Full-consumption aircraft process design optimization method
By building a relationship model between aircraft process design and manufacturing completion, and using MBD assembly structure tree and BP neural network model to optimize aircraft process design data, the problem of lack of optimization and adjustment reference in the existing technology is solved, and the efficiency and accuracy of aircraft manufacturing are improved sex.
Patent Information
- Application Number
- CN202411098016.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-11-12
AI Technical Summary
The existing technology has fewer establishments on the relationship between aircraft process design data and manufacturing completion, resulting in a lack of reference for process design optimization and adjustment, resulting in low aircraft manufacturing efficiency.
By collecting aircraft process design data and manufacturing completion in historical periods, building a relational model, using MBD assembly structure tree model and BP neural network model, predict manufacturing completion and optimize process design data.
It effectively solves the problem of low aircraft manufacturing efficiency, improves the optimization and adjustment capabilities of process design, and ensures the accuracy and efficiency of the manufacturing process.
Smart Images

Figure CN118940413A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of aircraft assembly and manufacturing, and relates to a method for optimizing the process design of a fully consumable aircraft. Background Art
[0002] With the rapid development of aerospace technology, the manufacturing industry is facing increasingly complex design and production challenges. Traditional manufacturing processes often use two-dimensional drawings and manual process design, resulting in poor information transmission, delayed data updates, and frequent errors in the manufacturing process. In order to improve production efficiency and product quality and promote the intelligent transformation of aviation manufacturing, three-dimensional model-based process design (MBD) and full-consumption process design concepts have emerged. MBD (Model-Based Definition) provides a more intuitive and efficient design solution by integrating all design, process and production information of the product into a three-dimensional model. It enables engineers to share and collaborate on data on a unified platform, thereby accelerating the product development cycle and improving the traceability of the design. Full-consumption process design emphasizes the comprehensive use of product data throughout the entire process to ensure that every step in the manufacturing process is consistent with the design. This method can not only improve the efficiency of data use, but also effectively reduce the error rate and ensure that the manufacturing compliance of the final product reaches 100%. In the fierce market competition, this efficient process design process will bring greater competitive advantages to aviation manufacturing companies.
[0003] However, whether the manufacturing compliance of the final product can reach 100% is closely related to the process design of the entire manufacturing process. The existing technology rarely establishes the relationship between process design data and its corresponding manufacturing completion status, thus lacking a reference for optimizing and adjusting the process design, resulting in a significant reduction in aircraft manufacturing efficiency. Summary of the invention
[0004] The present invention provides a method for optimizing the process design of a fully expendable aircraft, which aims to construct a relationship model between process design data and manufacturing completion status based on an empirical data set of a historical period, provide a reference for optimizing and adjusting the process design, and solve the problem of low aircraft manufacturing efficiency.
[0005] The purpose of the present invention can be achieved through the following technical solutions:
[0006] The present application provides a method for optimizing the process design of a fully expendable aircraft, comprising the following steps:
[0007] S1. Collect a data set sample consisting of aircraft process design data and corresponding manufacturing completion status in historical periods, wherein the process design data includes aircraft component information, component structure, component assembly sequence and process parameters; the manufacturing completion status includes compliance and non-compliance;
[0008] S2. Construct an MBD assembly structure tree model based on process design data;
[0009] S3, structure tree pruning: by comparing and removing the component nodes with the same process design data between all assembly structure trees, the main features of each assembly structure tree are retained;
[0010] S4, using the main features of the assembly structure tree as explanatory variables and the corresponding manufacturing completion status as the response variable, training the BP neural network model;
[0011] S5, obtaining the process design data of the aircraft to be manufactured, inputting the trained BP neural network model, and predicting the manufacturing completion status;
[0012] S6. Optimize and adjust the process design data based on the predicted manufacturing completion status of the aircraft to be manufactured.
[0013] Furthermore, in step S2, the MBD assembly structure tree model includes the following construction steps:
[0014] S21, hierarchical classification: classifying by levels according to the collected component information, the levels including system level, subsystem level, component level and part level;
[0015] S22, determining the parent-child relationship between components according to the collected component structure and component assembly sequence;
[0016] S23, Data Integration: Integrate the process parameters of each component into the structure tree to ensure the information integrity of each component during the assembly process;
[0017] S24, Visualization and Analysis: After completing the MBD assembly structure tree model, the model is displayed through visualization software.
[0018] Furthermore, in step S23, the process parameters include material parameters, dimensions and tolerances, processing parameters, surface treatment parameters, detection and testing parameters, and environmental parameters.
[0019] Furthermore, in step S24, the model is displayed by means of visualization software, specifically by highlighting the assembly relationship and dependency between the components, and displaying the assembly sequence of the components.
[0020] Further, in step S3, the main features of each assembly structure tree are retained by comparing and eliminating the component nodes with the same process design data between all assembly structure trees, including the following steps:
[0021] S31, creating a feature matrix: constructing a feature matrix, wherein the rows of the feature matrix represent different assembly structure trees, and the columns represent component nodes;
[0022] S32. Similarity comparison: compare component nodes, identify and delete component nodes with the same process design data in each row.
[0023] Furthermore, in step S4, the BP neural network model includes the following training steps:
[0024] S41, determining the model structure: determining the number and quantity of neuron layers of the neural network, wherein the neurons include an input layer, a hidden layer, and an output layer, wherein the input layer is the main features of the assembly structure tree, and the output layer is the manufacturing completion status;
[0025] S42, forward propagation: The input data is calculated through the neural network, starting from the input layer, passing through the hidden layers layer by layer, and finally reaching the output layer. In each neuron, the weighted sum is calculated based on the input data and weights, and finally the output of the neuron is obtained through the activation function;
[0026] S43, calculating error: comparing the output of the neural network with the expected output, and calculating the error;
[0027] S44, back propagation: Based on the error, starting from the output layer, back propagate the error layer by layer, and adjust the connection weights and biases using the gradient descent algorithm based on the error;
[0028] S45, repeat iteration: repeat step S42 to step S44 until the error reaches a preset error threshold;
[0029] S46. Model evaluation: Use an independent test data set to test the trained neural network and use the determination coefficient to evaluate the predictive performance of the model.
[0030] Furthermore, in step S41, the model structure has five hidden layers, and the number of layers of the input layer and the output layer is the same as the number of samples; connection weights and biases are provided between the input layer units and the hidden layer units, and between the hidden layer units and the output layer units, and activation functions are provided for the hidden layer and the output layer. The calculation formulas for the neuron outputs of the hidden layer and the output layer are as follows:
[0031]
[0032] Where y is the neuron output; x i is the input; b is the bias; w i is the connection weight between neurons; F is the activation function; k is the number of inputs.
[0033] Furthermore, in step S41, the "compliant" of the output layer manufacturing completion status is converted into "1", and the "non-compliant" is converted into 0.
[0034] Further, in step S43, the calculation error is calculated as follows:
[0035]
[0036] Where E is the output layer unit error function; t l is the expected output; z l Output layer output.
[0037] Furthermore, in step S44, the gradient descent algorithm is configured as a stochastic gradient descent method.
[0038] Beneficial effects of the present invention:
[0039] By collecting a data set sample consisting of aircraft process design data and corresponding manufacturing completion status in historical periods; constructing an MBD assembly structure tree model according to the process design data; retaining the main features of each assembly structure tree by comparing and eliminating component nodes with the same process design data between all assembly structure trees; training a BP neural network model with the main features of the assembly structure tree as explanatory variables and the corresponding manufacturing completion status as response variables; obtaining the process design data of the aircraft to be manufactured, inputting the trained BP neural network model, and predicting the manufacturing completion status; optimizing and adjusting the process design data according to the predicted manufacturing completion status of the aircraft to be manufactured. The present invention solves the problem that the prior art rarely establishes the relationship between process design data and manufacturing completion status, lacks reference for process design optimization and adjustment, and leads to reduced aircraft manufacturing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.
[0041] Figure 1 The present invention is a flowchart of a fully expendable aircraft process design optimization method. DETAILED DESCRIPTION
[0042] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.
[0043] See also Figure 1 The present application provides a method for optimizing the process design of a fully expendable aircraft, comprising the following steps:
[0044] S1. Collect a data set sample consisting of aircraft process design data and corresponding manufacturing completion status in historical periods, wherein the process design data includes aircraft component information, component structure, component assembly sequence and process parameters; the manufacturing completion status includes compliance and non-compliance;
[0045] In this embodiment, in the history of aircraft manufacturing, systematically collecting and organizing process design data and its corresponding manufacturing completion is crucial to improving manufacturing efficiency and product quality. First, the process design data covers various key components of the aircraft, including component structure, materials and assembly sequence. The component structure involves the three-dimensional structural information of major components such as wings, fuselage, and tail wing. This information can be obtained from old aircraft design manuals, technical literature, and CAD drawings. At the same time, the component structure shows the dependencies between the components. In addition, in order to better understand the physical properties of the components, the data collection of materials cannot be ignored. The mechanical properties and durability of the main types of materials such as aluminum alloys and carbon fiber composite materials need to be recorded in detail. Most of this information can be obtained through material specifications and databases provided by manufacturers. In terms of the assembly sequence of components, detailed assembly steps are also an indispensable part of the entire design process. Assembly manuals and factory production records will become important data sources. Through these documents, it can be clear which component should be assembled when, as well as the tools and equipment required. At the same time, in the process of data collection, special attention should be paid to the manufacturing completion. This part of the data will directly reflect the fit between the process design and actual manufacturing. The compliance and non-compliance can be recorded in detail through the quality management system and production records. For example, if a step in the manufacturing process is found to be not carried out according to the established design, the conformity score of the final product will be reduced, which is a "non-conformity" situation. When defining the standards for conformity and non-conformity, conformity means that each step in the manufacturing process is consistent with the design, and the manufacturing conformity of the final product reaches 100%. Non-conformity means that deviations have occurred in some key links, resulting in a final pass rate of less than 100%. Specific non-conformities may include improper use of materials, incorrect assembly sequence, and substandard process parameters. Recording these problems in a timely manner is an important part of the manufacturing process. By integrating the above data into a database, a complete historical record can be formed, which lays the foundation for subsequent analysis. Using data analysis tools, common failure points and common problems in the manufacturing process can be identified. This process not only helps to summarize the important factors affecting conformity, but also provides valuable lessons for future design and manufacturing processes. At the same time, based on the analysis results, companies can develop targeted improvement measures to better consider possible manufacturing obstacles in the design link and effectively prevent them in the manufacturing process.
[0046] S2. Construct an MBD assembly structure tree model based on process design data;
[0047] In the present embodiment, the assembly structure tree model is a model that shows each component of the product and the relationship between them in a tree structure, and each node represents a component, subassembly or system, and the edge represents the assembly relationship between the components. This structure not only clearly shows the composition of the product, but also can make the collaboration of links such as design, manufacture and maintenance more efficient. Before building the MBD assembly structure tree model, it is necessary to comprehensively collect and organize the process design data related to the aircraft. After having sufficient data, the MBD assembly structure tree model can be started to be built.
[0048] Furthermore, in step S2, the MBD assembly structure tree model includes the following construction steps:
[0049] S21, hierarchical classification: classifying by levels according to the collected component information, the levels including system level, subsystem level, component level and part level;
[0050] In this embodiment, the structure tree can generally be divided into the following levels:
[0051] System level: The aircraft as a whole, as the highest level.
[0052] Subsystem layer: such as aircraft power system, control system, fuselage structure, etc.
[0053] Component layer: such as engines, wings, tail fins, etc.
[0054] Component layer: includes specific parts, such as screws, rivets, etc.
[0055] S22, determining the parent-child relationship between components according to the collected component structure and component assembly sequence;
[0056] For example, a wing is a component in a subsystem, which can contain multiple parts, such as ribs, wing surfaces, skins, etc. In this way, a complete structure tree can be built layer by layer.
[0057] S23, Data Integration: Integrate the process parameters of each component into the structure tree to ensure the information integrity of each component during the assembly process;
[0058] This can be achieved by using the MBD module of the CAD software, combining the three-dimensional model with the structure tree. In this embodiment, the process parameters are important indicators and data used to define the process flow during the production and manufacturing process. Specifically, process parameters usually include the following categories: material parameters (material types such as aluminum alloy, steel, composite materials, etc.; material properties, including physical and chemical properties such as strength, toughness, hardness, and corrosion resistance), dimensions and tolerances (dimensional requirements: specific dimensions of parts such as length, width, height, etc.; tolerances: allowable deviations in dimensions to ensure interchangeability and functionality of parts), processing parameters (processing methods: such as cutting, welding, casting, forging, forming, etc.; processing conditions: such as cutting speed, feed speed, processing depth, etc.; process path: specific processes and steps for material processing), surface treatment parameters (processing methods: such as spraying, anodizing, electroplating, polishing, etc.; processing standards: such as surface roughness, coating thickness, etc.), inspection and testing parameters (detection methods: such as dimensional measurement, appearance inspection, non-destructive testing, etc.; test standards: test standards and specifications that products and components should comply with), and environmental parameters (including the impact of temperature, humidity, dust, etc. on the production process).
[0059] Furthermore, in step S23, the process parameters include material parameters, dimensions and tolerances, processing parameters, surface treatment parameters, detection and testing parameters, and environmental parameters.
[0060] S24, Visualization and Analysis: After completing the MBD assembly structure tree model, the model is displayed through visualization software.
[0061] This process can be carried out using PLM (Product Lifecycle Management) software or dedicated 3D visualization tools:
[0062] Visualize component relationships: Highlight the assembly relationships and dependencies between components.
[0063] Analyze assembly sequence: Use the structure tree to analyze the assembly sequence of each component, which can be used later to identify potential assembly conflicts and problem areas.
[0064] Furthermore, in step S24, the model is displayed by means of visualization software, specifically by highlighting the assembly relationship and dependency between the components, and displaying the assembly sequence of the components.
[0065] S3, structure tree pruning: by comparing and removing the component nodes with the same process design data between all assembly structure trees, the main features of each assembly structure tree are retained;
[0066] In this embodiment, the structure tree pruning aims to extract the main features of each assembly structure tree by comparing and eliminating duplicate features, thereby improving the efficiency of subsequent modeling. The main features of the assembly structure tree represent the parts where the process design data between all structure trees are different. These parts are the key features that determine whether the aircraft manufacturing completion status is in compliance. By extracting the main features of the assembly structure tree, the accuracy of the relationship model between the process design data and the manufacturing completion status can be improved.
[0067] Further, in step S3, the main features of each assembly structure tree are retained by comparing and eliminating the component nodes with the same process design data between all assembly structure trees, including the following steps:
[0068] S31, creating a feature matrix: constructing a feature matrix, wherein the rows of the feature matrix represent different assembly structure trees, and the columns represent component nodes;
[0069] S32. Similarity comparison: compare component nodes, identify and delete component nodes with the same process design data in each row.
[0070] S4, using the main features of the assembly structure tree as explanatory variables and the corresponding manufacturing completion status as the response variable, training the BP neural network model;
[0071] In this embodiment, the BP artificial neural network, also known as the back propagation neural network (BackpropagationNeural Network), is a common artificial neural network model. It consists of an input layer, an output layer and at least one hidden layer, and the neurons between the layers are connected by weights. The BP neural network continuously adjusts the weights through training data to minimize the prediction error of the model. The training process of the BP neural network is divided into two stages: forward propagation and back propagation. During the forward propagation process, the input data is propagated through each layer of the neural network to obtain the output result; and the back propagation process is to adjust the connection weights between the layers according to the error of the output result, thereby continuously optimizing the prediction effect of the model. Through this back propagation method, the BP neural network can gradually improve the prediction ability of the input data, and is suitable for various tasks, such as classification, regression and pattern recognition. BP neural network has been widely used in the fields of machine learning and deep learning, and is one of the important components of deep learning models.
[0072] It should be noted that when training the BP neural network model, the explanatory variables usually need to be numerical data. This is because the neural network is easier to understand and process numerical data when processing input data. However, the main features of the assembly structure tree contain non-numeric data (such as categorical variables or character types), which can be processed in the following ways: Label encoding: Convert categorical features to integers. For example, a feature "material" may have three values: "steel", "aluminum", and "plastic", which can be encoded as 0, 1, and 2 respectively. One-Hot Encoding: Create new binary features to represent each category. For example, the material features "steel", "aluminum", and "plastic" can be converted into three features ("material_steel", "material_aluminum", and "material_plastic"), each of which is 1 only in the corresponding category and 0 in the rest. The response variable (manufacturing completion status) can convert the data into 0, 1 data, that is, "conform" is converted to "1" and "non-conform" is converted to 0.
[0073] Furthermore, in step S4, the BP neural network model includes the following training steps:
[0074] S41, determining the model structure: determining the number and quantity of neuron layers of the neural network, wherein the neurons include an input layer, a hidden layer, and an output layer, wherein the input layer is the main features of the assembly structure tree, and the output layer is the manufacturing completion status;
[0075] The hidden layer is a layer of one or more layers of neurons located between the input layer and the output layer in an artificial neural network. In a neural network, neurons in the hidden layer do not directly receive input data, but receive information from neurons in the previous layer (input layer or hidden layer) through weighted connections, and pass the processed information to neurons in the next layer. The existence of hidden layers enables neural networks to learn complex nonlinear relationships and improve the predictive ability of the model.
[0076] S42, forward propagation: The input data is calculated through the neural network, starting from the input layer, passing through the hidden layers layer by layer, and finally reaching the output layer. In each neuron, the weighted sum is calculated based on the input data and weights, and finally the output of the neuron is obtained through the activation function;
[0077] In the process of forward propagation, the input data first passes through the input layer, and then passes through the neurons in the hidden layer. Each neuron calculates the weighted sum of the input data and the corresponding weight, and then inputs the weighted sum into the activation function for nonlinear transformation. After being processed by the activation function, the output is used as the input of the next layer of neurons. The calculation is passed layer by layer until the output layer produces the final result.
[0078] The activation function is a key component in neural networks, used to introduce nonlinear characteristics so that neural networks can learn and represent complex nonlinear relationships. The activation function receives the weighted input of the neuron and outputs the activation value of the neuron after nonlinear transformation. This activation value is passed to the next layer as the output of the neuron.
[0079] S43, calculating error: comparing the output of the neural network with the expected output, and calculating the error;
[0080] It should be noted that the expected output here refers to the actual value of the output vector sample.
[0081] S44, back propagation: Based on the error, starting from the output layer, back propagate the error layer by layer, and adjust the connection weights and biases using the gradient descent algorithm based on the error;
[0082] Backpropagation is the process of calculating the output error in a neural network and transferring it back to each layer of the network, using the gradient descent algorithm to continuously adjust the connection weights and biases in order to optimize model performance, reduce prediction errors, and improve the model's learning and generalization capabilities.
[0083] S45, repeat iteration: repeat step S42 to step S44 until the error reaches a preset error threshold;
[0084] S46. Model evaluation: Use an independent test data set to test the trained neural network and use the determination coefficient to evaluate the predictive performance of the model.
[0085] Furthermore, in step S41, the model structure has five hidden layers, and the number of layers of the input layer and the output layer is the same as the number of samples; connection weights and biases are provided between the input layer units and the hidden layer units, and between the hidden layer units and the output layer units, and activation functions are provided for the hidden layer and the output layer. The calculation formulas for the neuron outputs of the hidden layer and the output layer are as follows:
[0086]
[0087] Where y is the neuron output; x i is the input; b is the bias; w i is the connection weight between neurons; F is the activation function; k is the number of inputs.
[0088] Furthermore, in step S41, the "compliant" of the output layer manufacturing completion status is converted into "1", and the "non-compliant" is converted into 0.
[0089] Further, in step S43, the calculation error is calculated as follows:
[0090]
[0091] Where E is the output layer unit error function; t l is the expected output; z l Output layer output.
[0092] Furthermore, in step S44, the gradient descent algorithm is configured as a stochastic gradient descent method.
[0093] S5, obtaining the process design data of the aircraft to be manufactured, inputting the trained BP neural network model, and predicting the manufacturing completion status;
[0094] S6. Optimize and adjust the process design data based on the predicted manufacturing completion status of the aircraft to be manufactured.
[0095] Beneficial effects of the present invention:
[0096] By collecting a data set sample consisting of aircraft process design data and corresponding manufacturing completion status in historical periods; constructing an MBD assembly structure tree model according to the process design data; retaining the main features of each assembly structure tree by comparing and eliminating component nodes with the same process design data between all assembly structure trees; training a BP neural network model with the main features of the assembly structure tree as explanatory variables and the corresponding manufacturing completion status as response variables; obtaining the process design data of the aircraft to be manufactured, inputting the trained BP neural network model, and predicting the manufacturing completion status; optimizing and adjusting the process design data according to the predicted manufacturing completion status of the aircraft to be manufactured. The present invention solves the problem that the prior art rarely establishes the relationship between process design data and manufacturing completion status, lacks reference for process design optimization and adjustment, and leads to reduced aircraft manufacturing efficiency.
[0097] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.
Claims
1. A method for optimizing the process design of a fully expendable aircraft, characterized in that: The following steps are involved: S1. Collect a data set sample consisting of aircraft process design data and corresponding manufacturing completion status in historical periods, wherein the process design data includes aircraft component information, component structure, component assembly sequence and process parameters; the manufacturing completion status includes compliance and non-compliance; S2. Construct an MBD assembly structure tree model based on process design data; S3, structure tree pruning: by comparing and removing the component nodes with the same process design data between all assembly structure trees, the main features of each assembly structure tree are retained; S4, using the main features of the assembly structure tree as explanatory variables and the corresponding manufacturing completion status as the response variable, training the BP neural network model; S5, obtaining the process design data of the aircraft to be manufactured, inputting the trained BP neural network model, and predicting the manufacturing completion status; S6. Optimize and adjust the process design data based on the predicted manufacturing completion status of the aircraft to be manufactured.
2. A fully expendable aircraft process design optimization method according to claim 1, characterized in that: In step S2, the MBD assembly structure tree model includes the following construction steps: S21, hierarchical classification: classifying by levels according to the collected component information, the levels including system level, subsystem level, component level and part level; S22, determining the parent-child relationship between components according to the collected component structure and component assembly sequence; S23, Data Integration: Integrate the process parameters of each component into the structure tree to ensure the information integrity of each component during the assembly process; S24, Visualization and Analysis: After completing the MBD assembly structure tree model, the model is displayed through visualization software.
3. A fully expendable aircraft process design optimization method according to claim 2, characterized in that: In step S23, the process parameters include material parameters, dimensions and tolerances, processing parameters, surface treatment parameters, detection and testing parameters, and environmental parameters.
4. The method for optimizing the process design of a fully expendable aircraft according to claim 2, characterized in that: In step S24, the model is displayed by means of visualization software, specifically by highlighting the assembly relationship and dependency between the components, and displaying the assembly sequence of the components.
5. The method for optimizing the process design of a fully expendable aircraft according to claim 1, characterized in that: In step S3, the main features of each assembly structure tree are retained by comparing and eliminating the component nodes with the same process design data among all assembly structure trees, including the following steps: S31, creating a feature matrix: constructing a feature matrix, wherein the rows of the feature matrix represent different assembly structure trees, and the columns represent component nodes; S32. Similarity comparison: compare component nodes, identify and delete component nodes with the same process design data in each row.
6. The method for optimizing the process design of a fully expendable aircraft according to claim 1, characterized in that: In step S4, the BP neural network model includes the following training steps: S41, determining the model structure: determining the number and quantity of neuron layers of the neural network, wherein the neurons include an input layer, a hidden layer, and an output layer, wherein the input layer is the main features of the assembly structure tree, and the output layer is the manufacturing completion status; S42, forward propagation: The input data is calculated through the neural network, starting from the input layer, passing through the hidden layers layer by layer, and finally reaching the output layer. In each neuron, the weighted sum is calculated based on the input data and weights, and finally the output of the neuron is obtained through the activation function; S43, calculating error: comparing the output of the neural network with the expected output, and calculating the error; S44, back propagation: Based on the error, starting from the output layer, back propagate the error layer by layer, and adjust the connection weights and biases using the gradient descent algorithm based on the error; S45, repeat iteration: repeat step S42 to step S44 until the error reaches a preset error threshold; S46. Model evaluation: Use an independent test data set to test the trained neural network and use the determination coefficient to evaluate the predictive performance of the model.
7. A fully expendable aircraft process design optimization method according to claim 6, characterized in that: In step S41, the model structure has five hidden layers, and the number of layers of the input layer and the output layer is the same as the number of samples; connection weights and biases are set between the input layer units and the hidden layer units, and between the hidden layer units and the output layer units, and activation functions are set for the hidden layer and the output layer. The calculation formulas for the neuron outputs of the hidden layer and the output layer are as follows: Where y is the neuron output; x i is the input; b is the bias; w i is the connection weight between neurons; F is the activation function; k is the number of inputs.
8. The method for optimizing the process design of a fully expendable aircraft according to claim 6, characterized in that: In step S41, the "compliant" of the output layer manufacturing completion status is converted into "1", and the "non-compliant" is converted into "0".
9. The method for optimizing the process design of a fully expendable aircraft according to claim 6, characterized in that: In step S43, the calculation error is as follows: Where E is the output layer unit error function; t l is the expected output; z l Output layer output.
10. The method for optimizing the process design of a fully expendable aircraft according to claim 6, characterized in that: In step S44, the gradient descent algorithm is configured as a stochastic gradient descent method.
Citation Information
Patent Citations
Three-dimensional assembly process generation method and system for aerospace product
CN103413004A
Method for modeling and predicting casting quality based on unbalanced multi-source information neural network
CN114139917A
Multi-parameter fusion detection method based on BP neural network
CN114241701A
Process scheme optimization method suitable for whole vehicle assembly quality control
CN115062413A
Oscillation scanning laser arc hybrid welding process parameter optimization method and system
CN117483956A
Cited By
System and method for processing optimization through similarity analysis
TWI941122B
