A method and system for designing expansion joints of tube sheets and tube bodies

By generating an adversarial network design generator and discriminator, combining data from different expansion and connection methods, optimizing the expansion and connection design scheme, the problem that it is difficult for the existing technology to comprehensively consider the advantages and disadvantages of multiple expansion and connection methods is achieved, and high quality and diversification are achieved swelling design supports.

CN118504398BActive Publication Date: 2024-11-26FUNKE HEAT EXCHANGER SYST CHANGZHOU
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410631952.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-21
Publication Date
2024-11-26
Estimated Expiration
2044-05-21

AI Technical Summary

Technical Problem

The existing expansion and connection technology is difficult to comprehensively consider the advantages and disadvantages of mechanical expansion, hydraulic expansion and inductive expansion, which makes it difficult to obtain the optimal production and processing parameters when facing new design needs.

Method used

Generative adversarial network design generator and discriminator are used to collect and preprocess product data of different expansion methods, generate design solutions that approximate real data, and optimize the design solutions through feedback loops to balance the advantages and disadvantages of different methods.

Benefits of technology

Generative adversarial networks can generate high-quality expansion and connection design solutions, simulate the expansion and connection effects of different methods, provide diversified data support that conforms to actual conditions, ensure the authenticity and practicality of the design, and improve design efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118504398B_ABST
    Figure CN118504398B_ABST
Patent Text Reader

Abstract

The present invention belongs to the technical field of heat exchanger production and processing, and in particular, relates to a method for designing expansion of tube sheets and tube bodies, including: collecting and preprocessing data of various products; establishing a generative adversarial network based on the preprocessed product data, designing a generator and a discriminator; continuously optimizing the performance of both parties through a feedback loop between the generator and the discriminator; and selecting the design with the best performance from the designs generated by the generator. In the expansion design of tube sheets and tube bodies, the generative adversarial network can generate high-quality data samples that are close to the actual situation. By collecting actual product data of mechanical expansion, hydraulic expansion, and induction expansion, the generator can simulate the expansion effects of different methods and provide highly realistic sample data for design analysis; different expansion methods have differences in the distribution of materials and process parameters, and the generative adversarial network can effectively capture and simulate complex distributions, providing diversified and realistic data support for the design.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of deep learning technology, and in particular relates to a method and system for designing expansion joints of a tube sheet and a tube body. Background Art

[0002] At present, the tube body and tube sheet of the heat exchanger are usually connected by the following expansion methods:

[0003] Mechanical expansion is the process of expanding the end of the tube body by mechanical force so that it can be in close contact with the tube sheet hole wall and form a seal. This method can be achieved by a tube expansion machine or other tube expansion tools. It is simple to operate and highly efficient.

[0004] Hydraulic expansion, in this method, the high pressure generated by the hydraulic tool is used to expand the tube end so that it fits tightly into the tube sheet hole. Hydraulic expansion can provide more uniform expansion force, improve contact pressure distribution, and improve the sealing and strength of the joint.

[0005] Induction expansion heats the tube end through electromagnetic induction, causing it to expand and then be inserted into the cold tube sheet hole. The tube end then cools and shrinks and is tightly fixed to the tube sheet. This method can reduce physical impact and improve the accuracy and reliability of the connection.

[0006] The above expansion methods often have their own advantages and disadvantages for different product materials, application scenarios, heat exchange media, working requirements, etc., for example:

[0007] Mechanical expansion often uses simple mechanical tools, such as tube expanders, which are low-cost, intuitive and easy to understand, easy to master and execute, and usually faster than hydraulic expansion and induction expansion, making it suitable for mass production. However, during implementation, mechanical expansion may cause uneven force on the pipe, increasing the risk of leakage, and harder or more brittle materials may not be effectively handled by mechanical expansion. In addition, mechanical damage to the pipe body may be greater, especially when high-strength expansion is used.

[0008] Hydraulic expansion can evenly expand the pipe ends through the high pressure generated by hydraulic tools, which can improve the overall quality of the expansion. It is suitable for materials of various hardness and thickness, including harder alloys, and can provide better sealing effects and reduce the possibility of leakage. However, hydraulic expansion requires the use of special hydraulic equipment, with a large initial investment, professional knowledge required to operate hydraulic equipment, high maintenance costs and operational complexity, and the speed of hydraulic expansion may be slower than mechanical expansion.

[0009] A large amount of existing product data is currently stored for each of the above-mentioned expansion methods. However, when faced with new design requirements, how to comprehensively and intelligently consider the advantages and disadvantages of multiple solutions, and combine the real product data required for processing to obtain more targeted expansion design solutions and provide optimal production and processing parameters has become a technical problem that needs to be solved. Summary of the invention

[0010] The present invention provides a method and system for designing expansion joints of a tube sheet and a tube body, which can effectively solve the problems in the background technology.

[0011] A method for designing expansion joints of tube sheets and tube bodies, the method being based on a number of heat exchanger product data obtained by using various expansion joint methods, the method comprising:

[0012] Collect existing heat exchanger product data processed based on various expansion methods, and pre-process the product data;

[0013] A generative adversarial network is established based on the preprocessed product data, a generator and a discriminator are designed, the generator generates a design close to the real product data, and the discriminator judges the authenticity of the design according to the output of the generator and the real data;

[0014] Through the feedback loop between the generator and the discriminator, the performance of both parties is continuously optimized;

[0015] Select the design with the best performance from those produced by the generator.

[0016] Furthermore, a generative adversarial network is established based on the preprocessed product data, and a generator and a discriminator are designed, including:

[0017] Determine the network type to meet the needs of processing continuous data;

[0018] Determine the generator architecture and the discriminator architecture;

[0019] The preprocessed product data is divided into a training set and a test set to respectively implement training and verification of a generative adversarial network;

[0020] Setting training parameters and executing a training process to achieve alternating training of the generator and the discriminator;

[0021] The training progress is monitored according to the training parameters.

[0022] Furthermore, the generator and the discriminator adopt the same architecture.

[0023] Furthermore, the generator includes a plurality of deconvolution layers, and the discriminator includes a number of convolution layers equal to the number of the deconvolution layers.

[0024] Furthermore, the method for determining the number of convolutional layers and deconvolutional layers includes:

[0025] Defining a search space for the convolutional layer and the deconvolutional layer;

[0026] Design evaluation functions to evaluate the performance of the generator and discriminator;

[0027] Using an automated search algorithm, searching in the defined search space, gradually adjusting the number and parameters of convolutional layers and deconvolutional layers to optimize the value of the evaluation function;

[0028] The search process is stopped until the value of the evaluation function meets the set conditions.

[0029] Furthermore, the calculation formula of the evaluation function is:

[0030] E total =α·E generator +(1-α)·E discriminator

[0031] Among them, E generator is the generator loss, E discriminator is the discriminator loss, α is the weight parameter used to adjust the proportion of generator loss and discriminator loss in the comprehensive evaluation function;

[0032] The weight parameter α is automatically adjusted according to the real-time difference between the generator loss and the discriminator loss.

[0033] Furthermore, the adjustment process of the weight parameter α includes:

[0034] Set the initial weight parameter α 0 , used to start the training process;

[0035] In each training iteration or set time interval, calculating the real-time difference ΔE between the generator loss and the discriminator loss;

[0036] An adjustment threshold is set. When the real-time difference ΔE exceeds the adjustment threshold, the value of the weight parameter α is adjusted according to the size and direction of the real-time difference ΔE, so that the weight parameter α tends to affect the part with a larger loss value.

[0037] Furthermore, the weight parameter α is adjusted by linear attenuation.

[0038] Furthermore, the weight parameter α is adjusted by exponential decay.

[0039] A tube sheet and tube body expansion design system, the design system is based on a number of heat exchanger product data obtained by using various expansion methods, the design system includes:

[0040] Data collection module, which collects the existing data of heat exchanger products processed by various expansion methods;

[0041] A preprocessing module, for preprocessing the product data;

[0042] A network generation module, which establishes a generative adversarial network based on the preprocessed product data, designs a generator and a discriminator, generates a design that is close to the real product data through the generator, and judges the authenticity of the design through the discriminator according to the output of the generator and the real data;

[0043] A network optimization module, which continuously optimizes the performance of both the generator and the discriminator through a feedback loop between the generator and the discriminator;

[0044] The scheme selection module selects the design with the best performance from the designs generated by the generator.

[0045] The technical solution of the present invention can achieve the following technical effects:

[0046] In the expansion design of tube sheets and tube bodies, the generative adversarial network can generate high-quality data samples that are close to the actual situation. By collecting actual product data of mechanical expansion, hydraulic expansion, induction expansion, etc., the generator can simulate the expansion effects of these different methods and provide highly realistic sample data for design and analysis. The distribution of heat exchanger product data is complex, and there are significant differences in the distribution of materials and process parameters of different expansion methods. The generative adversarial network can effectively capture and simulate these complex distributions, providing diverse and realistic data support for the design.

[0047] During the implementation process, through the adversarial training of the generator and the discriminator, the generative adversarial network can continuously optimize the design scheme. In the expansion design, the generator simulates the parameter adjustment and effect optimization of different expansion methods, while the discriminator judges the feasibility and advantages of the generated scheme based on the actual data to ensure the authenticity and practicality of the final scheme. When the expansion design data is limited, the generative adversarial network can generate a large amount of synthetic data to enhance the diversity and representativeness of the data set.

[0048] Through the present invention, the limitations of traditional methods are effectively solved. It is difficult for traditional methods to simultaneously consider the advantages and disadvantages of methods such as mechanical expansion, hydraulic expansion and induction expansion. The generative adversarial network can balance and integrate the advantages and disadvantages of these methods by comprehensively processing and optimizing the data of different methods, and provide a comprehensive optimized design solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, 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 recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 A flow chart of the design method for expansion of tube sheet and tube body;

[0051] Figure 2 To build a generative adversarial network based on preprocessed product data, design a flow chart of the generator and discriminator;

[0052] Figure 3 Flowchart of the method for determining the number of convolutional layers and deconvolutional layers. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0055] Embodiment 1

[0056] A method for designing expansion joints of tube sheets and tube bodies is based on data of several existing heat exchanger products obtained by using various expansion joint methods, such as Figure 1 As shown, the design method includes:

[0057] S1: Collect the existing product data of expansion design of tube sheet and tube body, and pre-process the product data;

[0058] This step involves collecting existing product data on the expansion design of tube sheets and tube bodies, including production data, material properties, performance parameters, etc. The collected data is standardized, cleaned or converted so that the data can be effectively processed by the generative adversarial network; preprocessing may include removing outliers, filling missing values, data encoding and normalization, etc., aiming to improve the quality of data and the efficiency of network training;

[0059] S2: Based on the preprocessed product data, a generative adversarial network is established, and a generator and a discriminator are designed. The generator generates a design that is close to the real product data, and the discriminator judges the authenticity of the design based on the output of the generator and the real data;

[0060] The goal of the generator is to create a seemingly real expansion design. It receives random noise as input and outputs new design data. The task of the discriminator is to evaluate the similarity between the design produced by the generator and the real product data. It determines whether the data provided is real or generated by the generator. The generator tries to deceive the discriminator so that it mistakes the generated data for real, while the discriminator learns how to distinguish between the real and the fake and improves its accuracy in identifying authenticity.

[0061] S3: Through the feedback loop between the generator and the discriminator, the performance of both sides is continuously optimized;

[0062] Through the continuous interaction between the generator and the discriminator, the performance of both is gradually improved. The generator learns how to generate more realistic data, while the discriminator improves its ability to judge authenticity. Based on the feedback during the training process, the network parameters such as learning rate, number of layers, number of neurons, etc. are continuously adjusted to find the best generation and discrimination strategies.

[0063] S4: Select the design with the best performance from the designs produced by the generator.

[0064] When selecting the optimal design solution, the generated design can be evaluated based on specific performance indicators. The following are some commonly used evaluation indicators:

[0065] Sealing: Measure the sealing effect of the design in actual application and evaluate its leakage risk;

[0066] Strength: Evaluate the mechanical strength of the design, including compressive and tensile strength;

[0067] Reliability: Analyze the reliability of the design under different working conditions, such as high temperature, high pressure, and corrosive environment;

[0068] Cost-effectiveness: Consider the processing cost and material cost of the design solution and evaluate its economic feasibility.

[0069] Processing feasibility: Evaluate the feasibility of the design in actual production, including processing difficulty and required equipment.

[0070] During the implementation process, the corresponding weight coefficients can be set according to the importance of different evaluation indicators. As a specific comprehensive scoring formula, it can be as follows:

[0071]

[0072] in:

[0073] SS is the comprehensive score;

[0074] wi is the weight of evaluation index i;

[0075] Ei is the value of evaluation index i;

[0076] n is the total number of evaluation indicators.

[0077] After the calculation is completed, the design with the highest comprehensive score can be selected as the design with the best performance based on the comprehensive scoring results; if the comprehensive scores of multiple design schemes are close, their advantages and disadvantages can be further analyzed and the final selection can be made based on the specific application scenario.

[0078] In the expansion design of tube sheets and tube bodies, the generative adversarial network can generate high-quality data samples that are close to the actual situation. By collecting actual product data of mechanical expansion, hydraulic expansion and induction expansion, the generator can simulate the expansion effects of these different methods and provide highly realistic sample data for design and analysis. The distribution of heat exchanger product data is complex, and there are significant differences in the distribution of materials and process parameters of different expansion methods. The generative adversarial network can effectively capture and simulate these complex distributions, providing diverse and realistic data support for design.

[0079] During the implementation process, through the adversarial training of the generator and the discriminator, the generative adversarial network can continuously optimize the design scheme. In the expansion design, the generator simulates the parameter adjustment and effect optimization of different expansion methods, while the discriminator judges the feasibility and advantages of the generated scheme based on actual data to ensure the authenticity and practicality of the final scheme. For example, in mechanical expansion, the expansion degree of the tube body and the contact pressure of the tube sheet are key parameters. The generative adversarial network can automatically optimize these parameters so that the design scheme can ensure sufficient sealing without causing excessive mechanical stress. In addition, the process parameters of hydraulic expansion and induction expansion, such as hydraulic pressure and induction heating temperature, can also be automatically optimized by the generative adversarial network to obtain the best combination of parameters.

[0080] When expansion design data is limited, the generative adversarial network can generate a large amount of synthetic data to enhance the diversity and representativeness of the data set; for example, it can simulate the expansion effects under various working conditions to provide rich data support for design and verification. If the data of certain expansion methods is incomplete, the generative adversarial network can generate the missing data to ensure the comprehensiveness and consistency of the design and analysis; for example, when the data of induction expansion is insufficient, the generative adversarial network can generate simulated data under this method to fill the data gap.

[0081] The present invention effectively solves the limitations of traditional methods. It is difficult for traditional methods to simultaneously consider the advantages and disadvantages of mechanical expansion, hydraulic expansion, induction expansion and other methods. The generative adversarial network can balance and integrate the advantages and disadvantages of these methods by comprehensively processing and optimizing the data of different methods, and provide a comprehensive optimized design solution. For example, in a specific application scenario, the generative adversarial network can give priority to hydraulic expansion with better sealing according to actual needs, while taking into account the ease of operation of mechanical expansion.

[0082] As a preferred embodiment of the above, a generative adversarial network is established based on the preprocessed product data, and a generator and a discriminator are designed, such as Figure 2 As shown, including:

[0083] A1: Determine the network type to meet the needs of processing continuous data; Selecting a network type suitable for processing continuous data is key, because the design data of the expansion design of the tube sheet and tube body usually contains continuous physical and engineering parameters. This step ensures that the network can handle the data type in actual applications;

[0084] A2: Determine the generator architecture and the discriminator architecture; this step is the core of achieving efficient generative adversarial networks;

[0085] A3: Split the preprocessed product data into a training set and a test set to train and verify the generative adversarial network, allowing the model to be optimized during training and verifying the generalization ability of the model on independent data;

[0086] A4: Set training parameters such as learning rate, batch size, number of iterations, etc., and execute the training process to alternately train the generator and the discriminator;

[0087] A5: Monitoring the training progress based on the training parameters, including the loss function and other performance indicators, is an important step to ensure that the GAN works as expected and to adjust the training strategy in a timely manner.

[0088] The above optimization scheme uses a generative adversarial network to process the expansion design of tube sheets and tube bodies. By determining the network type suitable for processing continuous data and designing efficient generator and discriminator architectures, the preprocessed product data is divided into training sets and test sets. The training parameters are set reasonably, and the training process is monitored. This achieves efficient generation of design schemes with excellent performance. This method can effectively improve design efficiency, save human resources, and ensure that the generated design schemes meet actual needs.

[0089] Among them, as an optimization method, the generator and the discriminator adopt the same architecture. The same architecture means that the generator and the discriminator have the same number of layers, the same layer types, such as convolutional layers, fully connected layers, etc., and the same number of parameters. This design can ensure the balance of "power" between the two. If the network structure of one side is much more complex than the other, it may be overly powerful, causing the training of the generative adversarial network to enter an invalid learning cycle; specifically, if the generator is stronger than the discriminator, it may generate very complex samples, and the discriminator will find it difficult to distinguish between true and false. As a result, the error rate of the discriminator increases, and the generator does not obtain effective gradient information for further optimization; if the discriminator is too powerful, it may easily distinguish all the outputs of the generator, resulting in extremely high losses for the generator. The generator may fall into a stagnant state and fail to learn a valid data distribution.

[0090] When the architectures of the generator and discriminator are symmetrical, they can make more stable progress during training because neither side will be significantly ahead or behind in the learning process, thus avoiding oscillations or collapse during training; symmetry helps ensure that the gradient flow during training remains healthy, that is, the gradient neither explodes nor disappears. This balance helps the generative adversarial network converge to the equilibrium point more smoothly.

[0091] Considering another direction, as an optimization method: the generator includes several deconvolution layers, and the discriminator includes convolution layers equal to the number of deconvolution layers. In the specific implementation process, the generator is responsible for converting the input random noise vector into an output that looks like real data. The generator structure includes multiple deconvolution layers, which map the input low-dimensional feature space to a high-dimensional space to generate output similar to real data; the deconvolution layer usually gradually increases the spatial resolution of the output while reducing the number of channels to gradually build more detailed generated data. In the expansion design of tube sheets and tube bodies, the generator can simulate different expansion process parameters, such as hydraulic pressure, mechanical expansion force, induction heating temperature, etc., to generate various possible expansion design schemes, and provide high-resolution and detailed simulation data for subsequent analysis and optimization.

[0092] The discriminator is responsible for analyzing the input data and judging its authenticity, that is, judging whether the input is fake data generated by the generator or real data. When it includes the same number of convolutional layers as the deconvolutional layers, the convolutional layers are used to extract features from the input data and classify them. These convolutional layers can be understood as a feature extractor, which extracts the features of the input data through convolution operations for subsequent classification tasks; the convolutional layers gradually reduce the spatial resolution and increase the number of channels to extract higher-level abstract features from the input data for authenticity discrimination.

[0093] An equal number of convolutional and deconvolutional layers helps ensure information flow between the generator and the discriminator. Since the number of layers is equal, information can be transmitted and exchanged more naturally in the network without causing an unbalanced flow of information due to too many layers in one network. This helps the generator learn how to construct realistic data and helps the discriminator more accurately evaluate the difference between generated data and real data. In the expansion design of tube sheets and tube bodies, maintaining a balanced flow of information helps the generator better simulate the impact of different process parameters on the expansion effect and improve the accuracy and reliability of the design.

[0094] As a preferred embodiment of the above, Figure 3 As shown, the method for determining the number of convolutional layers and deconvolutional layers includes:

[0095] B1: Define the search space of convolutional and deconvolutional layers; including the minimum and maximum range of the number of convolutional and deconvolutional layers, the range of the number of filters per layer, the range of the filter size per layer, the range of the stride per layer, etc., to ensure that the search process is carried out within a reasonable parameter range; considering the product data of the expansion design of the tube sheet and tube body, these ranges may be limited by physical constraints and engineering requirements, so a reasonable definition of the search space can ensure that the generated design meets the actual feasibility and performance requirements;

[0096] B2: Design an evaluation function to evaluate the performance of the generator and the discriminator. For the design of the expansion joint of the tube sheet and the tube body, the evaluation function can include the generator loss, the discriminator loss, and other performance indicators, such as the quality of the samples generated by the generator and the accuracy of the discriminator for the real and generated samples. Such an evaluation function can comprehensively evaluate the performance of the generative adversarial network and ensure that the generated design meets the design requirements.

[0097] B3: Use an automated search algorithm to search in the defined search space and gradually adjust the number and parameters of convolutional and deconvolutional layers to optimize the value of the evaluation function;

[0098] Considering the complexity and diversity of the expansion design of tube sheets and tube bodies, it is very important to select a suitable automated search algorithm that can quickly and accurately find the optimal network architecture, such as genetic algorithms, evolutionary algorithms, reinforcement learning, etc., to perform iterative searches in the search space and adjust the number and parameters of convolutional layers and deconvolutional layers according to the value of the evaluation function; for example, genetic algorithms can be used to optimize the pressure parameters of hydraulic expansion, evolutionary algorithms can be used to optimize the force parameters of mechanical expansion, and reinforcement learning can be used to optimize the temperature parameters of induction heating;

[0099] B4: Stop the search process until the value of the evaluation function meets the set conditions.

[0100] In this step, setting conditions may include:

[0101] Performance converges to a stable threshold: Monitor the value of the evaluation function. If the value of the evaluation function remains basically unchanged after several consecutive iterations, or changes within a very small range, it can be considered that the performance has converged to a stable threshold, and the search process can be stopped at this time.

[0102] Achieve preset performance goals: Set a predefined performance goal. When the value of the evaluation function reaches or exceeds the performance goal, stop the search process. This goal can be a predetermined indicator of the performance of the generator and discriminator, such as the quality of the generated samples reaches a certain level or the accuracy of the discriminator reaches a certain level.

[0103] Combined with the characteristics of the expansion design of the tube sheet and the tube body, by reasonably defining the search space, designing the evaluation function, using the automated search algorithm and setting the stop condition, the number of convolutional and deconvolutional layers of the generator and discriminator can be effectively determined, thereby achieving the optimal design of the generative adversarial network.

[0104] As a preferred method, the calculation formula of the evaluation function is:

[0105] E total =α·E generator +(1-α)·E discriminator

[0106] Among them, E generator is the generator loss, E discriminator is the discriminator loss, α is the weight parameter used to adjust the proportion of generator loss and discriminator loss in the comprehensive evaluation function;

[0107] The weight parameter α is automatically adjusted according to the real-time difference between the generator loss and the discriminator loss.

[0108] By adjusting the weight parameter α, the relative importance of the generator and discriminator during the training process can be controlled. For example, when α is close to 0, more emphasis is placed on the discriminator loss to improve the performance of the discriminator; when α is close to 1, more emphasis is placed on the generator loss to improve the performance of the generator. During the training process, the value of α can be dynamically adjusted to achieve the purpose of balancing the performance of the generator and the discriminator.

[0109] In specific expansion design, the generator loss can measure the gap between the generated expansion scheme and the real product data, such as the pressure distribution of hydraulic expansion, the contact force distribution of mechanical expansion, and the temperature field distribution of induction expansion. The discriminator loss can measure the accuracy of the discriminator in distinguishing between real expansion data and generated expansion data, such as the comparison of the sealing effect and strength distribution of real data with the generated data. In the expansion design of tube sheets and tube bodies, different process parameters, such as pressure, force, temperature, etc., have different effects on the final design effect. By adjusting the value of α, the importance of the generator and the discriminator in the optimization of these parameters can be dynamically balanced, for example:

[0110] Initial training: Pay more attention to the discriminator loss to ensure that the discriminator can accurately distinguish between real data and simulated data generated by different expansion processes, and improve the discriminator's recognition ability.

[0111] Late training: Pay more attention to the generator loss to ensure that the generator can generate high-quality expansion solutions that are close to real data and improve the generator's generation ability.

[0112] The advantage of the comprehensive evaluation function is that it can consider the performance of both the generator and the discriminator at the same time, thereby better guiding the training process of the generative adversarial network, making it converge faster and generate higher quality design data. By adjusting the weight parameter α in real time according to the difference between the generator loss and the discriminator loss, the generative adversarial network can be made more flexible and adaptable during the training process, and a better balance between the generator and the discriminator can be achieved, thereby improving the performance and training efficiency of the network.

[0113] Specifically, it can be implemented according to the following steps: The adjustment process of the weight parameter α includes:

[0114] C1: Set the initial weight parameter α 0 , used to start the training process;

[0115] C2: Calculate the real-time difference ΔE between the generator loss and the discriminator loss at each training iteration or set time interval;

[0116] C3: Set an adjustment threshold. When the real-time difference ΔE exceeds the adjustment threshold, adjust the value of the weight parameter α according to the size and direction of the real-time difference ΔE, so that the weight parameter α tends to affect the part with a larger loss value.

[0117] The weight parameter α can be adjusted by linear or exponential decay.

[0118] In the linear attenuation method, the adjustment of the weight parameter α is linear, that is, in each training step or a certain time interval, the value of α decreases or increases with the same step size. This method can be implemented relatively simply and maintains stability to a certain extent, but it may cause the adjustment process to be too rapid, making the performance changes of the network at different stages not smooth enough.

[0119] In the exponential decay mode, the adjustment of the weight parameter α is carried out according to the exponential function, usually decreasing or increasing with the exponential function. The exponential decay mode can make the adjustment of the weight parameter smoother and more in line with the changing trend of network performance. In the early stage of training, α changes faster and can quickly adapt to the initial state of the network; in the later stage of training, α changes slower, which helps to maintain the stability of the network.

[0120] Regardless of whether linear attenuation or exponential attenuation is used, by adjusting the weight parameter α, the generative adversarial network can be made more flexible and adaptive during the training process. This method can enable the generator and discriminator to receive appropriate attention during the training process at different stages, thereby better improving the performance of the network and training efficiency.

[0121] Embodiment 2

[0122] A tube sheet and tube body expansion design system, the design system is based on a number of heat exchanger product data obtained by using various expansion methods, the design system includes:

[0123] Data collection module, which collects the existing data of heat exchanger products processed by various expansion methods;

[0124] Preprocessing module, preprocessing each product data;

[0125] The network generation module builds a generative adversarial network based on the preprocessed product data, designs a generator and a discriminator, generates a design that is close to the real product data through the generator, and judges the authenticity of the design based on the output of the generator and the real data through the discriminator;

[0126] The network optimization module continuously optimizes the performance of both the generator and the discriminator through a feedback loop between them.

[0127] The scheme selection module selects the design with the best performance from the designs generated by the generator.

[0128] The technical effect achieved by this embodiment is the same as that of the above-mentioned embodiment 1, and will not be repeated here.

[0129] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for designing expansion joints of a tube sheet and a tube body, characterized in that: The design method is based on several heat exchanger product data obtained by mechanical expansion, hydraulic expansion and induction expansion, and the design method includes: Collect existing heat exchanger product data processed based on mechanical expansion, hydraulic expansion and induction expansion, the product data including production data, material properties, performance parameters, and pre-process each of the product data; A generative adversarial network is established based on the preprocessed product data, and a generator and a discriminator are designed. The generator generates a design close to the real product data, simulates different expansion process parameters, including hydraulic pressure, mechanical expansion force, and induction heating temperature, and generates various possible expansion design schemes. The generator receives random noise as input; The authenticity of the design is judged by the discriminator according to the output of the generator and the real data, including the comparison results of the sealing effect and strength distribution of the real data and the generated data; Through the feedback loop between the generator and the discriminator, the performance of both parties is continuously optimized; in mechanical expansion, the expansion degree of the tube body and the contact pressure of the tube sheet are key parameters, and the generative adversarial network automatically optimizes the key parameters, the hydraulic pressure and induction heating temperature of hydraulic expansion and induction expansion, and the best combination parameters are obtained through the generative adversarial network. Select the design with the best performance from the designs generated by the generator; when selecting the best design, evaluate the generated design according to specific performance indicators, the following are the evaluation indicators; Sealing: measure the sealing effect of the design in actual application and evaluate the leakage risk of the design; Strength: Evaluate the mechanical strength of the design, including compressive and tensile strength; Reliability: Analyze the reliability of the design under different working conditions, including performance in high temperature, high pressure, and corrosive environments; Cost-effectiveness: Consider the processing cost and material cost of the design solution and evaluate the economic feasibility of the design solution; Processing feasibility: Evaluate the feasibility of the design in actual production, including the processing difficulty and required equipment; The generator includes a plurality of deconvolution layers, and the discriminator includes a number of convolution layers equal to the number of the deconvolution layers, so as to ensure a balanced flow of information between the generator and the discriminator; the method for determining the number of convolution layers and deconvolution layers includes: Define the search space of the convolution layer and the deconvolution layer; design an evaluation function for evaluating the performance of the generator and the discriminator; use an automated search algorithm to search in the defined search space, and gradually adjust the number and parameters of the convolution layer and the deconvolution layer to optimize the value of the evaluation function; stop the search process until the value of the evaluation function meets the set conditions.

2. The expansion design method for tube sheet and tube body according to claim 1, characterized in that: A generative adversarial network is established based on the preprocessed product data, and a generator and a discriminator are designed, including: Determine the network type to meet the needs of processing continuous data; Determine the generator architecture and the discriminator architecture; The preprocessed product data is divided into a training set and a test set to respectively implement training and verification of a generative adversarial network; Setting training parameters and executing a training process to achieve alternating training of the generator and the discriminator; The training progress is monitored according to the training parameters.

3. The expansion design method for tube sheet and tube body according to claim 1, characterized in that: The calculation formula of the evaluation function is: ; in, is the generator loss, is the discriminator loss, is a weight parameter used to adjust the proportion of generator loss and discriminator loss in the comprehensive evaluation function; The weight parameter 𝛼 is automatically adjusted according to the real-time difference between the generator loss and the discriminator loss.

4. The expansion design method of tube sheet and tube body according to claim 3 is characterized in that: The weight parameter The adjustment process includes: Set the initial weight parameters , used to start the training process; Calculate the real-time difference between the generator loss and the discriminator loss at each training iteration or set time interval ; Set the adjustment threshold. When the adjustment threshold is exceeded, the real-time difference The size and direction of the weight parameter The value of the weight parameter Tends to affect the part with larger loss value.

5. The expansion design method of tube sheet and tube body according to claim 4, characterized in that: The weight parameters are realized by linear attenuation. adjustments.

6. The expansion design method of tube sheet and tube body according to claim 4, characterized in that: The weight parameter is realized by exponential decay. adjustments.

7. A tube sheet and tube body expansion design system, using a tube sheet and tube body expansion design method as claimed in claim 1, characterized in that: The design system is based on a number of heat exchanger product data obtained using various expansion methods, and the design system includes: Data collection module, which collects the existing data of heat exchanger products processed by various expansion methods; A preprocessing module, for preprocessing the product data; A network generation module, which establishes a generative adversarial network based on the preprocessed product data, designs a generator and a discriminator, generates a design that is close to the real product data through the generator, and judges the authenticity of the design through the discriminator according to the output of the generator and the real data; A network optimization module, which continuously optimizes the performance of both the generator and the discriminator through a feedback loop between the generator and the discriminator; The scheme selection module selects the design with the best performance from the designs generated by the generator.