An aircraft structural component design method based on convolutional neural network

Abstract

The present invention discloses a design method for aircraft structural components based on a convolutional neural network, comprising the following steps: (1) Designers collect two-dimensional design drawings of different aircraft structural components, input the feature points of the aircraft structural components through a human-computer interaction interface, and build a convolutional neural network model; (2) Through rule reasoning machine for reasoning and analysis, obtain the relative spatial position information and curve information between the feature points in the feature information; (3) The CNN convolutional neural network trains on the data set, detects key points, and establishes a convolutional neural network training library; (4) Process the feature information, automatically detect important parts in the nodes and establish a parametric three-dimensional model, and save this instance to the knowledge-model library in real time. The present invention fully combines the convolutional neural network with the design of aircraft structural components, and can quickly establish a three-dimensional parametric model of aircraft structural components from two-dimensional design drawings, shortening the design cycle and R & D cycle of aircraft structural components.

Description

Technical Field

[0001] The present invention relates to the technical field of aircraft design, and particularly to an aircraft structural member design method based on a convolutional neural network. Background Art

[0002] With the rapid development of advanced aircraft manufacturing technologies, higher requirements are put forward for the design and manufacturing of aircraft structural members. In order to improve the design efficiency of aircraft and save economic costs, the United States first adopted digital technologies in the field of aircraft structural member design, and at the same time modified and optimized the design process of traditional aircraft structural members through digital technologies. Aircraft structural members are gradually developing towards integralization and enlargement of part structures, and the complexity is also increasing. There are problems such as low design efficiency, cumbersome and complex design modeling processes, and low knowledge utilization rate in the field of aircraft structural member design, which also become important obstacles on the road to realizing intelligent manufacturing of large aircraft structural members.

[0003] The current patent disclosures and literature show that: 1) Patent (CN201911335707.7) Automatic construction method for the web processing area of aircraft structural members based on images. This method is applicable to constructing the web processing area of aircraft structural members that contain various complex elements such as broken surfaces, broken edges, opening and closing angles, and clamping bosses in the part model, but it does not realize the construction of various types of aircraft structural members and cannot meet the requirements of the current diverse types and design diversities of aircraft structural members. 2) Patent (CN201510577537.9) Aircraft structural member feature point automatic acquisition system and its acquisition method. The aircraft structural member feature point automatic acquisition system and acquisition method provided by this method can realize the automatic acquisition of feature points after setting the reference position, solving the problem of large errors in feature point acquisition by traditional workers. However, this method is only applicable to the acquisition of feature points, with a narrow scope of application and low knowledge utilization rate in aircraft structural member design.

[0004] In summary, although the existing research results and methods can, to a certain extent, realize the intelligent design of complex aircraft structural members, there are problems such as a relatively high error tolerance rate, limited design methods, and a large variety of aircraft structural members with complex structures. The existing methods have a low design knowledge utilization rate and low design efficiency, and cannot meet the requirements of high efficiency, high design quality, and parametric design in the current field of aircraft structural member design in China. Summary of the Invention

[0005] The technical problem to be solved by the present invention is as follows: Due to the low utilization rate of aircraft structural member design knowledge by designers or systems and low design efficiency, the present invention proposes an aircraft structural member design method based on a convolutional neural network. This method fully combines the convolutional neural network and aircraft structural member design, and can quickly establish a three-dimensional parametric model of an aircraft structural member from a two-dimensional design drawing, shortening the design cycle and R & D cycle of aircraft structural members.

[0006] The present invention achieves the above object through the following technical solutions: A design method for aircraft structural parts based on a convolutional neural network, whose technical architecture is divided into four parts, namely, a knowledge acquisition and construction module, a rule-based reasoning and feature information integration module, a convolutional neural network model training module, and a three-dimensional model establishment and storage system module. Specifically:

[0007] (1) Knowledge acquisition and construction module

[0008] The knowledge acquisition and construction module collects a large number of two-dimensional design drawings of different aircraft structural parts. Designers input the feature points of the aircraft structural parts through a man-machine interaction interface, describe the feature parameters of the aircraft structural parts, and build a convolutional neural network model.

[0009] (2) Rule-based reasoning and feature information integration module

[0010] The rule-based reasoning and feature information integration module conducts reasoning and analysis through a rule inference engine to obtain the relative spatial position information and curve information between the feature points in the feature information.

[0011] (3) Convolutional neural network model training module

[0012] The convolutional neural network model training module uses the relative spatial position information and curve information of the feature points as a data set to be screened and trained through the CNN convolutional neural network algorithm, detects key points, realizes the refinement and typification of the feature information of the aircraft structural parts, connects multi-level feature information, enables the convolutional network to have the mapping ability between input and output pairs, and thus establishes a training library for the convolutional neural network model.

[0013] (4) Three-dimensional model establishment and storage system module

[0014] The three-dimensional model establishment and storage system module processes the feature information in the training library of the convolutional neural network model, detects important parts in the nodes through OpenCV, processes the information modules of nodes, curves, construction elements, and dimension elements, establishes a parametric three-dimensional model, completes the design of the aircraft structural parts, and saves this instance to the knowledge-model library in real time, continuously updating the knowledge-model library.

[0015] Furthermore, the two-dimensional design drawings of typical aircraft structural parts in the knowledge acquisition and construction module include the front view, top view, end view, and partial view of the aircraft structural parts.

[0016] Furthermore, the establishment principle of the feature points in the knowledge acquisition and construction module should include: 1) The selection of feature points each time can be parametrically designed; 2) The selection of feature points can initially determine the type of a certain aircraft structural part.

[0017] Further, the relative spatial position information in the rule-based reasoning and feature information integration module is the coordinate value in the spatial rectangular coordinate system with x, y, and z.

[0018] Further, the rule-based reasoning in the rule-based reasoning and feature information integration module contains domain knowledge related to the characteristics of aircraft structural parts, and has an IF (condition) THEN (action) structure. When the condition of the rule is satisfied, the rule is triggered, and then the action is executed.

[0019] Further, the feature information integration in the rule-based reasoning and feature information integration module means that based on the design requirements of the structural parts, the feature parameters are obtained through reasoning in the rule library, and then the spatial position information and curve information of the feature points are analyzed to preliminarily determine the feature parameters of the structural parts. Finally, the final parameter scheme of the structural parts is optimized and determined through manual evaluation.

[0020] Further, at least three feature points need to be retained in both the relative spatial position information and the curve information in the convolutional neural network model training module.

[0021] Further, the refinement of the feature information in the convolutional neural network model training module is achieved through the pooling layer in the CNN convolutional network structure. That is, when there is a lot of feature information, some information is not very useful or redundant for constructing the three-dimensional model. The pooling layer collects features, and then extracts the features in the frequency domain of the image through Fourier transform for sparse processing, so as to remove the redundant information and extract the most important features.

[0022] Further, the parametric model in the three-dimensional model establishment and storage system module is to optimize the parameters of the aircraft structural parts based on the spatial relative position constraints between nodes and the feature information of the aircraft structural parts in the human-computer interaction system based on CATIA CAA, and form a three-dimensional model.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] The aircraft structural part design method based on the convolutional neural network of the present invention can quickly establish a three-dimensional parametric model of the aircraft structural part through a two-dimensional design drawing, improve the utilization rate of aircraft structural part design knowledge, improve the design efficiency, and shorten the design cycle and R & D cycle of the aircraft structural part. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is the operation flowchart of the present invention;

[0026] Figure 2 is the schematic diagram of the human-computer interaction system of the present invention;

[0027] Figure 3It is the flowchart of feature information integration of the present invention;

[0028] Figure 4 It is the schematic diagram of the rule base for rule reasoning in the present invention;

[0029] Figure 5 It is the schematic diagram of feature information integration of the present invention. Detailed implementation manners

[0030] The present invention will be further described in detail below with reference to the accompanying drawings. The accompanying drawings described herein are a part of this application and are used to further explain the present invention, but do not constitute a limitation to the present invention.

[0031] The present invention provides a method for designing aircraft structural parts based on a convolutional neural network, including the following steps:

[0032] (1) Knowledge acquisition and construction is to collect a large number of two-dimensional design drawings of different aircraft structural parts, including the front view, top view, end view, and partial view of the aircraft structural parts. Then, designers input the feature points of the aircraft structural parts through the human-computer interaction interface, describe the feature parameters of the aircraft structural parts, and build a convolutional neural network model. The human-computer interaction system is as Figure 2 shown.

[0033] (2) Rule-based reasoning and feature information integration is to perform reasoning and analysis through a rule inference engine. The rule inference engine has an IF (condition) THEN (action) structure. When the condition of the rule is satisfied, the rule is triggered, and then the action is executed, thereby obtaining the relative spatial position information and curve information between the feature points in the feature information. The rule base is as Figure 4 shown.

[0034] (3) Convolutional neural network model training is to use the relative spatial position information and curve information of the feature points as a data set to be screened and trained through the CNN convolutional network structure, detect the key points, refine and typicalize the feature information of the aircraft structural parts, connect the multi-level feature information, and enable the convolutional network to have the mapping ability between the input and output pairs, thereby establishing a training library for the convolutional neural network model. The feature information integration step is as Figure 3 shown.

[0035] (4) The three-dimensional model establishment and storage system processes the feature information in the training library of the convolutional neural network model, detects the important parts in the nodes through OpenCV, processes the information modules of nodes, curves, construction elements, and dimension elements, and utilizes the spatial relative position constraints between nodes and the feature information of aircraft structural parts to complete the parameter optimization of aircraft structural parts based on CATIA CAA in the human-computer interaction system, establish a parametric three-dimensional model, complete the design of aircraft structural parts, and save this instance to the knowledge-model library in real time, continuously updating the knowledge-model library.

[0036] In summary, the present invention combines the convolutional neural network and the design of aircraft structural parts for product design, converts the two-dimensional drawing into a three-dimensional parametric modeling model, greatly improves the design efficiency, and significantly improves the utilization rate of design knowledge, and can shorten the development cycle of aircraft structural parts.

[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art can still modify or equivalently replace the specific implementation manners of the present invention, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention are within the scope of the claims of the present invention pending approval.

Claims

1. A convolutional neural network-based aircraft structural component design method, characterized by: Includes the following sections: (1) Knowledge acquisition and construction module: The knowledge acquisition and construction module collects a large number of two-dimensional design drawings of different aircraft structural parts. Designers input the characteristic points of aircraft structural parts through the human-computer interaction interface, describe the characteristic parameters of aircraft structural parts and build a convolutional neural network model; (2) Based on the rule reasoning and feature information integration module, the rule reasoning and feature information integration module uses the rule reasoning engine to perform reasoning analysis to obtain the relative spatial position information and curve information between the feature points in the feature information; (3) Convolutional neural network model training module The convolutional neural network model training module uses the relative spatial position information and curve information of feature points as a data set to be screened and trained through the CNN convolutional neural network algorithm, detects key points, realizes the refinement and typicalization of the feature information of aircraft structural parts, connects multi-level feature information, and enables the convolutional network to have the mapping ability between input and output pairs, thereby establishing a training library for the convolutional neural network model; (4) 3D model building and storage system module The 3D model building and storage system module processes the feature information in the training library of the convolutional neural network model, detects the important parts of the nodes through OpenCV, and processes the information modules of nodes, curves, construction elements, and size elements to build a parametric 3D model, complete the design of aircraft structural parts, and save this instance to the knowledge-model library in real time, and continuously update the knowledge-model library; The two-dimensional design drawings of typical aircraft structural parts in the knowledge acquisition and construction module include the main view, top view, azimuth view and partial view of the aircraft structural parts; The principles for establishing the feature points in the knowledge acquisition and construction modules should include: 1) Each feature point selection can be parameterized; 2) The selection of feature points can preliminarily determine the type of a certain aircraft structural component; The relative spatial position information in the rule-based reasoning and feature information integration module is the coordinate value of x, y, z in the spatial rectangular coordinate system.

2. The aircraft structural component design method based on convolutional neural network according to claim 1, characterized in that: The rule reasoning in the rule-based reasoning and feature information integration module includes domain knowledge related to the characteristics of aircraft structural parts and has an IF condition THEN behavior structure. When the condition of the rule is met, the rule is triggered and the behavior is then executed.

3. The aircraft structural component design method based on convolutional neural network according to claim 1, characterized in that: The feature information integration in the rule-based reasoning and feature information integration module refers to obtaining feature parameters through rule base reasoning based on the structural part design requirements, and then analyzing the spatial position information and curve information of the feature points to preliminarily determine the feature parameters of the structural part. Finally, manual evaluation is used to optimize and determine the final parameter solution of the structural part.

4. The aircraft structural component design method based on convolutional neural network according to claim 1, characterized in that: At least three feature points need to be retained in the relative spatial position information and curve information in the convolutional neural network model training module.

5. The aircraft structural component design method based on convolutional neural network according to claim 1, characterized in that: The refinement of feature information in the convolutional neural network model training module is achieved through the pooling layer in the CNN convolutional network structure. That is, there is a lot of feature information, and some information is not very useful or repeated for building a three-dimensional model. The pooling layer collects features and then extracts the features of the image frequency domain after Fourier transform for sparse processing, thereby removing redundant information and extracting the most important features.

6. The aircraft structural component design method based on convolutional neural network according to claim 1, characterized in that: The parametric model in the three-dimensional model establishment and storage system module is formed by optimizing the parameters of the aircraft structural parts based on CATIA CAA in the human-computer interaction system by utilizing the spatial relative position constraints between nodes and the characteristic information of the aircraft structural parts to form a three-dimensional model.

Images