Method and system for perceiving state of TBM (Tunnel Boring Machine) under-excavation rock mass of end-to-end model

Abstract

The invention provides an end-to-end model TBM tunneling rock mass state sensing method and system, and relates to the technical field of image processing and deep learning. The method comprises the following steps: collecting TBM tunneling parameters and rock slag picture data in a TBM construction case or collecting TBM tunneling parameters and rock slag picture data of a current TBM construction tunnel under construction, and establishing a data storage warehouse module; training an end-to-end neural network model by using the TBM tunneling parameters and the corresponding rock mass state parameters; a trained end-to-end deep neural network model is deployed on a cloud server and used for receiving real-time TBM tunneling parameters and rock slag picture data images and executing rock mass state classification, and surrounding rock grades of tunneling rock masses are displayed on a visual interface of a TBM upper computer. According to the method, the rock mass state information can be displayed on the visual interface of the TBM upper computer in real time, and the defects that traditional rock mass condition parameters are difficult to obtain, and the means are backward are overcome.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing and deep learning, and in particular to a method and system for perceiving the state of a TBM in rock excavation based on an end-to-end model. Background Art

[0002] TBM (Tunnel Boring Machine) is an important tunnel construction equipment that integrates mechanical, electrical and hydraulic technologies. It can realize excavation, slag transportation and tunnel wall support at one time, and has a high level of mechanization and automation, fast excavation speed, and obvious advantages such as construction safety and environmental protection. Therefore, TBM has been widely used in underground engineering construction fields such as water conservancy, railways, coal mines and urban subways around the world. However, due to the limitations of time, cost and technical level, the geological survey before TBM tunnel construction is often not detailed and accurate, and poor geological conditions are often encountered during the construction process, which are not indicated by geological data. At the same time, the current information on the state of the rock mass excavated by TBM (such as parameters such as compressive strength and joint conditions) still relies on manual on-site sketches and indoor tests, lacks a priori and real-time information, and cannot adjust the excavation plan and control parameters in real time to cope with stratum changes or complex geological conditions. This situation leads to major accidents such as TBM jamming, damage, scrapping and even casualties.

[0003] During the TBM excavation process, a large amount of equipment operating parameter data such as thrust, torque, penetration and cutter head speed will be generated in real time. These electromechanical and hydraulic data directly reflect the rock conditions of the TBM's current excavation.

[0004] Due to the backwardness of existing methods for obtaining rock mass status information during TBM excavation, it is impossible to reflect the rock mass status of the TBM tunnel in real time to optimize the excavation plan. Therefore, a system and method that can sense the rock mass status in real time is needed to overcome the shortcoming of the existing technology that it cannot accurately reflect the rock mass status when selecting an appropriate excavation plan. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for perceiving the state of a TBM in rock excavation based on an end-to-end model in view of the deficiencies of the prior art, so as to solve the problem of perceiving the state of the TBM in rock excavation in real time.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0007] In a first aspect, the present invention provides a method for sensing the state of a TBM in rock mass excavation based on an end-to-end model, the method comprising:

[0008] Collect TBM excavation parameters and rock slag image data in TBM construction cases or collect TBM excavation parameters and rock slag image data of TBM construction tunnels currently under construction, put TBM excavation parameters and rock mass state parameters into the rock machine database, and establish a data storage warehouse module for rock mass state parameters;

[0009] The end-to-end neural network model is trained using the TBM excavation parameters and corresponding rock mass state parameters stored in the data warehouse storage module to form an end-to-end neural network kernel. The rock-machine relationship model is established using regression and clustering algorithms to form the kernel of the model calculation module.

[0010] The trained end-to-end deep neural network model is deployed on the cloud server to receive real-time TBM excavation parameters and rock slag image data images and perform rock mass state classification, and display the surrounding rock grade of the excavated rock mass on the visualization interface of the TBM host computer.

[0011] Optionally, the end-to-end deep neural network model includes a VGG16 convolutional neural network framework, a fully connected layer, a multimodal fusion layer, a relative attention layer, and a decision layer. If the surrounding rock grade prediction is a classification problem, a Softmax activation function is used, and the number of neurons in the output layer is equal to the number of classifications of the surrounding rock grade.

[0012] Optionally, the decision layer includes a 3-layer fully connected network, and finally a Softmax activation function output layer, where the Softmax activation function formula is: Where Z j Represents the value of the output layer neuron, where the number of rock mass classifications and the number of neurons are both equal to 5.

[0013] Optionally, the cross entropy loss function formula for the end-to-end deep neural network model is: Among them, y i is the true value of the i-th sample, is the predicted value, and 5 represents the number of classifications of surrounding rock grade.

[0014] In the second aspect, the present invention provides an end-to-end model TBM rock state perception system, which is used to implement the end-to-end model TBM rock state perception method according to the first aspect, and the system includes a data storage warehouse module, a rock slag image acquisition module, a TBM current excavation parameter storage, an end-to-end deep neural network module, and a real-time output display module; the data storage warehouse module is connected to the end-to-end deep neural network module for training the end-to-end deep neural network; the TBM current excavation parameter storage is connected to the end-to-end deep neural network module; the rock slag image acquisition module is connected to the end-to-end deep neural network module; the end-to-end deep neural network module is connected to the real-time output display module.

[0015] The beneficial effects of the present invention include:

[0016] The end-to-end model TBM rock state perception method provided by the present invention comprises: collecting TBM excavation parameters and rock slag image data in TBM construction cases or collecting TBM excavation parameters and rock slag image data of a TBM construction tunnel currently under construction, putting the TBM excavation parameters and rock state parameters into a rock machine database, and establishing a data storage warehouse module for rock state parameters; using the TBM excavation parameters and corresponding rock state parameters stored in the data warehouse storage module to train an end-to-end neural network model to form an end-to-end neural network kernel, using regression algorithms and clustering algorithms to establish a rock machine relationship model to form the kernel of a model calculation module; deploying the trained end-to-end deep neural network model on a cloud server to receive real-time TBM excavation parameters and rock slag image data images and perform rock state classification, and displaying the surrounding rock grade in the rock excavation on a visualization interface of a TBM host computer. The present invention can display rock mass status information in real time on the visualization interface of the TBM host computer, making up for the shortcomings of traditional rock mass condition parameter acquisition, backward means, and unclear rock mass status in front of the tunnel face; it can reduce TBM energy consumption, ensure the safety of equipment and personnel, improve TBM excavation efficiency, and greatly enhance the safe and efficient excavation capabilities of TBM. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A structural block diagram of an end-to-end deep neural network model training provided by an embodiment of the present invention is shown;

[0019] Figure 2 A schematic block diagram of a TBM rock mass excavation state perception system of an end-to-end model provided by an embodiment of the present invention is shown;

[0020] Figure 3 A system workflow block diagram provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] In order to solve the backwardness of the existing methods for acquiring rock mass status information during TBM excavation, the present invention provides a TBM rock mass status perception system and method based on an end-to-end deep network model. This system analyzes the equipment operation data and real-time slag image data generated during the TBM excavation process, and perceives the rock mass status parameters in real time, thereby overcoming the defect that the existing technology cannot effectively reflect the rock mass status of the TBM tunnel in real time so as to select an appropriate excavation plan.

[0023] Figure 1 A structural block diagram of an end-to-end deep neural network model training provided by an embodiment of the present invention is shown; Figure 2 A schematic block diagram of a TBM rock mass excavation state perception system of an end-to-end model provided by an embodiment of the present invention is shown; Figure 3 A system workflow block diagram provided by an embodiment of the present invention is shown.

[0024] The present invention provides a method for perceiving the state of a TBM in excavation of a rock mass in an end-to-end model. The method comprises: collecting TBM excavation parameters and rock slag image data in TBM construction cases or collecting TBM excavation parameters and rock slag image data of a TBM construction tunnel currently under construction, putting the TBM excavation parameters and rock mass state parameters into a rock machine database, and establishing a data storage warehouse module for the rock mass state parameters; using the TBM excavation parameters and the corresponding rock mass state parameters stored in the data warehouse storage module to train an end-to-end neural network model to form an end-to-end neural network kernel, using a regression algorithm and a clustering algorithm to establish a rock machine relationship model to form the kernel of a model calculation module; deploying the trained end-to-end deep neural network model on a cloud server to receive real-time TBM excavation parameter and rock slag image data images and perform rock mass state classification, and displaying the surrounding rock grade in the excavation rock mass on a visualization interface of a TBM host computer.

[0025] like Figures 1 to 3As shown, the present invention provides an end-to-end model of TBM rock state perception system, the system is used to implement the above-mentioned end-to-end model of TBM rock state perception method, the system includes a data storage warehouse module, a rock slag image acquisition module, a TBM current excavation parameter storage, an end-to-end deep neural network module, and a real-time output display module; the data storage warehouse module is connected to the end-to-end deep neural network module for training the end-to-end deep neural network; the TBM current excavation parameter storage is connected to the end-to-end deep neural network module; the rock slag image acquisition module is connected to the end-to-end deep neural network module; the end-to-end deep neural network module is connected to the real-time output display module.

[0026] The data storage warehouse module specifically includes a rock slag image database, an excavation parameter database, and a TBM engineering database. The rock slag image acquisition module refers to obtaining high-definition image information of on-site rock slag and compressing the acquired images into a standard RGB format. The module includes a set of high-resolution sensors and can work under different lighting conditions. The acquired images will be enhanced by a grayscale histogram equalization algorithm. The TBM current excavation parameter memory is used to store the current excavation parameters of the TBM. The current excavation parameters of the TBM include the operating parameters of the TBM, thrust, torque, and control parameters of penetration and speed. These parameters can be directly obtained from the TBM host computer. The real-time output display module is used to display the rock mass status information obtained by the end-to-end neural network algorithm - rock mass compressive strength, surrounding rock grade on a visual interface for reference by construction personnel.

[0027] The present invention can display rock mass status information in real time on the visualization interface of the TBM host computer, making up for the shortcomings of traditional rock mass condition parameter acquisition difficulties and backward means.

[0028] The deep neural network model is an end-to-end deep neural network model specially trained to perceive the state of the TBM in the rock mass excavation. Optionally, the end-to-end deep neural network model includes a VGG16 convolutional neural network framework, a fully connected layer, a multimodal fusion layer, a relative attention layer, and a decision layer. If the surrounding rock grade prediction is a classification problem, the Softmax activation function is used, and the number of neurons in the output layer is equal to the number of classifications of the surrounding rock grade. Specifically, the number of neurons in the output layer is equal to the surrounding rock grade 5.

[0029] The model is described in detail below.

[0030] Feature extraction layer: For the feature extraction of rock slag images, the VGG16 convolutional neural network framework in the convolutional neural network (CNN) is used as the basic skeleton for feature extraction. As a deep residual network, it has been proven to provide excellent performance when processing visual data, especially in image recognition and classification tasks. In order to adapt to the extraction of rock slag image features and rock slag classification, the VGG16 model is trained on the rock slag image dataset, so that it can accurately distinguish the rock slag image categories. Among them, for the classification of rock slag images, F p =CNN p (Image), here F p Indicates different categories of rock slag, CNN p (Image) is a VGG16 network model trained on a rock slag image dataset. It uses a fully connected network to extract the excavation parameter feature vector for TBM excavation parameters, including TBM operating parameters thrust, torque, and control parameters penetration and speed. The fully connected network is a 3-layer fully connected network with a Softmax activation function output layer at the end. Multimodal fusion layer: This layer combines the features extracted from the pointer and the number. The pointer features and the number features are first standardized and then merged with the image features F c =Concat(F p ,F d ), through the Concat operation, the pointer features and digital features are fused into a composite feature tensor Fc.

[0031] Relative attention layer: In order to better focus on key features and improve the adaptability of the network, the present invention introduces a relative attention mechanism. The relative attention layer strengthens the model's ability to distinguish information by calculating the relative importance weights between different features. The attention mechanism is used to further select and weight features: A = Attention (F c )F a =A⊙F c , where A is the feature weight matrix calculated by the attention model, “⊙” represents the Hadamard product, or element-by-element multiplication, and F a is the weighted feature representation.

[0032] Decision layer: This layer is responsible for integrating all the extracted and weighted features mentioned above, and making the final reading result through a deep neural network. The final fusion decision is made through a deep neural network, Output = FC (F a ), where FC is a fully connected network for decision making. The decision layer includes a 3-layer fully connected network, and finally a Softmax activation function output layer, where the Softmax activation function formula is: Where Z jRepresents the value of the output layer neuron, where the number of classifications of surrounding rock grades and the number of neurons are both equal to 5. There are 5 types of surrounding rock grades here, so 5 neurons are needed.

[0033] Loss function: In the end-to-end deep neural network model, selecting a suitable loss function is a key step in the model training process, which affects the model's adjustment of parameters and the final rock mass perception performance. Since the model needs to process both excavation parameters and rock slag image tasks at the same time, the cross-entropy loss function can be used, which is suitable for multi-classification problems:

[0034]

[0035] where y j is the one-hot vector of the target label corresponding to the true category, 5 is the number of true categories of the target label surrounding rock grade, is the model's predicted probability for each classification. In other words, the cross-entropy loss function formula of the end-to-end deep neural network model is: Among them, y i is the true value of the i-th sample, is the predicted value, and 5 represents the number of classifications of surrounding rock grade.

[0036] The present invention adopts a multi-task learning framework, sets a specific loss function for each task, and weights them reasonably to obtain the total loss function. For the pointer reading recognition task, the common loss function is the mean square error loss function:

[0037]

[0038] Among them, y i is the true value of the i-th sample, is the predicted value, N is the number of samples, and the end-to-end deep neural network model includes the VGG16 convolutional neural network framework, a fully connected layer, a multimodal fusion layer, a relative attention layer, a decision layer, and the surrounding rock grade prediction is a classification problem. The Softmax activation function is used, and the number of neurons in the output layer is equal to the surrounding rock grade 5.

[0039] In summary, the present invention discloses a system and method for perceiving the state of a TBM in rock digging with an end-to-end model. The system includes an image acquisition module, current excavation parameters of the TBM, an end-to-end neural network module, and a real-time output display module. In the method for perceiving the state of a rock digging according to the present invention, a designed end-to-end deep neural network model is used to specifically train, identify, and analyze images from the excavation parameters and rock slag size of the TBM. The model learns how to perceive the surrounding rock grade in real time through training data. The end-to-end neural network module includes a VGG16 convolutional neural network framework for extracting features of rock slag images, a fully connected layer for extracting feature vectors of excavation parameters, a multimodal fusion layer, a relative attention layer, and a decision layer. If the prediction of the surrounding rock grade is a classification problem, a Softmax activation function is used, and the number of neurons in the output layer is equal to the number of classifications (for example, 5 levels). The present invention can display the rock state information in real time on the visualization interface of the TBM host computer in the real-time output display module. The present invention makes up for the shortcomings of traditional methods such as difficulty in obtaining rock mass condition parameters, backward means, and unclear rock mass status in front of the tunnel face; it reduces TBM energy consumption, ensures the safety of equipment and personnel, improves TBM excavation efficiency, and greatly enhances the safe and efficient excavation capabilities of TBM.

[0040] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable ordinary technicians in the field to understand the content of the present invention and implement it. They cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. An end-to-end model TBM excavation rock state perception method, characterized in that: The method comprises: Collect TBM excavation parameters and rock slag image data in TBM construction cases or collect TBM excavation parameters and rock slag image data of TBM construction tunnels currently under construction, put TBM excavation parameters and rock mass state parameters into the rock machine database, and establish a data storage warehouse module for rock mass state parameters; Using the TBM excavation parameters and corresponding rock mass state parameters stored in the data warehouse storage module, an end-to-end neural network model is trained to form an end-to-end neural network kernel, and a rock-machine relationship model is established using a regression algorithm and a clustering algorithm to form the kernel of a model calculation module; The trained end-to-end deep neural network model is deployed on the cloud server to receive real-time TBM excavation parameters and rock slag image data images and perform rock mass state classification, and display the surrounding rock grade of the excavated rock mass on the visualization interface of the TBM host computer.

2. The method for sensing the state of a TBM during excavation of a rock mass based on an end-to-end model according to claim 1, characterized in that: The end-to-end deep neural network model includes a VGG16 convolutional neural network framework, a fully connected layer, a multimodal fusion layer, a relative attention layer, and a decision layer. If the surrounding rock grade prediction is a classification problem, the Softmax activation function is used, and the number of neurons in the output layer is equal to the number of classifications of the surrounding rock grade.

3. The method for sensing the state of a TBM in rock mass excavation based on an end-to-end model according to claim 2, characterized in that: The decision layer includes a 3-layer fully connected network, and finally a Softmax activation function output layer, where the Softmax activation function formula is: Where Z j Represents the value of the output layer neuron, where the number of rock mass classifications and the number of neurons are both equal to 5.

4. The method for sensing the state of a TBM in rock mass excavation based on an end-to-end model according to claim 3 is characterized in that: The cross entropy loss function formula of the end-to-end deep neural network model is: Among them, y i is the true value of the i-th sample, is the predicted value, and 5 represents the number of classifications of surrounding rock grade.

5. An end-to-end model of TBM excavation rock state perception system, characterized by: The system is used to implement the TBM rock state perception method in the end-to-end model according to any one of claims 1 to 4, and the system includes a data storage warehouse module, a slag image acquisition module, a TBM current excavation parameter storage, an end-to-end deep neural network module, and a real-time output display module; the data storage warehouse module is connected to the end-to-end deep neural network module for training the end-to-end deep neural network; the TBM current excavation parameter storage is connected to the end-to-end deep neural network module; the slag image acquisition module is connected to the end-to-end deep neural network module; the end-to-end deep neural network module is connected to the real-time output display module.

Images