An optimization design method for all-terrain travel structure of drainage pipeline

Abstract

The present invention discloses an optimization design method for an all-terrain traveling structure of a drainage pipeline, and relates to the technical field of traveling structure design. The method comprises: obtaining a first spiral propulsion wheel design drawing under different working conditions; establishing a first spiral propulsion wheel design drawing database; constructing a generative adversarial network model based on an attention mechanism, and training the generative adversarial network model; inputting random noise points into the trained generative adversarial network model to generate a second spiral propulsion wheel design drawing, and integrating the second spiral propulsion wheel design drawing and the first spiral propulsion wheel design drawing database into a target spiral propulsion wheel design drawing database; simulating the propulsion characteristics of three-dimensional spiral propulsion wheels corresponding to each target spiral propulsion wheel design drawing in the target spiral propulsion wheel design drawing database; judging whether each target spiral propulsion wheel design drawing meets the force condition according to the propulsion characteristics; the present invention solves the problem of unreasonable structural design of spiral propulsion wheels.

Description

Technical Field

[0001] The invention relates to the technical field of traveling structure design, and in particular to an optimization design method for an all-terrain traveling structure of a drainage pipeline. Background Art

[0002] Drainage pipes are the lifeline of cities. They are widely buried underground in cities and play an important role in urban drainage and sewage discharge. In recent years, the total length of my country's drainage network has increased year by year. By 2020, the total length of my country's urban underground drainage pipes has reached 800,000 kilometers, which is a huge scale. However, after long-term interactions such as sewage scouring, corrosion, and human damage, various complex defects will appear on the inner surface of the pipes, posing great safety risks to people's lives and the economy. With the rapid development of underground drainage pipe network construction and the aging and disrepair of a large number of in-service projects, engineering safety hazards have become prominent, and diseases such as leakage, corrosion, subsidence, cracking, and siltation are prevalent, resulting in frequent accidents such as environmental pollution, urban waterlogging, and road collapse, causing significant losses and social impacts.

[0003] In the improvement of the traveling structure, the internal activity space conditions during the actual drainage pipe inspection are not considered. According to relevant regulations, TV inspection should not be carried out with water. When the on-site conditions cannot be met, measures should be taken to lower the water level to ensure that the water level in the pipe is not greater than 20% of the pipe diameter; therefore, the road surface conditions for the drainage pipe inspection robot can be summarized as waterless or water conditions with possible structural and functional defects. For such complex road conditions, the introduction of a spiral structure as a mechanism for generating travel power is considered.

[0004] However, the current spiral propulsion wheel design method that relies on manual experience is time-consuming and labor-intensive, and the design results of different designers are different, making it difficult to fully absorb historical design experience; the existing computer-aided structural design method consumes a lot of computing resources, is time-consuming, and it is difficult to draw on existing mature design results. Summary of the invention

[0005] Based on the above technical problems, an optimization design method for an all-terrain travel structure of a drainage pipe is provided to solve the problems that the existing computer-aided structural design method consumes large computational resources, is time-consuming, and is difficult to draw on existing mature design results.

[0006] Based on this, the present invention provides an optimization design method for an all-terrain traveling structure of a drainage pipeline, comprising:

[0007] Step 1: Using Ansys software to obtain the design drawings of the first spiral propulsion wheel under different working conditions, wherein the different working conditions include different water depths, water flow velocities, and robot weights;

[0008] Step 2: unifying the size of the first spiral propulsion wheel design drawing and establishing a first spiral propulsion wheel design drawing database;

[0009] Step 3: construct a generative adversarial network model based on an attention mechanism, initialize the generative adversarial network model through a transfer learning method, and train the generative adversarial network model through a first spiral propulsion wheel design diagram database;

[0010] Step 4: inputting random noise into the trained generative adversarial network model to generate a second spiral propulsion wheel design drawing. If the clarity and recognition of the second spiral propulsion wheel design drawing meet the preset conditions, integrating the second spiral propulsion wheel design drawing and the first spiral propulsion wheel design drawing database into a target spiral propulsion wheel design drawing database;

[0011] Step 5: Simulate the propulsion characteristics of the three-dimensional spiral propulsion wheel corresponding to each target spiral propulsion wheel design in the target spiral propulsion wheel design database; and determine whether each target spiral propulsion wheel design meets the force condition based on the propulsion characteristics.

[0012] Furthermore, the method of using Ansys software to obtain the design drawing of the first spiral propulsion wheel under different working conditions includes:

[0013] Set different water flow conditions, use Ansys software to perform overall simulation analysis, and obtain the optimal spiral propulsion wheel drum shape diagram;

[0014] Taking the maximum stress as the control variable, the simulation analysis was carried out using Ansys software to obtain the optimal screw propulsion wheel size;

[0015] Different working conditions were set, and Ansys software was used to perform topological optimization design of the spiral propulsion wheel structure to obtain the optimal material utilization.

[0016] Furthermore, in step 2, the size of the first spiral propulsion wheel design drawing is unified to 512 pixels×512 pixels.

[0017] Furthermore, the constructing of a generative adversarial network model based on an attention mechanism, initializing the generative adversarial network model by a transfer learning method, and training the generative adversarial network model by a first spiral propulsion wheel design diagram database comprises:

[0018] Constructing a generative adversarial network model, wherein the generative adversarial network model includes a generative network and a discriminative network;

[0019] The pre-trained model that performs well on public face datasets is used to initialize the generative adversarial network to improve the convergence speed.

[0020] The generative adversarial network is trained using the mini-batch gradient descent method. If the loss function graph reaches a stable state, the training is stopped.

[0021] Furthermore, the training of the generative adversarial network model by using the first spiral propulsion wheel design diagram database includes:

[0022] According to the smoothness and convergence of the loss value descending curve of the image generated by the generative adversarial network model, the hyperparameters of the generative adversarial network are adjusted, and the hyperparameters include learning rate, total number of iterations, number of small batch images and momentum coefficient.

[0023] Furthermore, the attention mechanism includes:

[0024] Dividing the first spiral propulsion wheel design into 9 two-dimensional blocks;

[0025] Vectorize each two-dimensional block into a one-dimensional valid sequence that can be accepted by the Transformer model;

[0026] Through the fully connected layer, the one-dimensional valid sequence is converted into a fixed-length content vector;

[0027] Position coding information is added to each two-dimensional block and embedded into the content vector to obtain a semantic sequence.

[0028] Further, simulating the propulsion characteristics of the three-dimensional spiral propulsion wheel corresponding to each target spiral propulsion wheel design in the target spiral propulsion wheel design database; judging whether each target spiral propulsion wheel design meets the force condition according to the propulsion characteristics includes:

[0029] Establishing a three-dimensional model of a spiral propulsion wheel, and dividing the three-dimensional model of the wheel into mixed grids according to quadrilaterals and triangles to obtain a grid model of the spiral propulsion wheel;

[0030] The mesh model of the spiral propulsion wheel is saved as a file in inp format, and the file is imported into Ansys software to be assembled with the main body of the robot into an overall three-dimensional mesh model.

[0031] Beneficial effects of the present invention:

[0032] The present invention provides an optimization design method for an all-terrain travel structure of a drainage pipe, constructs a generative adversarial network, and realizes the generation of a variety of spiral propulsion wheel design drawings through random noise vectors, performs motion performance simulation analysis of a three-dimensional model of the various spiral propulsion wheel design drawings, selects the best spiral propulsion wheel design drawing from the various spiral propulsion wheel design drawings, and manufactures a spiral propulsion wheel according to the best spiral propulsion wheel design drawing.

[0033] Moreover, the generative adversarial network of the present invention introduces an attention mechanism, which replaces the convolutional neural network as a feature extractor, improves the performance of image generation, thereby improving the operating performance of the spiral propulsion wheel and solving the problem of unreasonable structural design of the spiral propulsion wheel. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] 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 labor.

[0035] Figure 1 It is a flow chart of an optimization design method of a drainage pipeline all-terrain traveling structure provided by the present invention;

[0036] Figure 2 It is a schematic diagram of the structure of the generative adversarial network provided by the present invention;

[0037] Figure 3 It is a schematic diagram of the spiral propulsion structure and parameters provided by the present invention. DETAILED DESCRIPTION

[0038] In the following description, for the purpose of explanation rather than limitation, specific details such as specific system structures and technologies are provided to facilitate a thorough understanding of the embodiments of the present invention. In order to illustrate the technical solution of the present invention, a specific embodiment is provided below for illustration.

[0039] In one embodiment, a method is provided as follows Figure 1 The optimization design method of the drainage pipeline all-terrain travel structure shown in the figure may include the following steps:

[0040] Step 1: Use Ansys software to obtain the design drawing of the first spiral propulsion wheel under different working conditions, wherein the different working conditions include different water depths, water flow speeds, and robot weights.

[0041] In this embodiment, it is based on management software computer-aided design (CAD). CAD refers to the process of using computer software to design management software on a graphical development interface, designing the flow structure and data structure of the management software, and finally generating management software that can be independently applied through automatic data loading and analysis of the computer software system.

[0042] Ansys used in this embodiment is a powerful finite element software for engineering simulation, which can solve problems ranging from relatively simple linear analysis to many complex nonlinear problems. It includes a rich unit library that can simulate any geometric shape, and has various types of material model libraries that can simulate the performance of typical engineering materials.

[0043] In this embodiment, different working conditions include different water depths, water flow velocities and robot weights.

[0044] In this embodiment, the optimization design includes shape optimization design, size optimization design and topology optimization design.

[0045] Shape optimization design: Set different water flow conditions, use Ansys software to perform overall simulation analysis, and obtain the optimal spiral propulsion wheel drum shape diagram.

[0046] Size optimization design: Taking the maximum stress as the control variable, Ansys software is used for simulation analysis to obtain the optimal screw propulsion wheel size.

[0047] Topology optimization design: Set different working conditions and use Ansys software to perform topology optimization design of the spiral propulsion wheel structure to obtain the best material utilization.

[0048] In this embodiment, the spiral propulsion structure is mainly composed of a drum and spiral blades. The drum is responsible for rotating around the axis to provide the main force. The spiral blades are wound around the outside of the drum at a certain spiral angle and interact with the external environment as the drum rotates. Its propeller-free structure effectively eliminates the influence of entanglement of floating objects and linear objects in the pipeline.

[0049] It should be understood that water flow refers to Newtonian fluid, and the generation of its shear stress is often related to the flow velocity gradient; solid refers to elastomer, and the generation of its shear stress is often related to strain; and the mud environment with a high water content exhibits properties between the two and can be regarded as a semi-fluid soil; when the spiral structure is in a mud environment, the spiral blades can easily cut into the mud through rotation, forming an obvious three-dimensional elliptical shear trajectory, and the shear stress generated by the extrusion and deformation of the mud acts along the elliptical trajectory, and the propulsion force generated is greater than that under water storage conditions.

[0050] When the spiral structure is in a waterless and hard road environment, the location where the interaction force is generated is mainly the contact surface between the spiral sheet and the pipeline. In this case, its driving force is smaller than that in soft environments such as water storage or silt deposition, but the friction traction it generates can still meet the needs of moving forward, and the more complex the road surface conditions, the greater the corresponding traction force. Moreover, this kind of waterless and disease-free road surface conditions are relatively rare in actual testing, and most of them are still soft environments.

[0051] In this embodiment, the shape optimization design includes the major axis and minor axis dimensions of the roller elliptical diameter, and the size optimization design includes the roller thickness, spiral blade thickness, and spiral blade spacing.

[0052] Step 2: Unify the size of the first spiral propulsion wheel design drawing and establish a first spiral propulsion wheel design drawing database.

[0053] In this embodiment, the size of the unified spiral propulsion wheel design drawing is 512 pixels×512 pixels.

[0054] According to the different optimization design methods mentioned above, the design drawings of the spiral propulsion wheels after unified size are summarized into three types of databases, namely, shape optimization design drawing database, size optimization design drawing database and topology optimization design drawing database.

[0055] Step 3: Construct a generative adversarial network model based on the attention mechanism, initialize the generative adversarial network model through the transfer learning method, and train the generative adversarial network model through the first spiral propulsion wheel design drawing database.

[0056] In this embodiment, CUDA and CUDNN based on NVIDIA GPU are installed, and Python and corresponding API libraries such as Tensorflow and keras are configured.

[0057] See also Figure 2 , which is a general structural diagram of a drainage pipe all-terrain travel structure design method based on a generative adversarial network provided in this embodiment. It can be seen from the figure that the generative adversarial network model constructed in this embodiment includes two parts: a generative network and a discriminative network; it should be understood that the generative adversarial network (GAN) uses adversarial training to make the samples generated by the generative network obey the real data distribution; in the generative adversarial network, there are two networks for adversarial training, one is the discriminative network, the goal is to judge as accurately as possible whether a sample comes from real data or is generated by the generative network; the other is the generative network, the goal is to generate samples whose sources cannot be distinguished by the discriminative network as much as possible, and these two networks with opposite goals are constantly trained alternately.

[0058] Transfer learning is relative to random parameter initialization. It uses a pre-trained model that performs well on public face datasets to initialize the generative adversarial network to improve the convergence speed of the model.

[0059] In this embodiment, the designed generative adversarial network model is trained by the small batch gradient descent method, and the loss value function graph is observed. When the model converges, the training is stopped.

[0060] In this embodiment, during the model training process, the main hyperparameters adjusted are: momentum coefficient, total number of iterations, number of mini-batch images, and learning rate.

[0061] The main basis for adjusting the above hyperparameters is: the performance of the model-generated images when setting different hyperparameters, that is, the smoothness and convergence of the loss value decline curve.

[0062] In this embodiment, the construction of the attention mechanism includes the following steps:

[0063] The first spiral propulsion wheel design image is divided into 9 two-dimensional blocks, and each two-dimensional block is vectorized into a one-dimensional valid sequence (x1, x2, x3, ..., x9) that can be accepted by the Transformer model, where xi is the i-th element in the one-dimensional valid sequence.

[0064] The one-dimensional valid sequence is converted into a fixed-length content vector through a fully connected layer. In order to increase the position information of the two-dimensional block, position coding information is added to each two-dimensional block.

[0065] The one-dimensional valid sequence is embedded into the content vector to obtain a semantic sequence (z1, z2, z3, ..., z9), where zj is the jth element in the semantic sequence.

[0066] Specifically, in step 3, by calculating the attention distribution of the input sequence, the attention mechanism obtains the attention value related to the current prediction value. This process is actually a reflection of how the attention mechanism alleviates the complexity of the neural network model. Through the size of the weights, the attention mechanism simulates the focus of human attention in processing information, effectively improving the performance of the model and reducing the amount of computation.

[0067] During the training process of the generative adversarial network, the input of the discriminative network is the first spiral propulsion wheel design database, which is used to distinguish the authenticity of the generated image and the original image.

[0068] Step 4: Input random noise into the trained generative adversarial network model to generate a second spiral propeller wheel design drawing. If the clarity and recognition of the second spiral propeller wheel design drawing meet the preset conditions, integrate the second spiral propeller wheel design drawing and the first spiral propeller wheel design drawing database into a target spiral propeller wheel design drawing database.

[0069] After the training of the generative adversarial network model is completed, random noise points are input into the trained generative adversarial network model to generate a second spiral propulsion wheel design drawing.

[0070] In this embodiment, the detection numerical indicators mainly include FID score, perceptual path length and linear separability.

[0071] The purpose of setting the preset conditions in this embodiment is to obtain a second spiral propulsion wheel design drawing with higher clarity and recognition. The clarity and recognition of the second spiral propulsion wheel design drawing can be compared with a clarity threshold and a recognition threshold to determine whether the clarity and recognition meet the preset conditions. The specific preset conditions can be set by oneself when applying.

[0072] Step 5: Simulate the propulsion characteristics of the three-dimensional spiral propulsion wheel corresponding to each target spiral propulsion wheel design in the target spiral propulsion wheel design database; and determine whether each target spiral propulsion wheel design meets the force condition based on the propulsion characteristics.

[0073] See also Figure 3 , which is a schematic diagram of a spiral propulsion structure and parameters provided in this embodiment, is based on the actual load of the vehicle body, and the load is continuously increased. At the same time, considering that the increase in weight will also increase the requirement for the traction force of the robot, the influence of the load change on the robot's motion performance is analyzed, so as to find a reasonable weight range.

[0074] Specifically, when the pipeline robot (including the spiral propulsion wheel and the body) moves in a pipeline with silt, the spiral propulsion wheel is the main part in contact with the silt. The meshing of the spiral propulsion wheel will have a great influence on the accuracy of the calculation, so the model of the spiral propulsion wheel needs to be processed separately; first, a three-dimensional model of the spiral propulsion wheel is established, and then the meshing is performed. Since the structure of the spiral propulsion wheel is relatively complex, this embodiment performs a mixed meshing according to quadrilaterals and triangles.

[0075] After the mesh model of the propeller wheel is established, the mesh file can be saved as an .inp format file and then imported into the Ansys software to assemble it with the car body into an overall three-dimensional mesh model.

[0076] Since sludge is a fluid, its shape is determined by the shape of the pipe. Once the geometric dimensions of the pipe are determined, the shape of the sludge is basically determined.

[0077] Specifically, a geometric space is first established. This space should be able to accommodate the geometric models of the pipeline robot and the sludge, but it should not be too large. If it is established too large, the geometric space will eventually be divided into a large number of grids, which will increase the calculation time. In this embodiment, the longest dimension of the space is designed according to the length of the pipeline; when the geometric model is established, it is defined as an Euler grid space. This grid is relative to the Lagrangian grid. During the simulation analysis process, the grid is fixed and the material can flow in the grid; Euler bodies are generally regular shapes, and the use of hexahedral grid division is simple and efficient.

[0078] In such an Euler grid space, the silt is located above the pipe. First, the geometric shape of the silt must be designed according to the analysis requirements, and then this geometric shape is subjected to Boolean operation with the Euler body. The space obtained by subtracting the two is the initial geometric shape of the silt in the Euler space. The material properties of the silt are then assigned to this space, and the two sides of the pipe are constrained so that the silt can only flow and splash inside the pipe, so that the material can flow in the Euler space. The pipeline robot is then placed on the silt as a whole, and a load is applied to the pipeline robot. The size of this load is exactly the weight of the pipeline robot. Under the action of this load, the pipeline robot will interact with materials such as silt. Since the pipeline robot itself is rigid, that is, it cannot be deformed, such interaction just acts on the contact surface between the two. When the pipeline robot falls to the surface of the pipe under the action of gravity, it is just in the initial position of the movement. The torque is then loaded onto each wheel to start the simulation analysis.

[0079] It is understandable that in the principles of terrestrial mechanics, the machine structure will have a certain impact on the kinematic performance of the wheeled structure, and the influence of the vehicle's weight on the kinematic performance and the influence of the drive wheel helix angle on traction should be considered.

[0080] Different vehicle weights mean different driving forces are required. The heavier the weight, the higher the driving force required, and the greater the torque requirement for the motor. In this way, the size of the robot will increase, which will also have a great impact on the structure. From a control perspective, the lighter the weight, the better, so that the robot is small and flexible, but a certain gravity load is required to allow the robot to fall to the surface of the pipe so that the contact between the wheel and the pipe can provide sufficient adhesion, thereby increasing rolling friction and achieving the effect of increasing traction. Based on the actual load of the vehicle body, the load is continuously increased, and at the same time, considering that the increase in weight will also increase the requirements for the robot's traction, the impact of load changes on the robot's motion performance is analyzed, so as to find a reasonable weight range and determine whether the corresponding target spiral propulsion wheel design meets the force conditions.

[0081] Different drive wheel helix angles also have a great influence on the robot's motion performance; when the drive wheel helix angle is considered as a single design variable, the drive wheel helix angle is increased from 0. As the helix angle increases, the shear distance increases. According to MSA's test data, in sand or silt, the propulsion characteristics are best when the helix angle is 30°.

[0082] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. An optimization design method for an all-terrain traveling structure of a drainage pipeline, characterized in that: include: Step 1: Using Ansys software to obtain the design drawings of the first spiral propulsion wheel under different working conditions, wherein the different working conditions include different water depths, water flow velocities, and robot weights; Step 2: unifying the size of the first spiral propulsion wheel design drawing and establishing a first spiral propulsion wheel design drawing database; Step 3: construct a generative adversarial network model based on an attention mechanism, initialize the generative adversarial network model through a transfer learning method, and train the generative adversarial network model through a first spiral propulsion wheel design diagram database; Step 4: inputting random noise into the trained generative adversarial network model to generate a second spiral propulsion wheel design drawing. If the clarity and recognition of the second spiral propulsion wheel design drawing meet the preset conditions, integrating the second spiral propulsion wheel design drawing and the first spiral propulsion wheel design drawing database into a target spiral propulsion wheel design drawing database; Step 5: simulating the propulsion characteristics of the three-dimensional spiral propulsion wheel corresponding to each target spiral propulsion wheel design in the target spiral propulsion wheel design database; judging whether each target spiral propulsion wheel design satisfies the force condition according to the propulsion characteristics; The attention mechanism includes: Dividing the first spiral propulsion wheel design into 9 two-dimensional blocks; Vectorize each two-dimensional block into a one-dimensional valid sequence that can be accepted by the Transformer model; Through the fully connected layer, the one-dimensional valid sequence is converted into a fixed-length content vector; Adding position coding information to each two-dimensional block and embedding the content vector to obtain a semantic sequence; Simulating the propulsion characteristics of the three-dimensional spiral propulsion wheel corresponding to each target spiral propulsion wheel design in the target spiral propulsion wheel design database; judging whether each target spiral propulsion wheel design meets the force condition according to the propulsion characteristics includes: Establishing a three-dimensional model of a spiral propulsion wheel, and dividing the three-dimensional model of the spiral propulsion wheel into mixed grids according to quadrilaterals and triangles to obtain a grid model of the spiral propulsion wheel; The mesh model of the spiral propulsion wheel is saved as a file in inp format, and the file is imported into Ansys software to be assembled with the vehicle body into an integral three-dimensional mesh model; Based on the actual load of the vehicle body, the load is continuously increased to determine whether the corresponding target spiral propulsion wheel design meets the force conditions.

2. The optimization design method of the drainage pipeline all-terrain traveling structure according to claim 1 is characterized in that: The method of using Ansys software to obtain the first screw propulsion wheel design drawing under different working conditions includes: Set different water flow conditions, use Ansys software to perform overall simulation analysis, and obtain the optimal spiral propulsion wheel drum shape diagram; Taking the maximum stress as the control variable, the simulation analysis was carried out using Ansys software to obtain the optimal screw propulsion wheel size; Different working conditions were set, and Ansys software was used to perform topological optimization design of the spiral propulsion wheel structure to obtain the optimal material utilization.

3. The optimization design method of the drainage pipeline all-terrain traveling structure according to claim 1 is characterized in that: In step 2, the size of the first spiral propulsion wheel design is unified to 512 pixels×512 pixels.

4. The optimization design method of the drainage pipeline all-terrain traveling structure according to claim 1 is characterized in that: The constructing of a generative adversarial network model based on an attention mechanism, initializing the generative adversarial network model by a transfer learning method, and training the generative adversarial network model by a first spiral propulsion wheel design diagram database comprises: Constructing a generative adversarial network model, wherein the generative adversarial network model includes a generative network and a discriminative network; The pre-trained model that performs well on public face datasets is used to initialize the generative adversarial network to improve the convergence speed. The generative adversarial network is trained using the mini-batch gradient descent method. If the loss function graph reaches a stable state, the training is stopped.

5. The optimization design method of the drainage pipeline all-terrain traveling structure according to claim 4 is characterized in that: The training of the generative adversarial network model by using the first spiral propulsion wheel design diagram database comprises: According to the smoothness and convergence of the loss value descending curve of the image generated by the generative adversarial network model, the hyperparameters of the generative adversarial network are adjusted, and the hyperparameters include learning rate, total number of iterations, number of small batch images and momentum coefficient.

Images