the convolutional neural network, the features in the image are extracted layer by layer through convolution operations and activation functions, and the representation state of the concrete is finally output. These representation states include key information such as the degree of surface bubbling, the degree of slurry overflow, and the location of defects.

[0062] Furthermore, the concrete characterization state includes the degree of surface bubbling, the degree of slurry bleeding and the location of defects.

[0063] It should be noted that the representation state extracted by the convolutional neural network, in which the surface bubbling degree can be obtained by analyzing the number and distribution density of bubbles in the image, the system evaluates the bubbling degree of the concrete surface. The degree of concrete surface slurry can be calculated by image texture analysis and gray value changes. The defect location is the relative plane coordinates of cracks and holes on the concrete surface.

[0064] Furthermore, S3 includes: generating a continuous state set according to the concrete representation state, defining the state parameters of the reinforcement learning model according to the continuous state set and the height difference matrix, using vibration positioning, vibration time and vibration intensity as action parameters, generating a reward value according to the concrete state after vibration and the height difference matrix, and training the reinforcement learning model.

[0065] In some embodiments, first, the system generates a continuous state set according to the concrete characterization state. These characterization states include the degree of surface bubbling, the degree of slurry overflow and the defect location, which are obtained in real time through convolutional neural network (CNN) analysis. The system standardizes these characterization state data to form a continuous state set. The standardization process includes steps such as normalization and noise removal to ensure the stability and consistency of the state data. Next, the system combines the continuous state set with the height difference matrix to define the state parameters of the reinforcement learning model. The height difference matrix records the concrete height difference at each point in the distribution area and is obtained through the distance perception module. The system combines the height difference matrix with the continuous state set to form the input state of the reinforcement learning model. This process requires weighting the height difference matrix so that it has the same magnitude and importance as the concrete characterization state. The core of the reinforcement learning model is to define reasonable state parameters and action parameters. In this embodiment, vibration positioning, vibration time and vibration intensity are used as action parameters. Vibration positioning refers to the position of the vibration equipment on the concrete surface, the vibration time is the duration of the vibration applied by the equipment at each position, and the vibration intensity is the strength of the vibration. These parameters are represented by multidimensional vectors as the output action of the reinforcement learning model. To optimize these action parameters, the system is trained using a reinforcement learning algorithm. During the training process, the system generates a reward value based on the concrete state after vibration and the height difference matrix. The generation of the reward value is based on the uniformity and density of the concrete. If the concrete state after vibration meets expectations, a high reward value is given; otherwise, a low reward value or penalty value is given. The calculation of the reward value takes into account multiple factors, including no bubbles on the surface, reduced slurry, and reduced height difference. The reinforcement learning model is trained using algorithms such as Q-learning or deep Q network (DQN). During the training process, the system updates the Q value function or policy network through repeated iterations, so that the model can gradually learn the optimal vibration parameter settings. After each iteration, the system selects an action based on the current strategy, applies the vibration operation, and updates the model based on the actual effect. For example, in a vibration operation, the system selects to apply medium-intensity vibration at a certain location based on the current state and continues for a certain period of time. After vibration, the system collects the real-time state of the concrete and compares it with the target state. If the results show that the uniformity and density of concrete have improved, the system records a high reward value; if the effect is not good, a low reward value is recorded. Through multiple similar operations and feedback, the reinforcement learning model is gradually optimized and finally forms an optimal vibration parameter setting strategy.

[0066] Furthermore, the formula of the reinforcement learning is as follows:

[0067] Q(s t , a t )=Q(s t , a t )+α[r t+γ(∑ s′ P(s t+1 =s′|s t , a t )max a′ Q(s′, a′))-Q(s t , a t )];

[0068] Among them, s t is the state parameter at time step t; a t is the state parameter set s t The action parameters under the condition include vibration positioning, vibration time and vibration intensity; Q(s t , a t ) is the state parameter s t Next, execute the action parameter a t The quality value of α is the learning rate, which is used to control the learning speed; r t is the reward value at time step t; γ is the discount factor; P(s t+1 =s′|s t , a t ) is in state s t Next, perform action a t Then transfer to the next state s t+1 = probability of s′; max a′ Q(s′, a′) is the maximum quality value of selecting the optimal action a′ in the state s′ at the next step t+1.

[0069] It should be noted that, first, the system generates a continuous state set based on the concrete characterization state and the height difference matrix. The concrete characterization state includes uniformity, thickness, and defect information, which are extracted from the real-time acquired images through a convolutional neural network (CNN). These data are combined with the height difference matrix to form the state parameter s at time step t. t , which is used to describe the overall situation of concrete in the distribution area. The height difference matrix records the height difference of concrete at different points, ensuring that the system can accurately evaluate the uniformity of concrete distribution. Next, the system defines the action parameter a t , including vibration positioning, vibration time and vibration intensity. Vibration positioning refers to the position of the vibration equipment on the concrete surface, vibration time is the duration of the vibration applied by the equipment at each position, and vibration intensity is the strength of the vibration. These parameters determine the operation mode of the vibration equipment and directly affect the density and uniformity of the concrete. After the vibration operation is performed, the system generates a reward value r based on the concrete state after vibration and the height difference matrix t. The reward value reflects the impact of the current operation on the quality of concrete, including the improvement of density and uniformity. If the state of the concrete after vibration meets expectations, the system will give a higher reward value; otherwise, a lower reward value or penalty value will be given. The reward value is calculated based on a series of criteria, such as no bubbling on the surface, reduced slurry and uniform thickness. The reinforcement learning model uses the Q-learning algorithm to optimize the vibration strategy by continuously updating the Q-value function.

[0070] Furthermore, the reward value r t The formula is as follows:

[0071] r t =w 1 ·f b (s t+1 )+w 2 ·f p (s t+1 )+w 3 ·f s (s t+1 );

[0072] Among them, f b (s t+1 ) is the state parameter s of concrete at time step t+1 t+1 The degree of bubbling on the lower surface; f p (s t+1 ) is the state parameter s of concrete at time step t+1 t+1 The value of the flooding degree under f s (s t+1 ) is the state parameter s of concrete at time step t+1 t+1 The sedimentation degree value under 1 ,ω 2 and ω 3 is the weight parameter, where w 1 is a negative number, w 2 is a positive number, w 3 Is a negative number.

[0073] It should be noted that due to the reward value r t In the consideration, the greater the bubbling degree value in the next time step, the more incomplete the vibration is. The slurry degree value can be considered as a positive reward, which means that the vibration is sufficient, and the absence of obvious settlement is also one of the criteria for sufficient vibration.

[0074] Furthermore, the settlement degree value is determined according to the difference between the height difference matrix values ​​at time step t+1 and time step t.

[0075] It should be noted that the settlement degree is an important indicator for evaluating the compactness and uniformity of concrete during the vibration process. To achieve this goal, the system determines the settlement degree value by comparing the height difference matrix values ​​at time step t+1 and time step t.

[0076] Furthermore, the abnormal alarm module is used to perform the following steps:

[0077] Preset initial setting time of concrete;

[0078] Obtain the duration of concrete lining stoppage by monitoring the intelligent control module;

[0079] If the duration of the concrete lining stoppage is greater than or equal to the initial setting time of the concrete, an abnormal alarm will be issued;

[0080] If the duration of the concrete lining stoppage is less than the initial setting time of the concrete, no abnormal alarm will be issued.

[0081] In some embodiments, the initial setting time of concrete is first preset manually. The initial setting time is the time period from the beginning of mixing to the beginning of losing plasticity of concrete. During this period, the concrete needs to maintain a good construction state to ensure the smooth progress of subsequent construction and the quality of the final molding. The setting of the initial setting time is based on factors such as the concrete ratio, ambient temperature and humidity, and is determined through experiments and standard specifications. In the actual construction process, the abnormal alarm module obtains the duration of the stop of the concrete lining by monitoring the intelligent control module. While monitoring and controlling the material distribution and vibration equipment, the intelligent control module records the operation time of each link. When the system detects that the concrete lining operation stops, it starts timing and continuously monitors the accumulation of the stop time. If the duration of the stop of the concrete lining is greater than or equal to the preset initial setting time, the abnormal alarm module will immediately issue an abnormal alarm. At this time, the system will remind the operator by sound and light alarm or sending a notification, indicating that there may be a risk of initial setting of concrete and measures need to be taken in time. Measures may include restarting the vibration equipment, adjusting the concrete ratio or other emergency treatment plans to ensure construction quality.

[0082] In the absence of abnormal conditions, that is, the duration of the concrete lining stoppage is less than the preset initial setting time, the system will continue monitoring without triggering abnormal alarms. This design ensures that the system can operate continuously and stably during the construction process, and will not frequently issue false alarms due to short operation pauses.

[0083] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A tunnel secondary lining material distribution system based on neural network, characterized in that: include: A distance sensing module, an image acquisition module, an intelligent control module and an abnormal alarm module, wherein the distance sensing module, the image acquisition module and the abnormal alarm module are respectively connected to the intelligent control module; The distance sensing module is used to obtain concrete height data at several points during the laying process, and includes several laser radar sensors arranged at the laying operation window of the lining trolley; The image acquisition module is used to acquire the concrete lining status image in real time, and includes a plurality of image capture devices arranged at the material distribution operation window of the lining trolley; The intelligent control module is used to perform the following steps: S1. Select the material placing operation window according to the height difference of concrete at different points and control the placing machine to carry out the material placing operation; S2, judging the concrete representation state according to the concrete lining state image based on the convolutional neural network; S3, generating a continuous state set according to the concrete representation state, and performing reinforcement learning based on the continuous state set, the concrete height difference and the vibration parameters to obtain a reinforcement learning model; S4, based on the reinforcement learning model, real-time generation of vibration parameters to control the vibration equipment; The abnormal alarm module is used to record the duration of the concrete lining stoppage and issue an abnormal alarm.

2. According to the neural network-based tunnel secondary lining material distribution system of claim 1, it is characterized in that: The S1 comprises the following steps: Calculate the concrete height difference at different points based on the concrete height data at several points during the placement process; Generate a height difference matrix based on the concrete height difference at different points; Traversing the matrix to obtain the insufficient thickness area and generate a material distribution strategy, wherein the material distribution strategy includes determining the current material distribution operation window, planning the material distribution path and adjusting the material distribution speed; The material placing machine is controlled to perform material placing operations based on the material placing strategy.

3. The neural network-based tunnel secondary lining material distribution system according to claim 2 is characterized in that: The S2 comprises the following steps: The image acquisition module is used to collect the concrete lining state image during the laying process, and the concrete characterization state is annotated on the concrete lining state image; The convolutional neural network is trained with the concrete lining state image as input and the concrete state annotation result as output; Based on the trained convolutional neural network, the real-time collected concrete lining status images are processed to extract the real-time concrete representation status.

4. The neural network-based tunnel secondary lining material distribution system according to claim 3 is characterized in that: The concrete characterization state includes the degree of surface bubbling, the degree of slurry bleeding and the location of defects.

5. The neural network-based tunnel secondary lining material distribution system according to claim 4 is characterized in that: The S3 includes: generating a continuous state set according to the concrete representation state, defining the state parameters of the reinforcement learning model according to the continuous state set and the height difference matrix, using vibration positioning, vibration time and vibration intensity as action parameters, generating a reward value according to the concrete state after vibration and the height difference matrix, and training the reinforcement learning model.

6. The neural network-based tunnel secondary lining material distribution system according to claim 1 is characterized in that: The abnormal alarm module is used to perform the following steps: Preset initial setting time of concrete; Obtain the duration of concrete lining stoppage by monitoring the intelligent control module; If the duration of the concrete lining stoppage is greater than or equal to the initial setting time of the concrete, an abnormal alarm will be issued; If the duration of the concrete lining stoppage is less than the initial setting time of the concrete, no abnormal alarm will be issued.

Images