Centrifugal pump intelligent design and performance optimization system based on artificial intelligence large language model

Abstract

The invention relates to a centrifugal pump intelligent design and performance optimization system based on an artificial intelligence big language model, which combines a big language model, a multi-target genetic algorithm, fluid-solid coupling calculation and a kernel machine learning technology to realize automatic design, optimization and performance prediction of a centrifugal pump. According to the system, through the natural language processing and generating capacity of a large language model, a user can input requirements through a natural language, the system automatically generates a preliminary design scheme, geometric parameters and structural parameters of an impeller are optimized through a multi-target genetic algorithm, and the design effect is verified through fluid-solid coupling calculation. Meanwhile, based on a Gaussian process regression model, the system efficiently predicts pump performance (such as lift, efficiency and the like) under small sample data, and provides a confidence interval of a result. In addition, the large language model can feed back design optimization suggestions in real time, a new scheme is generated according to the modification requirements of the user, the number of design iterations is reduced, and the design efficiency is improved. The system is suitable for design optimization scenes of various centrifugal pumps, and has the characteristics of automation, high efficiency and intellectualization.

Description

Technical Field

[0001] The present invention relates to the field of pump mechanical design, and in particular to a centrifugal pump intelligent design and performance optimization system that combines an artificial intelligence large language model, a genetic algorithm, a multi-objective optimization technology, and a kernel machine learning model. The system can realize the automated design, optimization, and performance prediction of centrifugal pumps through intelligent language processing, optimization algorithms, and machine learning models. Background Art

[0002] Centrifugal pumps are widely used in various industrial fields, and their performance and design parameters directly affect the efficiency, stability and service life of the pumps. Traditional centrifugal pump design methods are mostly based on empirical formulas, CFD (computational fluid dynamics) simulation and experimental testing. These methods usually involve tedious calculations and experimental steps, and the optimization process is time-consuming, making it difficult to quickly achieve the optimal design goal.

[0003] With the development of computing technology, genetic algorithms and multi-objective optimization are gradually applied to the design process of centrifugal pumps. These technologies can be used to optimize impeller geometry parameters and evaluate performance more efficiently. However, existing design methods still rely on manual participation, making it difficult to achieve an automated and intelligent design process. On the other hand, machine learning techniques based on small sample learning, especially Gaussian process regression models, have been used to predict pump performance parameters, but they still lack intelligent interaction and improved feedback mechanisms.

[0004] Large language models (LLMs) such as Tongyi Qianwen have powerful natural language understanding, generation and data processing capabilities. Combining them with the optimization design and performance prediction technology of centrifugal pumps can significantly improve the automation of the design process and reduce manual intervention through intelligent feedback mechanisms, thereby achieving a more efficient and intelligent pump design system. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent design and performance optimization system for centrifugal pumps based on an artificial intelligence large language model, which realizes the automated design and optimization of centrifugal pumps by integrating genetic algorithms, multi-objective optimization, fluid-solid coupling calculations, and kernel machine learning techniques, and provides intelligent feedback and improvement suggestions through LLM.

[0006] The system of the present invention has the following characteristics: The preliminary design of the centrifugal pump was generated through LLM, and the impeller and structural geometric parameters were optimized in combination with a multi-objective genetic algorithm; Apply fluid-structure interaction calculation and nuclear machine learning models to intelligently predict centrifugal pump performance (such as head and efficiency); The system automatically provides feedback on optimization solutions, and users can input requirements and modification suggestions through natural language; The system has data-driven automatic learning capabilities, which can improve the accuracy of optimization and prediction through historical data and sample sets; Combined with small sample kernel machine learning technology, the system can make efficient performance predictions with a small number of samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 This is a flow chart of the intelligent design and performance optimization system for centrifugal pumps, showing the overall process from inputting design requirements to outputting the optimization results.

[0008] Figure 2 is the node graph of the input module.

[0009] Figure 3 Optimize design modules for LLM.

[0010] Figure 4 It is a fluid-structure interaction calculation module.

[0011] Figure 5 It is the performance prediction module. DETAILED DESCRIPTION

[0012] The overall structure of the system of the present invention is as follows: Figure 1 As shown, it includes the following main modules: Input module: responsible for receiving the user's design requirements and natural language input, and converting them into optimization goals that the system can recognize; Optimization module: Combine genetic algorithm and fluid-solid coupling calculation to perform multi-objective optimization of the pump's geometric parameters; Performance prediction module: uses Gaussian process regression model and small sample machine learning technology to intelligently predict pump performance; Feedback Module: LLM will generate detailed design feedback and improvement suggestions based on the calculation and prediction results to help users make design decisions. Example 1: Intelligent Optimization Design of Centrifugal Pump Impeller

[0013] In this embodiment, the geometric parameters of the centrifugal pump impeller are designed and optimized based on the combination of artificial intelligence large language model and multi-objective genetic algorithm.

[0014] Preliminary design input: The user inputs the basic design conditions of the centrifugal pump, such as flow rate Q, head H, speed n, etc. Through the LLM natural language processing module, the user can also input specific requirements through text descriptions, such as wanting to optimize the geometric parameters of the impeller to improve efficiency or reduce head fluctuations.

[0015] Multi-objective genetic algorithm optimization: The system first uses LLM to generate preliminary impeller design parameters based on parameters such as flow rate and head, including impeller inlet diameter D0, outlet diameter D2, blade placement angles β1, β2, number of blades z, etc. Then the system uses these parameters as the initial variables for multi-objective optimization, calculates multiple groups of optimization schemes through the genetic algorithm toolbox, and combines fluid-solid coupling simulation software (such as CFX) for performance calculation to obtain the pump head, efficiency and impeller stress distribution of each group of schemes.

[0016] Intelligent feedback and improvement: LLM will generate feedback information based on the calculation results. For example: "The lift of solution 1 meets the standard, but the efficiency needs to be improved. It is recommended to adjust the blade placement angle." Users can input modification requirements through natural language, and the system will automatically adjust the optimization parameters and perform secondary calculations.

[0017] Optimization result output: After multiple iterations of optimization, the system finally generates the optimal impeller geometry parameter combination and outputs design indicators such as pump efficiency, head and maximum stress. Users can generate the corresponding 3D model and view the detailed performance report of the impeller under the design conditions. Example 2: Performance prediction based on kernel machine learning

[0018] In this example, the performance of the centrifugal pump will be predicted using a kernel machine learning model and combined with a small sample data set for performance analysis.

[0019] Data preparation and standardization: First, a training sample set is constructed through the historical design data of centrifugal pumps. Each sample includes impeller parameters (such as the number of blades, impeller outlet width, blade outlet placement angle, etc.) and corresponding performance parameters such as head and efficiency. The system standardizes these samples to ensure that the input data meets the normal distribution.

[0020] Gaussian Process Regression Model Construction: The system selects the Gaussian Process Regression (GPR) model to construct the nonlinear relationship between impeller geometry and pump performance. With the assistance of LLM, the system can explain the working principle and prediction results of the model in natural language. For example, the user can ask: "Why is the head prediction value reduced?" The system will explain possible reasons, such as "The reduction in impeller outlet diameter leads to a decrease in pump flow, which affects the head." Prediction results and uncertainty assessment: Based on the training data, the system predicts the performance of the newly input impeller design parameters and outputs the predicted head, efficiency and corresponding variance (prediction uncertainty). Users can view the prediction results and further optimize the design based on the improvement suggestions provided by LLM.

[0021] De-normalization and verification: The system de-normalizes the prediction results to generate actual pump performance prediction values, and verifies the accuracy of the model through existing test data. If the prediction results deviate greatly from the experimental values, the system will automatically adjust the hyperparameters and retrain the model until a more accurate prediction result is obtained. Example 3: Intelligent feedback and interaction of the system

[0022] When users interact with the system through natural language, LLM can understand user needs in real time and generate appropriate feedback. For example, during the design process, users can enter: "Please optimize the impeller outlet width to improve efficiency and ensure that the head is not less than the design requirements." The system will automatically generate an optimization plan and feedback the results after the calculation is completed: "The optimized impeller outlet width is 35mm, the head is 48m, and the efficiency is increased by 2.5%." The system gradually improves its optimization capabilities and intelligent feedback level by continuously learning historical data and optimization results. For example, the system can automatically recommend new optimization strategies based on past successful experiences in design optimization, thereby reducing the user's trial and error costs.

Claims

1. An intelligent design and performance optimization system for centrifugal pumps based on an artificial intelligence large language model, characterized in that: The system combines a large language model and a machine learning algorithm to automatically generate a multi-objective optimization solution and intelligently predict the performance of the centrifugal pump by inputting the design parameters of the centrifugal pump. It includes the following steps: (1) Input the initial design conditions and geometric parameters of the centrifugal pump and generate a preliminary design solution using the large language model; (2) Automatically optimize the centrifugal pump impeller and structural parameters based on multi-objective genetic algorithms, fluid-structure interaction calculations and other technologies; (3) Combined with the optimization scheme generated by the large language model, intelligent prediction of the centrifugal pump's head, efficiency and other performance indicators; (4) Generate intelligent feedback and improvement suggestions for input data and design conditions through the natural language processing capabilities of the large language model; (5) The system automatically verifies the effectiveness of the optimization scheme and outputs the final design and optimization results.

2. The system according to claim 1, characterized in that By leveraging the language generation capabilities of a large language model, it is possible to provide users with real-time improvement suggestions during the design process and generate possible optimal design parameter combinations based on historical data and existing sample sets.

3. The system according to claim 1, characterized in that The system can automatically perform multi-objective optimization on the impeller inlet diameter, outlet diameter, blade placement angle, width and number of blades by connecting with the MATLAB genetic algorithm toolbox, thus reducing manual intervention.

4. The system according to claim 1, characterized in that Combining Gaussian process regression and kernel machine learning models, we can achieve accurate prediction of centrifugal pump performance and provide uncertainty assessment of prediction results based on the understanding ability of large language models.

5. The system according to claim 1, characterized in that Users can describe their requirements in natural language, and the system will automatically translate them into input conditions for the optimization model, generate corresponding design plans and conduct evaluations.

6. The system according to claim 1, characterized in that The large language model can reduce the number of design iterations and improve design efficiency by analyzing existing optimization results and combining deep learning and small sample machine learning techniques.

Images