CN-121980851-A - Sliding bearing multi-mode end-to-end intelligent design system based on large model

CN121980851ACN 121980851 ACN121980851 ACN 121980851ACN-121980851-A

Abstract

The invention relates to the technical field of sliding bearing design in mechanical engineering, in particular to a sliding bearing multi-mode end-to-end intelligent design system based on a large model. The invention drives the multi-modal input and the high-fidelity simulation closed loop through the large language model, thereby remarkably improving the design efficiency of the sliding bearing, greatly reducing the technical threshold, supporting the interaction of natural language and drawings, enabling non-experts to finish high-performance design, realizing full-flow automation from demand to drawings, avoiding manual intervention errors, optimizing in wide-area parameter space by utilizing the reasoning capability of the large model, realizing multi-objective global optimization, supporting the multi-modal input and output of texts, drawings and models, and enhancing engineering applicability.

Inventors

TONG XIAOMENG
ZHAO WEIKAI
GAO XIAOHUI
WANG YIXUAN

Assignees

北京航空航天大学

Dates

Publication Date: 20260505
Application Date: 20251231

Claims (8)

1. The sliding bearing multi-mode end-to-end intelligent design system based on the large model is characterized by comprising a multi-mode input and feature recognition module, a high-fidelity performance prediction module, a structural parameter intelligent optimization decision module and a parametric modeling and multi-mode output module which are sequentially connected and form an intelligent design closed loop, wherein each module is in seamless connection through a data interface, and a large language model is taken as a global decision and scheduling core to form a full-closed loop automatic design system from multi-mode demand input to engineering result output; the multi-mode input and feature recognition module is used for receiving and analyzing multi-mode design input of a user and outputting structured initial design parameters, optimization targets and performance constraints; the high-fidelity performance prediction module is used for calling a finite element solver to perform performance simulation through an automatic process according to the input design parameters, and extracting key performance indexes as evaluation results; The intelligent optimization decision module of the structural parameters is internally provided with a large language model, and is used for driving the large language model to carry out reasoning through a preset prompt project according to the evaluation result, the optimization target and the historical iteration data, generating optimization suggestions of design parameters, controlling the iterative optimization flow until the termination condition is met, and outputting an optimal design parameter set; and the parameterized modeling and multi-mode output module is used for automatically generating a corresponding three-dimensional bearing model/two-dimensional engineering drawing according to the optimal design parameter set.
2. The large model-based sliding bearing multi-modal end-to-end intelligent design system of claim 1, wherein the multi-modal design inputs in the multi-modal input and feature recognition module comprise natural language text and 2D/3D CAD engineering drawings/models, wherein: The natural language text is used for describing the application scene, the optimization target, the performance constraint and the initial design requirement of the sliding bearing; the 2D/3D CAD engineering drawing/model is used for improving the design of the existing bearing product and providing a basic structure parameter reference.
3. The sliding bearing multi-modal end-to-end intelligent design system based on large model as set forth in claim 2, wherein the multi-modal input and feature recognition module comprises: the natural language understanding unit is used for analyzing the optimization target, the performance constraint and the design parameter range which are input by a user in natural language through the large language model; the drawing recognition intelligent body is used for automatically extracting geometric parameters and product manufacturing information from engineering drawings uploaded by a user by fusing a target detection and optical character recognition deep learning model.
4. The large model-based sliding bearing multi-mode end-to-end intelligent design system according to claim 3, wherein the drawing recognition Agent is a mixed visual recognition Agent based on deep learning, and the mixed visual recognition Agent is used for analyzing 2D/3D CAD engineering drawings/models, and the technical implementation of the mixed visual recognition Agent comprises the steps of adopting a multi-stage mixed recognition frame: The first stage is to process the drawing image based on the target detection model of the directional bounding box, locate and extract the product manufacturing information block, wherein the PMI block contains size, tolerance and roughness information; The second stage processes the PMI block through a mixed analysis flow, wherein the mixed analysis flow combines a symbol content recognition model and a general optical character recognition model to separate engineering symbols and text information in the recognition block; And thirdly, carrying out structural recombination on the identification result according to a preset grammar rule, and extracting key geometric parameters of the shaft diameter, the width, the gap and the tile angle.
5. The sliding bearing multi-mode end-to-end intelligent design system based on the large model as set forth in claim 4, wherein the high-fidelity performance prediction module specifically comprises: The automatic preprocessing unit is used for automatically generating a corresponding three-dimensional entity model based on a preset parameterized geometric template according to input design parameters, and carrying out full-automatic grid division on the three-dimensional entity model according to preset rules to generate an input file required by finite element analysis; The solver calling and calculating unit is used for automatically calling an external finite element solver and executing solving calculation on the input file so as to complete the thermal fluid dynamics analysis of the bearing performance; And the automatic post-processing unit is used for automatically analyzing the output result file of the finite element solver and extracting key performance indexes including peak temperature, minimum oil film thickness, power loss, leakage flow and dynamic characteristic coefficients from the output result file.
6. The sliding bearing multi-mode end-to-end intelligent design system based on the large model of claim 5, wherein the prompt engineering in the structural parameter intelligent optimization decision module comprises system prompt and dynamic user prompt; the system prompt is used for defining expert roles, task flows and mandatory structuring output formats for the large language model; the dynamic user prompt is dynamically generated in each iteration, and the content comprises a current iteration state, an optimization target, performance constraints, design parameter boundaries, a last round of simulation results and a historical iteration record.
7. The sliding bearing multi-modal end-to-end intelligent design system based on large model as set forth in claim 6 wherein the workflow of the structural parameter intelligent optimization decision module follows an intelligent collaborative optimization algorithm that performs the following steps in each iteration: firstly, calling the high-fidelity performance prediction module to acquire performance indexes under current design parameters; Secondly, constructing prompt information comprising the current iteration state, an optimization target, performance constraints, design parameter boundaries, historical performance indexes and historical design parameters; Submitting the prompt information to the large language model, and receiving a structured response returned by the large language model, wherein the response comprises analysis of current performance, new design parameter suggestion, an reasoning process for making the suggestion and a mark of whether iteration is stopped; And fourthly, verifying whether the new design parameter suggestion meets the design parameter boundary, if so, updating the current design parameter, returning to the step S1, and if not, stopping the loop and outputting the current optimal result.
8. The large-model-based sliding bearing multi-mode end-to-end intelligent design method according to any one of claims 1 to 6, comprising the following steps: s1, receiving multi-mode design input, analyzing and extracting an initial design parameter set, an optimization target and performance constraints; S2, entering an intelligent optimization iteration loop, wherein in a single loop: s2.1, performing high-fidelity performance simulation through an automatic process based on a current design parameter set to obtain performance indexes; S2.2, forming prompts by the performance indexes, the optimization targets, the constraints and the historical data, and inputting the prompts into a large language model; s2.3, acquiring parameter optimization suggestions and reasoning generated by the large language model; s2.4, verifying and updating a design parameter set; S3, when the iteration meets the termination condition, exiting the loop, and outputting an optimal design parameter set; And S4, driving a parameterized modeling tool to automatically generate a three-dimensional bearing model/two-dimensional engineering drawing according to the optimal design parameter set.

Description

Sliding bearing multi-mode end-to-end intelligent design system based on large model Technical Field The invention relates to the technical field of sliding bearing design in mechanical engineering, in particular to a sliding bearing multi-mode end-to-end intelligent design system based on a large model. Background The sliding bearing is a core supporting component of the rotary machine, the performance of the sliding bearing directly determines the stability, efficiency and service life of the whole equipment, and the sliding bearing is widely applied to the key fields of aerospace, energy power, high-end equipment and the like. With the increasing urgent design demands of modern high-end equipment for high performance and rapid iteration, the limitations of the traditional sliding bearing design method are becoming more prominent. The current design method of the sliding bearing mainly has the following two types of bottlenecks: Based on experience and manual design, the design process is rapid but low in precision depending on personal experience of engineers, is difficult to cope with optimization requirements of multiple target and complex working conditions, and lacks accurate prediction of complex physical fields (such as flow fields and temperature fields) in bearings, so that design margin is unreasonable. The design based on numerical simulation adopts high-fidelity tools such as Finite Element Analysis (FEA), computational Fluid Dynamics (CFD) and the like to predict the performance, and has the following obvious defects although the accuracy is high: design-simulation-analysis iteration period is long (several hours to days), and efficiency is low; the requirements on the operation capability of simulation software of engineers are high, and the technical threshold is high; the design, modeling, simulation and optimization links are mutually fractured, the data transmission is dependent on manpower, and the full-flow automation is difficult to realize; the global optimization is extremely costly in a wide design parameter space, and usually only local trial and error can be performed. Therefore, the prior art is difficult to meet the design requirement of modern high-end equipment on high performance and rapid iteration of the bearing. Disclosure of Invention The invention aims to provide a sliding bearing multi-mode end-to-end intelligent design system based on a large model, so as to solve the problem that the design requirements of modern high-end equipment on high-performance and rapid iteration of a bearing are difficult to meet. The sliding bearing multi-mode end-to-end intelligent design system based on the large model comprises a multi-mode input and feature identification module, a high-fidelity performance prediction module, a structural parameter intelligent optimization decision module and a parametric modeling and multi-mode output module which are sequentially connected and form an intelligent design closed loop, wherein each module is in seamless connection through a data interface, and takes the large language model as a global decision and scheduling core to form a full-closed loop automatic design system from multi-mode demand input to engineering result output; the multi-mode input and feature recognition module is used for receiving and analyzing multi-mode design input of a user and outputting structured initial design parameters, optimization targets and performance constraints; the high-fidelity performance prediction module is used for calling a finite element solver to perform performance simulation through an automatic process according to the input design parameters, and extracting key performance indexes as evaluation results; The intelligent optimization decision module of the structural parameters is internally provided with a large language model, and is used for driving the large language model to carry out reasoning through a preset prompt project according to the evaluation result, the optimization target and the historical iteration data, generating optimization suggestions of design parameters, controlling the iterative optimization flow until the termination condition is met, and outputting an optimal design parameter set; and the parameterized modeling and multi-mode output module is used for automatically generating a corresponding three-dimensional bearing model/two-dimensional engineering drawing according to the optimal design parameter set. Preferably, the multi-modal design input in the multi-modal input and feature recognition module comprises a natural language text and a 2D/3D CAD engineering drawing/model, wherein: The natural language text is used for describing the application scene, the optimization target, the performance constraint and the initial design requirement of the sliding bearing; the 2D/3D CAD engineering drawing/model is used for improving the design of the existing bearing product and providing a basic structure parameter reference